Said at the event, not in the docs

The keynote cited Google CEO Sundar Pichai as saying Google is the biggest contributor of clicks to the open web and wants to remain so.

Speaker Lino CattaruzziEvidence transcript

• How AI Search links to websites

Google said that even the classic ten-blue-links layout was settled only after millions of experiments, as part of using as much data as possible for product decisions.

Speaker Lino CattaruzziEvidence transcript

• How Search is evolving
• Testing changes and quality raters

• Extended by D2-C452 Day 2: Search results have moved from ten blue links to feature-rich, media-rich results because users wanted more.
• Extended by D3-C147 Day 3: Google's quality talk said that at any moment thousands of Search experiments are probably running.

Google said that after the launch of Nano Banana, its image generation model, the Gemini app became the number one app in the app store in 2025.

Speaker Lino CattaruzziEvidence transcript

• Gemini and Search

• Extended by D2-C770 Day 2: Google attributed the September 2025 surge in worldwide search interest for Gemini to the viral launch of…

Google said it offers users a choice between AI Mode, AI Overviews and traditional search rather than focusing on one experience.

Speaker Lino CattaruzziEvidence transcript

• AI Mode usage and query trends
• AI Overviews usage

Google said 18 to 24 year olds are more engaged than ever with the AI-powered features in Search: they requery and ask more, and more complex, questions, which leads to more searches.

Speaker Lino CattaruzziEvidence transcript

• AI Overviews usage
• AI Mode usage and query trends

Google said its searches are growing year on year in both commercial and non-commercial queries.

Speaker Lino CattaruzziEvidence transcript

• AI Mode usage and query trends

Google said 77% of users of AI Overviews and AI Mode are able to make decisions faster because of them.

Speaker Lino CattaruzziEvidence transcript

• AI Overviews usage
• AI Mode usage and query trends
• Consumer decisions and the messy middle

The keynote added that people who also use a standalone LLM keep certain types of use anchored on Google (the example given is unclear in both recordings).

Speaker Lino CattaruzziEvidence transcript

• AI Mode usage and query trends

• Extends D1-C007 Day 1: Google says 94% of frequent LLM users are also frequent users of Google Search, as evidence that AI is…

Google said Gen Z users are a very large part of Google Search's user base and search differently from older users, for example with images.

Speaker Gary IllyesEvidence transcript

• How Search is evolving

Google said it noticed young users uploading pictures to reverse image search and expecting results that answer something about the picture, and that Google Lens was its answer to that behaviour.

Speaker Gary IllyesEvidence transcript

• Images

Used byglossary term Google Lens

Google said Search agents in AI Mode had launched globally a day or two before 30 September 2026.

Speaker Gary IllyesEvidence transcript

• Search Agents in AI Mode

Used byglossary term Search agents (information agents)

• Google rolls out AI Mode info monitoring to everyone globally Press Relevant Audience (29 September 2026) · checked 3 October 2026

• Extends D1-C022 Day 1: Search Agents in AI Mode let users set standing requests for updates, such as being told when a product goes…

Gary Illyes said searching by image or by video, as Search offers it today, was not possible before transformers were invented.

Speaker Gary IllyesEvidence transcript

• AI systems already inside Search
• Images

• Extends D1-C161 Day 1: Google said it created the Transformer architecture (the paper 'Attention Is All You Need', the T in ChatGPT)…

Google said there are trillions of URLs on the internet, or even more, that even Google cannot tell how many exist, and that some may never be discovered.

Speaker Cherry PrommawinEvidence transcript

• URL discovery

• Extended by D3-C602 Day 3: Google said it knows hundreds of trillions of URLs (as of October 2026).

Google said its index, printed on paper, would reach the Moon and back twelve times.

Speaker Cherry PrommawinEvidence transcript

• What the index looks like

Google said the more important documents in its index naturally tend to be documents created by humans.

Speaker Gary IllyesEvidence transcript

• AI-written vs human-written content

• Extends D1-C047 Day 1: Google's answer was that ML-based ranking systems are trained on content written by humans for humans, so…

Google said the natural content its ranking algorithms understand and promote better includes content created by humans or at least edited and reviewed by them.

“content that was created by humans, or at least edited and reviewed”

Speaker Gary IllyesEvidence transcript

• AI-written vs human-written content
• Content for people

• Extends D1-C047 Day 1: Google's answer was that ML-based ranking systems are trained on content written by humans for humans, so…

Google said a majority of its search features now use AI.

Speaker Cherry PrommawinEvidence transcript

• AI systems already inside Search
• The results page and search features

For visual search, Google breaks an image down into vectors, sends the vectors to the index and returns results based on them.

Speaker Gary IllyesEvidence transcript

• Images
• Retrieval: finding candidates in the index

• Repeated by D3-C311 Day 3: Google handles an image used as a search query much like a text query interpreted as an embedding: the image…

A community speaker said schema markup helps AI agents interpret a page: on a product page, marking up which number is the price saves the agent from guessing.

Speaker Carlos OrtegaEvidence transcript

• Why structured data still matters
• AI crawlers and agents on your site

• Extended by D1-C315 Day 1: Google's AI optimisation guide says structured data is not required for generative AI search and needs no…

A community speaker said a third-party study from early 2026 found that websites with JSON-LD were shown much more often, and that the industry reacted by adding schema markup everywhere.

Speaker Thiago PojdaEvidence transcript

• Why structured data still matters
• Measuring AI visibility and brand impact

A community speaker said a later controlled study, which added schema markup to pages that lacked it and kept everything else unchanged, found only a negative correlation: the markup had been a sign of teams doing everything else right, not a cause.

Speaker Thiago PojdaEvidence transcript

• Why structured data still matters
• Measuring AI visibility and brand impact

A community speaker cited a third-party study (heard as Peec AI's) finding that 43% of prompts produced fan-out queries in a language other than the one the user searched in.

Speaker Thiago PojdaEvidence transcript

• Query fan-out
• Localisation beyond translation
• Measuring AI visibility and brand impact

• Extended by D1-C504 Day 1: The study cited on stage matches Peec AI's analysis (reported in February 2026) of over 10 million ChatGPT…

A community speaker showed a Spanish-language AI prompt about places to eat in Barcelona whose fan-out queries were all in English, so the content used to answer may be written by and for a different audience than the one asking.

Speaker Thiago PojdaEvidence transcript

• Query fan-out
• Localisation beyond translation

A community speaker said Microsoft is ahead of Google in giving site owners the grounding queries its AI uses.

Speaker not identifiedEvidence transcript

• Measuring AI visibility and brand impact
• Query fan-out

A URL states how a resource is requested (the protocol, HTTP or HTTPS), where (the host, meaning which computer on the network) and what (the path to the exact page or file).

Speaker Cherry PrommawinEvidence transcript

• HTTP versions and Googlebot

AI agents are, technically, the same thing as crawlers: HTTP clients that accomplish something on behalf of a user or a service.

Speaker Gary IllyesEvidence transcript

• AI crawlers and agents on your site

Google runs probably hundreds, if not thousands, of crawlers on its crawler infrastructure; some of them are named and some are not.

Speaker Gary IllyesEvidence transcript

• The crawl pipeline
• AI crawlers and agents on your site

The scheduler hands the crawler an ordered list of URLs from the crawl queue, and the crawler works through the list from top to bottom.

Speaker Gary IllyesEvidence transcript

• The crawl pipeline

Used byglossary term Crawl scheduler

Crawling for Gemini may be set to care less about quality and more about the amount of content, because for large language models the number of tokens matters more than quality.

“the number of tokens is actually more important”

Wording checked against the slide or recording

Speaker Gary IllyesEvidence transcript

• Gemini and Search
• AI crawlers and agents on your site

• Extends D1-C039 Day 1: Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization…

Google described its crawling as a large-scale distributed swarm of simple HTTP clients, roughly what one would get by deploying many wget or curl libraries on cloud compute instances.

“a large-scale distributed swarm of simple HTTP clients”

Wording checked against the slide or recording

Speaker Gary IllyesEvidence transcript

• The crawl pipeline

1xx informational status codes, which only say that a request was received and more data is coming, have no meaning of their own for crawling.

Speaker Cherry PrommawinEvidence transcript

• Crawl errors

Most network errors happen somewhere between the site's origin server and Google's data centers, where they are invisible to both Google and the site owner; the way TCP networks work leaves no way to see what went wrong there.

Speaker Gary IllyesEvidence transcript

• Crawl errors

Used byrequirement DEV-SRV-01

• Extends D1-C069 Day 1: DNS issues and firewalls are blind spots: Google does not know a firewall is blocking it, it only sees that…

Google has seen an uptick in 402 Payment Required responses to its crawlers, an even more recent trend than the rise in 403 responses.

Speaker Gary IllyesEvidence transcript

• Crawl errors
• AI crawlers and agents on your site

Used byrequirement DEV-SRV-09

• How HTTP status codes affect Google's crawlers Google · checked 4 October 2026

• Extends D1-C366 Day 1: Google treats a 402 Payment Required response as a 404 Not Found, because Googlebot cannot pay for content.
• Extends D1-C365 Day 1: Google has recently seen more 403 responses to its crawlers (described on stage as 'authentication…

The robots.txt talk pointed to the robots.txt file of thebestfriedchickenever.com to show why comments in robots.txt are useful; the live file is mostly a large ASCII-art drawing written as # comment lines, followed by a single user-agent group (checked 2026-10-04).

Speaker GoogleEvidence notes, transcript

• robots.txt rules

Other companies have similar 'extended' robots.txt tokens; Google named Apple's Applebot-Extended.

Speaker GoogleEvidence transcript

• The generative AI opt-out in Search Console
• AI crawlers and agents on your site

Used byrequirement DEV-AIF-05

• Repeats D1-C487 Day 1: A Google panelist said Apple offers a robots.txt control for Applebot similar to Google-Extended, so a site…

Google called robots.txt extremely forgiving: a typo in a path only blocks the wrong path, a typo in a rule name such as disallow makes Google ignore that line, and the rest of the file is still used.

Speaker GoogleEvidence transcript

• robots.txt rules

Used byrequirement DEV-SRV-05

• Extended by D1-C532 Day 1: A typo in a rule name is not always ignored by Google: its open-source robots.txt parser deliberately accepts…

The real trouble is an unknown line placed between two user-agent lines: while writing RFC 9309, Google asked people whether the first crawler should then inherit the rules that follow, and opinions split roughly 50-50 (how the talk said the question was settled is unclear in the recording).

Speaker GoogleEvidence transcript

• robots.txt group parsing differences
• Web standards and how to take part

Used byrequirement DEV-SRV-06

• Extends D1-C087 Day 1: Googlebot ignores an unknown line, such as a Content-signal line, placed between two user-agent lines and…
• Extends D1-C123 Day 1: Bing closes the robots.txt group at an unknown line placed between two user-agent lines, so in the slide's…

Google said the robots.txt report helps catch CDNs that update a site's robots.txt without the owner's knowledge, and hosts that cloak the robots.txt file, which happens more often than people think.

Speaker GoogleEvidence transcript

• robots.txt rules
• What's new in Search Console

Used byrequirement DEV-SRV-06

Google said the robots.txt report uses the parser Google open-sourced at github.com/google/robotstxt.

Speaker GoogleEvidence transcript

• robots.txt rules
• What's new in Search Console

Used byglossary term robots.txt report

• Extends D1-C140 Day 1: Search Console's robots.txt report shows the robots.txt files Google found for the top 20 hosts of a Domain…

Dave Smart said robots.txt is checked for every URL in a redirect chain and crawling stops at the first blocked one; in his example a site redirected through /cart/ with JavaScript to set the local currency and back, and because /cart/ was disallowed the page was reported as blocked.

Speaker Dave SmartEvidence transcript

• robots.txt rules
• Crawl errors

Used byrequirement DEV-CAN-11glossary term Redirect chain

Dave Smart said this applies to all redirects, not only JavaScript ones; his examples: a redirect through an external authorisation service that is blocked by its own robots.txt, content that moved through several URLs over the years with one of them later blocked, and unexpected redirects, such as one served only to Googlebot's user agent.

Speaker Dave SmartEvidence transcript

• robots.txt rules
• Crawl errors

Used byrequirement DEV-CAN-11

The quality that drives crawl demand is the quality of the site as a whole, not the quality of an individual page.

Speaker Cherry PrommawinEvidence transcript

• Crawl demand

• Extends D1-C093 Day 1: Crawl demand is driven by the quality of the site, the change frequency of its URLs and their popularity on…
• Extended by D3-C623 Day 3: When Google notices the web getting excited about a few URLs on a site, it allocates more crawl demand to the…

Similar URLs usually come from programming errors or manually set links, but anyone can link to them from other sites, including malicious actors, so a site should be hardened against them.

Speaker Tobias SchwarzEvidence transcript

• Similar URLs and URL normalisation

Used byrequirement DEV-URL-11

A community speaker said that more than half of internet traffic in the previous year (2025) was non-human; the source of the figure was not named.

Speaker Jovana AvramovicEvidence transcript

• AI crawlers and agents on your site
• How Search is evolving

A community speaker cited more than 900 million weekly active ChatGPT users and 2.5 billion monthly users of a Google AI feature (the recording is unclear which), adding that these are not comparable market-share figures.

Speaker Jovana AvramovicEvidence transcript

• How Search is evolving

A community speaker linked the serial-position effect in human memory (people best remember the first and last items of a list) to the 'lost in the middle' pattern that research has found in large language models.

Speaker Jovana AvramovicEvidence transcript

• Chunking content for AI

Used byglossary term Lost in the middle

Crawling for AI model training differs from search crawling because training mainly needs a very large number of tokens, and it matters little which pages they come from.

Speaker not identifiedEvidence transcript

• AI crawlers and agents on your site

Used byrequirement DEV-AIF-05

• Answers D1-C422 Day 1: An audience member asked, in a question submitted at registration, what the biggest difference is between…

A Google panelist said they were working on a set of crawler best practices and offering research, malware-scanning and privacy crawlers an exemption from following them, apparently because such crawlers sometimes need to ignore robots.txt or probe URLs that other crawlers would not touch (the reason is a best reading: the audio has 'don't need', which would not explain an exemption).

Speaker not identifiedEvidence transcript

• Web standards and how to take part
• robots.txt rules

• Extended by D1-C433 Day 1: The crawler best practices mentioned on stage match the public IETF Internet-Draft 'Crawler best practices'…

Log files alone cannot show crawl inefficiency because they do not reveal where Google's crawl limit for the site lies; many similar URLs being crawled may not matter if the site is nowhere near that limit.

Speaker not identifiedEvidence transcript

• Crawl budget

Used byrequirement DEV-MON-04

• Answers D1-C441 Day 1: An audience member asked how to tell, from log files or by other methods, whether Google is spending crawl…

Many crawling complaints that reach Google's crawling team turn out to be caused by a plugin, for example a WordPress calendar plugin that adds calendar parameters to the URLs of every post, page, category and tag page.

Speaker not identifiedEvidence transcript

• Crawl budget
• Faceted navigation

Used byrequirement DEV-URL-08

• Extended by D1-C539 Day 1: A Google panelist said useless plugin-generated parameter URLs can be handled at the web server, for example…

To check that new content is crawled fast enough, Gary Illyes advised measuring the time between publishing a URL and Googlebot's first crawl of it: ideally it stays flat or falls, and only a rising trend calls for investigation.

Speaker Gary IllyesEvidence transcript

• Crawl budget
• How long Google's processes take

Used byrequirement DEV-MON-11

• Answers D1-C465 Day 1: An audience member asked how a large news site can tell whether crawl budget is limiting how fast new…
• Extended by D1-C543 Day 1: Gary Illyes said that if the time from publishing to Googlebot's first crawl starts rising, how much it rises…

The first priority in Google's own site consolidation was to identify the popular URLs people care about and make sure the migration did not damage them.

Speaker not identifiedEvidence transcript

• Redirects, site moves and alternate names

• Answers D1-C476 Day 1: An audience member asked how to plan a site migration so that it does not leave large numbers of URLs not…

A Google panelist said Apple offers a robots.txt control for Applebot similar to Google-Extended, so a site can allow Applebot for search while opting out of Apple's AI use.

Speaker not identifiedEvidence transcript

• The generative AI opt-out in Search Console
• AI crawlers and agents on your site

Used byrequirement DEV-AIF-05

• Answers D1-C481 Day 1: An audience member asked whether Google plans dedicated user agents for AI, as OpenAI has several, so that…
• Repeated by D1-C524 Day 1: Other companies have similar 'extended' robots.txt tokens; Google named Apple's Applebot-Extended.

When logs show an unknown crawler fetching a lot, search for its user agent online to find the robots.txt token that controls it; the mainstream crawlers that send the most traffic can all be controlled this way.

Speaker not identifiedEvidence transcript

• AI crawlers and agents on your site
• robots.txt rules

Used byrequirement DEV-AIF-05

• Answers D1-C481 Day 1: An audience member asked whether Google plans dedicated user agents for AI, as OpenAI has several, so that…

A Google panelist said, as a fun fact, that JavaScript was used for the language-consolidation redirects in Google's own site migration, because it was the only option available to the person doing it (the recording does not make fully clear whether the JavaScript performed the redirects or built the mapping).

Speaker not identifiedEvidence transcript

• Redirects, site moves and alternate names
• JavaScript SEO pitfalls and fixes

Used byrequirement DEV-CAN-01

• Extends D1-C479 Day 1: Where content was duplicated across languages, Google's own site consolidation redirected two language…
• Extended by D1-C542 Day 1: Google's redirects guide says Google Search follows JavaScript location redirects only after rendering, may…

Gary Illyes said that if the time from publishing to Googlebot's first crawl starts rising, how much it rises matters: for the crawling of fresh stories, something like two hours is probably not great (a smaller example figure is unclear in the recording).

Speaker Gary IllyesEvidence transcript

• Crawl budget
• How long Google's processes take

Used byrequirement DEV-MON-11

• Extends D1-C467 Day 1: To check that new content is crawled fast enough, Gary Illyes advised measuring the time between publishing a…

Check and focus on rich results for Google.

Evidence notes

• Rich results

• Extended by D2-C521 Day 2: Structured data makes pages eligible to appear as rich results, Google's structured data talk said in its…

Google contrasted out-of-stock products a buyer will wait for, such as a rare watch battery that matters to them, with easily replaced products, such as a particular cheese, for which the buyer simply picks a similar product instead of waiting.

Speaker not identifiedEvidence transcript

• Redirects, site moves and alternate names

• Answers D2-C009 Day 2: An audience member asked whether a product detail page that is out of stock for two to three months should…
• Extends D2-C010 Day 2: Google answered on a Q&A slide that whether to keep a temporarily out-of-stock product page or redirect it…

Google illustrated why it does not treat a robots.txt disallow as noindex with an extremely important site, such as a national tax authority, that blocks very important PDF files with robots.txt: Google cannot index the PDFs' content but can at least show their URLs in search results.

Speaker not identifiedEvidence transcript

• robots.txt rules
• Directives and crawl budget

Used byrequirement DEV-IDX-01

• Introduction to robots.txt Google Search Central · checked 4 October 2026

• Answers D2-C019 Day 2: An audience member noted that a robots.txt block does not mean noindex and that explaining this to developers…
• Extends D2-C020 Day 2: Google answered on a Q&A slide that it does not treat a robots.txt disallow as a noindex because some…

Google said very few URLs disallowed by robots.txt are in its index, compared with the index as a whole (no figure given).

Speaker not identifiedEvidence transcript

• robots.txt rules
• Index selection: deciding what gets indexed

Used byrequirement DEV-IDX-01

• Answers D2-C019 Day 2: An audience member noted that a robots.txt block does not mean noindex and that explaining this to developers…

The speaker said Google sometimes also extracts URLs that are typed out as plain text on a page without being hyperlinked; the remarks around this point were unclear in the recording.

Speaker Cherry PrommawinEvidence transcript

• Crawlable links
• URL discovery

John Mueller said AI crawlers do not really know what to do with nofollow links, because they look at the content rather than building a link graph; he did not say whether he meant Google's AI systems, other AI crawlers or both.

“AI crawlers don't really know what to do with a nofollow link, because they're looking at the content”

Speaker John MuellerEvidence transcript

• AI crawlers and agents on your site
• Directives and crawl budget

A max-snippet value too short for a useful snippet may lead Google to show no snippet at all, John Mueller said.

Speaker John MuellerEvidence transcript

• Snippets and preview controls

Used byrequirement DEV-IDX-05

John Mueller said Google Search does not always show an image thumbnail for a result, and when it does, the thumbnail is usually limited to the size of the result entry.

Speaker John MuellerEvidence transcript

• Images
• Snippets and preview controls

Compared with other search engines and AI crawlers, Google said its effort to mimic what the user sees is essential to keeping its knowledge of the web up to date and comprehensive; it did not say what the others do.

Speaker Erin SparlingEvidence transcript

• How Google renders pages
• AI crawlers and agents on your site

Being able to render and read JavaScript matters whether the client fetching a page is a search crawler or an AI system, Google said.

Speaker Erin SparlingEvidence transcript

• How Google renders pages
• AI crawlers and agents on your site

Google described the mission of its rendering as simply executing JavaScript, while the implementation is complex, expensive and difficult.

“our mission is very simple these days: execute JavaScript”

Speaker Erin SparlingEvidence transcript

• How Google renders pages

To train Gemini models, Google renders every page just as it does for Search, so a page that renders correctly for Search also works for Gemini training, provided the site allows its content to be used for training.

“if it works for Search, it works for Gemini for training”

Speaker Erin SparlingEvidence transcript

• Gemini and Search
• How Google renders pages
• Google-Extended

Used byrequirement DEV-IDX-11

• Extends D1-C039 Day 1: Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization…

When a Gemini user asks about a specific web page (for example, whether it says anything about the ruby HTML tag), the page is read at that moment rather than during crawling and may be used to ground the answer, which takes extra time.

Speaker Erin SparlingEvidence transcript

• Gemini and Search
• AI crawlers and agents on your site

Used byrequirement DEV-PRF-02

• Extends D1-C039 Day 1: Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization…

After the crawler has fetched a page, Google's rendering system executes it, checks that it loads properly and works out what it looks like at different sizes.

Speaker Erin SparlingEvidence transcript

• How Google renders pages

Google said its pipeline needs to ensure that content indexable without JavaScript can pass through without rendering, while content that appears only through JavaScript and CSS takes a longer rendering pass.

Speaker Erin SparlingEvidence transcript

• How Google renders pages
• JavaScript SEO pitfalls and fixes

Switching JavaScript off and on in the browser gives a quick first impression of which content on a page depends on rendering.

Speaker Sören BendigEvidence transcript

• Debugging rendering

Offer counts, discounts and other data should always be the same in the raw HTML and the rendered page, a community speaker advised.

Speaker Sören BendigEvidence transcript

• JavaScript SEO pitfalls and fixes
• Debugging rendering

Used byrequirement DEV-REN-07

A search engine's renderer does not wait forever before taking its snapshot, so content from slow internal systems or APIs can be missing and unresolved placeholders can end up in the final snapshot, a community speaker said.

Speaker Sören BendigEvidence transcript

• JavaScript SEO pitfalls and fixes
• How Google renders pages

Used byrequirement DEV-REN-06

• Extended by D2-C259 Day 2: Erin Sparling named server-side or hybrid rendering, fallbacks and not leaving placeholders in the DOM, among…
• Extended by D2-C263 Day 2: Content is missing from the rendered HTML either because the server does not serve it or because the content…

Unresolved placeholders caused by rendering timeouts are hard to catch because the problem moves around: it is not always the same page that is broken.

Speaker Sören BendigEvidence transcript

• JavaScript SEO pitfalls and fixes
• Debugging rendering

Used byrequirements DEV-MON-05, DEV-REN-06

To catch intermittent rendering problems, archive the full HTML and the resources of each page during a site audit and analyse them yourself, a community speaker advised.

Speaker Sören BendigEvidence transcript

• Debugging rendering

Used byrequirement DEV-MON-05

Content Security Policy lowers the risk of cross-site scripting and clickjacking by defining trusted hosts; a resource from a host that is not trusted is not used.

Speaker Sören BendigEvidence transcript

• JavaScript SEO pitfalls and fixes

Used byrequirement DEV-REN-09

A page can render completely blank while its raw HTML looks fine, so a team checking only the raw HTML thinks all is well; causes include server misconfiguration, overload and bad scripts.

Speaker Sören BendigEvidence transcript

• JavaScript SEO pitfalls and fixes
• Debugging rendering

Used byrequirement DEV-MON-05

A community speaker strongly advised SEOs to use Chrome DevTools more often to understand what happens when their pages render.

Speaker Sören BendigEvidence transcript

• Debugging rendering

• Extended by D2-C860 Day 2: Besides using Chrome DevTools more often, a community speaker advised checking rendering at scale with a…

Third-party systems billed per session that also run when bots render a page can cost a high-traffic site thousands of euros per month, a community speaker said, adding that he sees this regularly.

Speaker Sören BendigEvidence transcript

• Debugging rendering
• JavaScript SEO pitfalls and fixes

Used byrequirement DEV-PRF-03

If a page keeps loading content indefinitely, not all of that content will be indexed, because Google's rendering does not go on forever (this passage of the recording is partly unclear).

Speaker Rebecca YuEvidence transcript

• How Google renders pages
• JavaScript SEO pitfalls and fixes

Used byrequirement DEV-REN-06

If content is missing from the rendered HTML, find the script responsible in Chrome DevTools: open the Network tab, filter by Fetch/XHR and reload the page.

Speaker Rebecca YuEvidence transcript

• Debugging rendering

If the cause of missing JavaScript content is still unclear after checking the network requests, search the source code for a string related to the missing content.

Speaker Rebecca YuEvidence transcript

• Debugging rendering

Natalia Venditto named three typical failure modes by which an embedded app can break the application hosting it: collisions in the global JavaScript scope and module registry, CSS bleed, and fate sharing.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

CSS bleed, as Natalia Venditto described it, means styles leak between an embedded app and its host, so that suddenly everything looks like the embedded app or the embedded app takes on the look of the host.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Fate sharing means a failure in an embedded app spreads to its host, for example an unhandled exception in the embedded app that ends up breaking the host application.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Web Fragments is a lightweight library that puts a hidden iframe in the host page and uses it as an isolated sandbox to run the embedded app's JavaScript, not to render the app.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Web Fragments fetches the embedded app's assets from a remote endpoint through middleware that acts as a gateway.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Web Fragments reframes the embedded app's DOM by placing it inside a shadow DOM in the host page, while the app's JavaScript runs in the hidden iframe.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

With Web Fragments the page ends up as a single document in which the host application does not know the embedded app runs inside it and the embedded app does not know it lives in another document.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Web Fragments monkey-patches browser APIs to containerize the browser, recreating inside it an architecture similar to Docker containers on the back end.

“we monkey patch the browser to containerize it”

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

The browser APIs Web Fragments patches include document, history and location; Natalia Venditto said they are virtualized rather than hijacked, so controls work exactly as in a normal application.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Web Fragments retargets dispatchEvent calls to the fragment's shadow root and resolves DOM calls such as appendChild and element lookups against the main document.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Natalia Venditto said Web Fragments suits AI-generated apps because a web fragment used as a custom element sandboxes the app's rendering, avoiding collisions and CSS bleed between the app and its host in either direction.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Natalia Venditto said the five Web Fragments steps are all that is needed to run a containerized application fully on the client side.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Natalia Venditto said Web Fragments is framework-agnostic and vendor-agnostic and works the same whatever JavaScript framework the embedded app uses.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments

Natalia Venditto reported an enormous performance increase (no figure given) when an application is reframed inside a fully client-side host, because the reframed app is rendered and interactive immediately.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments
• JavaScript SEO pitfalls and fixes

Natalia Venditto said a server-rendered app reframed into the host page is fully indexable.

Speaker Natalia VendittoEvidence transcript

• Embedding apps safely: Web Fragments
• How Google renders pages

Used byrequirement DEV-REN-01

Natalia Venditto said ShadowRealm, a JavaScript standard proposal, was at stage 2.7 at the time of the talk, which she glossed as meaning that only implementation is missing.

“the only thing missing is implementation”

Speaker Natalia VendittoEvidence transcript

• Web standards and how to take part
• Embedding apps safely: Web Fragments

Natalia Venditto said the ShadowRealm proposal would do much of what Web Fragments does today with patches: containerizing and isolating execution.

Speaker Natalia VendittoEvidence transcript

• Web standards and how to take part
• Embedding apps safely: Web Fragments

• Extended by D2-C281 Day 2: Polyfills work in two directions: they patch backwards where a browser lacks support, or patch forward to…

Search engines such as Google and Bing would ignore HTML tags written inside a meta description, a community speaker said.

Speaker Sören BendigEvidence transcript

• JavaScript SEO pitfalls and fixes
• Snippets and preview controls

Used byrequirement DEV-HTM-03

• Extends D2-C142 Day 2: One site that builds its whole content by rendering also rendered its meta description with HTML tags inside…

Besides using Chrome DevTools more often, a community speaker advised checking rendering at scale with a modern crawler.

Speaker Sören BendigEvidence transcript

• Debugging rendering

Used byrequirement DEV-MON-05

• Extends D2-C162 Day 2: A community speaker strongly advised SEOs to use Chrome DevTools more often to understand what happens when…

Content is missing from the rendered HTML either because the server does not serve it or because the content has still not appeared after some period of time.

Speaker Erin SparlingEvidence transcript

• JavaScript SEO pitfalls and fixes
• How Google renders pages

Used byrequirement DEV-REN-06

• Extends D2-C149 Day 2: A search engine's renderer does not wait forever before taking its snapshot, so content from slow internal…

Instead of scrolling, Google renders a page in a very tall viewport, around 10,000 pixels high.

“it renders it in a very tall viewport. Specifically, around 10,000 pixels is what the viewport gets rendered as.”

Speaker Erin SparlingEvidence slide photo, transcript

• How Google renders pages
• Lazy loading and content behind interactions

Used byrequirements DEV-REN-03, DEV-URL-07glossary term Viewport expansion

• Extends D2-C199 Day 2: An Intersection Observer fires when the observed element enters the viewport, and the viewport Google renders…

Polyfills work in two directions: they patch backwards where a browser lacks support, or patch forward to build now for a feature expected in the future, as Web Fragments does.

Speaker Erin SparlingEvidence transcript

• JavaScript SEO pitfalls and fixes
• Web standards and how to take part
• Embedding apps safely: Web Fragments

• Extends D2-C254 Day 2: Natalia Venditto said the ShadowRealm proposal would do much of what Web Fragments does today with patches…

Erin Sparling called URL fragments (the part of a URL after #) the next most common JavaScript issue seen, after rendering problems and blocked rendering.

Speaker Erin SparlingEvidence transcript

• Crawlable links
• JavaScript SEO pitfalls and fixes

Real URLs also work as deep links: Android, iOS and desktop operating systems accept full URLs as keys to specific content in an app, so clean URLs simplify the cross-platform user experience, not only indexing.

Speaker Erin SparlingEvidence transcript

• Crawlable links
• JavaScript SEO pitfalls and fixes

Used byrequirement DEV-URL-03

A headless content management system serves its content as an API; even WordPress, often seen as one monolithic application, can be used headless, with only its editor or only its renderer.

Speaker Erin SparlingEvidence transcript

• JavaScript SEO pitfalls and fixes

Combining schema-based interfaces with web standards such as WebMCP lets a site expose tools to AI agents, which can then operate the interfaces, for example to test them.

Speaker Erin SparlingEvidence transcript

• AI crawlers and agents on your site
• Web standards and how to take part
• Debugging rendering

• Extends D1-C281 Day 1: A community speaker presented WebMCP (Web Model Context Protocol) as a way for a site to declare the actions…

Erin Sparling showed a Search Console result saying a URL would be indexed only under certain conditions, one of which had not been met, and contrasted it with the result for an available URL whose content loads.

Speaker Erin SparlingEvidence transcript

• Indexing statuses in Search Console
• Debugging rendering

Words in the footer of a page get a lower weight, so text placed in the footer is unlikely to contribute much to ranking the page.

“if you put something in a footer, it's more likely that it's not going to contribute much to ranking”

Speaker Gary IllyesEvidence transcript

• Main content and page sections

Used byrequirement DEV-HTM-02

Gary Illyes said not everything on a page can or should be important: if everything were placed in the main content, nothing would stand out as main content, which he called working as intended.

Speaker Gary IllyesEvidence transcript

• Main content and page sections

Google does not store the complete sentences or the full HTML of a page in the Search index, because large pieces of text would be unsearchable; it tokenizes the text into the smallest segments that still allow search.

Speaker Gary IllyesEvidence slide photo, transcript

• Tokenization: how text is stored
• What the index looks like

Used byglossary term Tokenization

• Repeated by D2-C721 Day 2: Google's Search index does not hold the full content of pages; Google said storing full pages and pulling…
• Extended by D3-C109 Day 3: The inspector metaphor blends two steps: Googlebot fetches pages during crawling, while tokenization happens…

For languages written with spaces between words, such as English and German, Search tokenization splits a sentence into its individual words.

Speaker Gary IllyesEvidence transcript

• Tokenization: how text is stored

Text in languages written without spaces, such as Thai and Chinese, would end up in the index as long strings that might never be searched for, so Google segments it into words with statistical models built from other web content in that language.

Speaker Gary IllyesEvidence transcript

• Tokenization: how text is stored
• Non-Latin scripts and word segmentation

• Extended by D3-C011 Day 3: Google named Thai as a language that makes query understanding more complex because it does not separate…
• Repeated by D3-C070 Day 3: Google's summary slide on query understanding noted that some languages do not use spaces between words…

For languages written without spaces, such as Thai and Chinese, Google uses exactly the same word segmentation when indexing a page as when interpreting the user's query, because otherwise the query could not be matched against the index.

Speaker Gary IllyesEvidence transcript

• Tokenization: how text is stored
• Non-Latin scripts and word segmentation

• Repeated by D2-C737 Day 2: A search query is broken into words with the same segmenter or tokenizer that Google used to build the index.
• Extended by D3-C013 Day 3: Google's query processing deliberately mirrors indexing: a query is transformed into something that can be…

When tokenizing for Search, Google attaches metadata to each token for use in ranking, such as whether the word appeared in the header or in the main content (tagged 'centerpiece' on the slide), in bold or in a heading; Gary Illyes said he was not showing all of the metadata.

Speaker Gary IllyesEvidence slide photo, transcript

• Tokenization: how text is stored
• Main content and page sections
• What the index looks like

Used byrequirement DEV-HTM-07

• Repeated by D2-C720 Day 2: Google's Search index stores the tokens produced by tokenizing each page, as they are, together with the…

Google also stores spam metadata with the tokens of a page, for example that text was white on a white background, so ranking can use that information.

Speaker Gary IllyesEvidence transcript

• Tokenization: how text is stored
• Spam detection and SpamBrain

Used byrequirement DEV-HTM-06

Tokenization for AI models such as Gemini, in training and in inference, differs from tokenization for Search, although Gary Illyes qualified this with 'or mostly'.

Speaker Gary IllyesEvidence slide photo, transcript

• Tokenization: how text is stored
• Gemini and Search

• Contradicts D1-C039 Day 1: Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization…

Gary Illyes added that, once chunk size is thought of in millions of tokens as Gemini's context window allows, chunking has perhaps lost its meaning anyway.

Speaker Gary IllyesEvidence transcript

• Chunking content for AI

Used byrequirement DEV-AIF-03

• Extends D2-C332 Day 2: Gemini does not need content cut into small chunks of 100 or 200 words, Gary Illyes said, since a smaller…

Because error pages are worded in endless variations, of which 'page not found' is only the classic one, Google cannot detect soft 404s with simple error, word or keyword matching.

Speaker Gary IllyesEvidence transcript

• Soft 404s

Google detects soft 404s with a language model, described as something like BERT, that is trained to understand the structure and layout of a page as well as its language, instead of reading the page as one flat wall of text.

“This is basically an LLM thing, something like BERT, that is specifically trained to understand page structure”

Speaker Gary IllyesEvidence transcript

• Soft 404s
• AI systems already inside Search
• Main content and page sections

• Extends D1-C042 Day 1: BERT is used in indexing to understand each word in the context of the whole sentence rather than one word at…

For soft 404 detection, the position of error text decides: an error in a less important part such as the navigation need not make a page a soft 404, but an error message alone in the main content, like 'Error establishing a database connection', does, much as a human would judge it.

Speaker Gary IllyesEvidence slide photo, transcript

• Soft 404s
• Main content and page sections

Used byrequirement DEV-ERR-03

• Extends D1-C078 Day 1: For temporary blocks return 503 or 429, never a 200 page with a captcha or error message, which becomes a…

Google's soft 404 detection tells the page chrome, such as navigation and footer, apart from the main content by analysing the page's visual hierarchy alongside its text.

Speaker Gary IllyesEvidence transcript

• Soft 404s
• Main content and page sections

Storage is a second reason for deduplication: Google's storage has many competing uses and storage prices have risen sharply, so the space for any one use is limited and Google has to draw a line somewhere.

Speaker John MuellerEvidence transcript

• Duplicate content
• Index selection: deciding what gets indexed

Google clusters duplicate pages by content in four ways: exact matches, near matches, structurally similar content, and soft 404s.

Speaker John MuellerEvidence transcript

• Duplicate content

AI agents that browse the web for users run into the same bot walls that sites put up against scrapers, and may give up and go to another site, for example to buy the product elsewhere.

Speaker John MuellerEvidence transcript

• AI crawlers and agents on your site
• Crawl errors

Used byrequirement DEV-AIF-04

• Extends D1-C436 Day 1: A Google panelist argued that it makes business sense for agents not to follow robots.txt: an agent sent to…

Google picks the canonical from a variety of criteria and uses some kind of machine learning to decide how much weight each criterion gets; the weighting changes from time to time.

“we use some kind of machine learning to understand how strong these criteria should be. And this changes from time to time.”

Speaker John MuellerEvidence transcript

• Canonical selection
• AI systems already inside Search

Google's speaker said that pointing rel=canonical from the pages of a paginated set to the first page can sometimes make sense depending on the goal, for example to make a category page more visible, but that it affects canonicalization and deduplication.

Speaker John MuellerEvidence transcript

• Pagination and "load more"
• Canonical selection

Used byrequirement DEV-URL-06

• Contradicts D1-C115 Day 1: Google's crawlers do not click buttons. Each page in a series needs its own URL and an <a href> link to the…

Google's speaker suggested checking rel=canonical links with a crawler such as Screaming Frog to make sure they are reasonable.

Speaker John MuellerEvidence transcript

• Canonical selection
• Canonical graphs: chains, loops and leaders

Used byrequirement DEV-MON-06

• Extended by D2-C408 Day 2: A community speaker advised auditing canonicals as whole canonical groups rather than pair by pair, paying…

Canonical chains can run through canonical links alone, through a server-side redirect between members, or through a client-side redirect between members.

Speaker Tobias SchwarzEvidence transcript

• Canonical graphs: chains, loops and leaders
• Redirects, site moves and alternate names

Canonical loops, such as two HTML pages whose canonical tags point to each other, are a structural conflict: the group has no canonical leader, so its canonical information cannot be used.

Speaker Tobias SchwarzEvidence transcript

• Canonical graphs: chains, loops and leaders
• Canonical selection

Used byrequirement DEV-CAN-04glossary term Canonical chain and canonical loop

Canonical loops can also run through server-side or client-side redirects, not only through canonical tags.

Speaker Tobias SchwarzEvidence transcript

• Canonical graphs: chains, loops and leaders
• Redirects, site moves and alternate names

A community speaker said that at best a search engine's heuristic would treat a canonical loop through redirects as self-referencing canonicals, and doubted that this is often done when the loop runs through a client-side redirect.

“But regarding the client-side redirect, I highly doubt that this is often done.”

Speaker Tobias SchwarzEvidence transcript

• Canonical graphs: chains, loops and leaders
• Canonical selection

A community speaker said pages carry different link equity, and a canonical leader that is not the strongest page in its group is technically valid but most likely suboptimal for ranking, so the strongest page should be the leader.

Speaker Tobias SchwarzEvidence transcript

• Canonical graphs: chains, loops and leaders
• Canonical selection

• Contradicts D2-C348 Day 2: Google forwards the signals attached to every URL in a duplicate cluster, such as links, to the…

Google's indexing includes a dedicated system, whose internal name Gary Illyes would not disclose, that extracts the parts of a page that are traditionally expensive to extract.

Speaker Gary IllyesEvidence transcript

• Processing: from fetched page to index

Google runs deduplication before feature extraction, so that the expensive extraction work is spent only on a trimmed-down, manageable set of documents.

Speaker Gary IllyesEvidence transcript, slide photo

• Processing: from fetched page to index
• Duplicate content

Used byrequirement DEV-SDA-10

• Extends D2-C026 Day 2: A Google pipeline slide placed processing between the crawler and the index and listed six processing steps…

Gary Illyes said feature extraction, which extracts page structures into a form Google's systems can consume internally, is still expensive, though not the most expensive operation.

Speaker Gary IllyesEvidence transcript

• Processing: from fetched page to index

AI Overviews and AI Mode launched as fairly text-heavy answers with little image content and few tables, and they have become more structured over time because that is what users want.

Speaker Ryan LeveringEvidence transcript

• How Search is evolving
• Why structured data still matters
• Google's five ecosystem principles

• Extends D1-C011 Day 1: Ecosystem principle 2, the SERP will evolve: besides organic links and ads, it will hold other elements such…

In the speaker's own tests, even the latest LLMs asked to generate schema.org markup for a page often invent properties that do not exist, get deeply nested schemas such as complex pricing models wrong, and duplicate content across several fields.

Speaker Ryan LeveringEvidence transcript

• Why structured data still matters
• Schema.org vocabulary

Used byrequirement DEV-SDA-11

Automatic LLM extraction of structured information makes a great demo but is not yet good enough for extraction that aims at something like 99.9% accuracy, the speaker said.

Speaker Ryan LeveringEvidence transcript

• Why structured data still matters
• AI systems already inside Search

Google's own extraction systems focus heavily on what is visible on a page, which improves precision but means they can miss content or interpret it incorrectly.

Speaker Ryan LeveringEvidence transcript

• Why structured data still matters
• Main content and page sections

Even Google cannot afford to run complex AI models on every page in its index, and distilling them into cheaper models makes them less precise.

Speaker Ryan LeveringEvidence transcript

• AI systems already inside Search
• Why structured data still matters

Rule-based parsing of markup is nearly free by comparison with AI models, so Google will always prefer extracting information from structured data over model-based extraction.

“So we're always going to prefer that particular approach.”

Speaker Ryan LeveringEvidence transcript

• Why structured data still matters

Google's systems sometimes fail to determine a page's main content correctly, and structured data helps because site owners tend to mark up what is actually important rather than boilerplate or ads.

Speaker Ryan LeveringEvidence transcript

• Main content and page sections
• Why structured data still matters

Used byrequirement DEV-SDA-07

• Extends D2-C028 Day 2: Google extracts every element of a page so that it can tell the header, the navigation and the main content…

As far as the speaker knows, in most of Google's main AI uses a page's schema.org markup is not turned into text and put directly into the model's context; the data is first sorted out, checked for quality and indexed before it is passed on as grounding context.

Speaker Ryan LeveringEvidence transcript

• Why structured data still matters
• SEO vs GEO

• Extends D1-C061 Day 1: Google's guide says new machine-readable files, AI text files or special markup are not needed to appear in…

The speaker said there are concerns that feeding raw schema.org markup straight into an AI model's context would be a big abuse vector, one reason Google does not do so.

Speaker Ryan LeveringEvidence transcript

• Why structured data still matters
• SEO vs GEO

Microdata has an advantage when payload size matters: embedded in the existing HTML, it avoids duplicating the page content in a separate JSON-LD block.

Speaker Ryan LeveringEvidence transcript

• Implementing structured data

Used byrequirement DEV-SDA-01

Several plug-ins emitting the same markup type is one of the most common structured data problems: the duplicates can make an event details page look like a list of events and change how Google interprets the page.

Speaker Ryan LeveringEvidence transcript

• Implementing structured data

Used byrequirement DEV-SDA-06

Schema.org, in which Google is a major participant, began publishing usage statistics for all its types and properties in 2026, showing in buckets how many domains use each one; the data is also in schema.org's GitHub repository.

Speaker Ryan LeveringEvidence transcript

• Schema.org vocabulary

Google often needs to see a markup type adopted on websites before it invests in a feature that uses it, a chicken-and-egg problem that the schema.org usage statistics are meant to help break.

Speaker Ryan LeveringEvidence transcript

• Schema.org vocabulary
• Requesting features from Google

• Extends D1-C120 Day 1: To get Google to build something, such as an addition to an API, report and request it publicly and in volume.

Schema.org added support for RDF lists and sets, which give a way to express ordered values, because RDF triples are not ordered by nature.

Speaker Ryan LeveringEvidence transcript

• Schema.org vocabulary

The SHACL rules are meant to run inside a site's content generation, so markup is sanity-checked before it is published and does not silently regress later, a breakage site owners might otherwise discover only through a Search Console report.

Speaker Ryan LeveringEvidence transcript

• Implementing structured data

Used byrequirement DEV-SDA-12glossary term SHACL

The SHACL rules will not replace Search Console as the canonical place for structured data reports, because some checks use Google's internal libraries and cannot be expressed in SHACL, but they will catch problems such as a missing required field.

Speaker Ryan LeveringEvidence transcript

• Implementing structured data

Used byrequirement DEV-SDA-12

Google plans a SHACL rule set for each of its structured data feature types and will release the rule sets gradually once they have been checked.

Speaker Ryan LeveringEvidence transcript

• Implementing structured data

Used byrequirement DEV-SDA-12

Gary Illyes suggested a div with a CSS background image as a way to keep an image from being picked up by Google.

Speaker Gary IllyesEvidence transcript

• Images
• Directives and crawl budget

Used byrequirement DEV-IMG-06

Of the attributes the HTML standard defines for the img element, Google uses three and ignores the rest; Gary Illyes named src and alt but not the third.

Speaker Gary IllyesEvidence transcript

• Images

Gary Illyes ranked the alt attribute below src in importance, while saying alt attributes are still important.

Speaker Gary IllyesEvidence transcript

• Images

Gary Illyes said the AVIF image format currently has hiccups and Google may have problems ingesting it, although it should technically be supported.

Speaker Gary IllyesEvidence transcript

• Images

Used byrequirement DEV-IMG-05

Gary Illyes cited a figure that more than 40% of Southeast Asian shoppers rely on videos to make purchase decisions.

Speaker Gary IllyesEvidence transcript

• Video in Search

Gary Illyes said over 219 million people in Southeast Asia consume content on or through YouTube daily (the daily scope was heard in two independent recordings but is not verified).

Speaker Gary IllyesEvidence transcript

• Video in Search

Gary Illyes said there are over 150 streaming apps in Southeast Asia.

Speaker Gary IllyesEvidence transcript

• Video in Search

Google knows from experiments that when a video's thumbnail is wrong, the share of viewers who drop out at the start of the video is extremely high.

Speaker Gary IllyesEvidence transcript

• Video in Search

Used byrequirement DEV-VID-03

Gary Illyes called the 'fast-loading pages' factor for video SEO a misnomer: what matters is that the video itself loads fast, because people no longer have the patience to wait for videos to load.

Speaker Gary IllyesEvidence transcript

• Video in Search

Used byrequirement DEV-VID-06

Serving videos from a content delivery network (CDN) that loads them faster than the site's own server is a win for video SEO, Gary Illyes said.

Speaker Gary IllyesEvidence transcript

• Video in Search

Used byrequirement DEV-VID-06

Google recommends the MP4 container for videos because of its browser and device compatibility.

Speaker Gary IllyesEvidence transcript

• Video in Search

Used byrequirement DEV-VID-06

Encode videos with standard codecs, because some people will not be able to play videos that use unusual ones.

Speaker Gary IllyesEvidence transcript

• Video in Search

Used byrequirement DEV-VID-06

For a video to be discovered it must be embedded prominently, above the fold; Gary Illyes said a video placed below the fold is not going to be indexed.

“If it's not above the fold, then you basically lost the game.”

Speaker Gary IllyesEvidence transcript

• Video in Search

Used byrequirement DEV-VID-01

Video sitemaps are not critical but good to have, because Google ingests video sitemaps much more often than it can process HTML pages.

Speaker Gary IllyesEvidence transcript

• Sitemaps
• Video in Search

Used byrequirement DEV-VID-04glossary term Video sitemap

Descriptive text around a video helps Google rank and retrieve the video.

Speaker Gary IllyesEvidence transcript

• Video in Search
• Retrieval: finding candidates in the index

Used byrequirement DEV-VID-02

The more people talk about a site's videos, the more likely Google is to surface them in search results, so videos should be promoted.

Speaker Gary IllyesEvidence transcript

• Video in Search

Google's media indexer processes the images and videos that feature extraction passes to it and attaches them to the URL of the page that hosts them.

Speaker Gary IllyesEvidence transcript

• Processing: from fetched page to index
• Images
• Video in Search

• Extends D2-C448 Day 2: For videos, Google's feature extraction takes the video itself and the data around it, to get a better sense…

Unlike a disallowed web page, whose bare URL can still appear in web results, a video blocked by robots.txt is not shown by its URL, because Google sees no good reason to show a video URL alone.

Speaker Gary IllyesEvidence transcript

• robots.txt rules
• Directives and crawl budget
• Video in Search

Used byrequirement DEV-VID-06

• Extends D2-C021 Day 2: Google's Q&A slide said Google does not index the content of pages disallowed in robots.txt, but users will…

Gary Illyes said noimageindex also affects videos on the page, because Google has to index a video's thumbnail, which is an image.

Speaker Gary IllyesEvidence transcript

• Robots meta tags and noindex
• Video in Search
• Images

Used byrequirement DEV-IDX-09glossary term noimageindex

Google serves AI-generated images and videos in search results when users are specifically looking for them.

“We will serve users AI-generated images and videos in search results if they are looking for them specifically.”

Speaker Gary IllyesEvidence transcript

• AI-written vs human-written content
• Images
• Video in Search

The diffusion models that generate images were built to generate images, not text, so they are typically poor at rendering text inside an image.

Speaker Gary IllyesEvidence transcript

• AI systems already inside Search
• Images

Used byrequirement DEV-IMG-08

• Extended by D3-C688 Day 3: Generative models, including image diffusion models, make things up when they lack information or because of…

Version 20 of the Screaming Frog SEO Spider, released in May 2024, was the first that could connect the crawler to ChatGPT through custom JavaScript, which let SEOs rewrite image alt text at scale.

Speaker not identifiedEvidence transcript

• Images
• Accessibility and ARIA

To test LLM-generated alt text before a rollout, a community speaker recommended running the crawler's custom JavaScript in Screaming Frog's List mode on a few chosen URLs and reviewing the generated alt text, instead of crawling the whole site in Spider mode.

Speaker not identifiedEvidence transcript

• Images
• Accessibility and ARIA

Used byrequirement DEV-IMG-02

A community speaker cited W3C guidance that, when cleaning a transcript, it is usually not appropriate to correct the speaker's grammar or other mistakes.

Speaker Patrick DomanicoEvidence transcript

• AI in SEO and content workflows
• Accessibility and ARIA

Google annotates each page with only one language in the index.

Speaker GoogleEvidence transcript

• Page language and language versions
• Signals calculated during indexing

Server location is not really a reliable country-targeting signal nowadays, so Google does not use it much; the recording is unclear on the word 'server'.

Speaker GoogleEvidence transcript

• Country targeting

Google usually understands non-English words typed 'in English' (probably meaning romanised, Latin-letter spellings, which the speaker did not spell out); the speaker said this belongs to query interpretation, a Day 3 (serving) subject.

Speaker GoogleEvidence transcript

• Non-Latin scripts and word segmentation

• Extended by D2-C965 Day 2: Users of non-Latin-script languages do not always search in their own script: the same Persian query may be…
• Extended by D3-C051 Day 3: Users expect content written the way they search: in some languages they search in Latin characters, in…

Whether machine-translated content is acceptable depends on the case and is the site owner's decision, after weighing three things machine translation can miss: translation quality, local conventions and cultural adaptation.

Speaker GoogleEvidence transcript

• Localisation beyond translation

Used byrequirement DEV-INT-10

Localisation should account for local conventions such as date formats, which differ between Europe, the UK, the US and other countries, and calendars (in Thailand the current year is 2569), or a date can point users to the wrong day.

Speaker GoogleEvidence transcript

• Localisation beyond translation

Used byrequirements DEV-INT-10, DEV-SDA-05

According to the speaker, shoppers in Europe and the US pay a lot of attention to promotions and discounts when deciding to buy (no source for this was captured), one of the cultural factors localisation should take into account.

Speaker GoogleEvidence transcript

• Localisation beyond translation

According to consumer data shown on a slide in the talk (source not captured), US consumers judge product quality more by user feedback and reviews, while European shoppers seem to look more at brand reputation.

Speaker GoogleEvidence transcript

• Localisation beyond translation

• Extended by D3-C341 Day 3: Some countries rely a lot on social proof such as reviews, Google's speaker said, recalling a point from…

According to consumer data shown on a slide in the talk (source not captured), German shoppers also rely heavily on expert recommendations and certifications when judging product quality.

Speaker GoogleEvidence transcript

• Localisation beyond translation

According to consumer data shown on a slide in the talk (source not captured), French shoppers particularly want to know where a product comes from.

Speaker GoogleEvidence transcript

• Localisation beyond translation

A community speaker cited a Semrush study that found at least one hreflang error on 75% of multilingual websites; the study, reported in February 2017, audited 20,000 sites with several language versions (the speaker recalled it as about five years old).

Speaker Alizée BaudezEvidence transcript

• hreflang annotations

• 75% of Multilingual Websites Have Hreflang Implementation Mistakes Press Search Engine Journal (15 February 2017) · checked 2 October 2026

Arabic and Persian contain letters that look identical to users but are different characters to software, with different Unicode code points; the example given was the letter ye, which has an Arabic and a Persian form.

Speaker a second community speakerEvidence transcript

• Non-Latin scripts and word segmentation

Because Arabic and Persian lookalike letters have different code points, a searcher may type one variant of a word while a website's text uses the other.

Speaker a second community speakerEvidence transcript

• Non-Latin scripts and word segmentation
• Localisation beyond translation

Lookalike Arabic and Persian character variants cause problems for data analysis and keyword research: data for one term can be split across the variants, which makes keyword research less accurate.

Speaker a second community speakerEvidence transcript

• Non-Latin scripts and word segmentation
• Localisation beyond translation

Persian and Arabic are written right to left and English left to right, so a title that mixes the two scripts can display in a confusing, unpredictable order even when its content is correct.

Speaker a second community speakerEvidence transcript

• Non-Latin scripts and word segmentation

Used byrequirement DEV-INT-12glossary term Bidirectional text

The display problem of mixing right-to-left and left-to-right text also affects product titles and URLs, the presenter of the non-Latin-script talk said, showing an example from a large Iranian e-commerce site.

Speaker a second community speakerEvidence transcript

• Non-Latin scripts and word segmentation
• Localisation beyond translation

Used byrequirement DEV-INT-12glossary term Bidirectional text

The presenter of the non-Latin-script talk said that in competitive Persian, Turkish and Arabic searches, bought backlinks and paid editorial content still visibly influence rankings and are widespread (an observation; no data was shown); the presenter stressed this described the situation and was not a recommendation.

Speaker a second community speakerEvidence transcript

• Spam policies and manual actions
• Non-Latin scripts and word segmentation

Smaller languages have fewer documents on the web and so worse coverage, from an indexing perspective, than large ones, for example Basque compared with Spanish or Portuguese, Hungarian compared with German, and Sicilian compared with Italian.

Speaker GoogleEvidence transcript

• Page language and language versions
• Index selection: deciding what gets indexed

Google detects the language of each document and weights language in index selection so that the index is not dominated by one or two languages.

Speaker GoogleEvidence transcript

• Page language and language versions
• Index selection: deciding what gets indexed
• Signals calculated during indexing

• Extends D2-C585 Day 2: Google determines a page's language for indexing from the page content, not from a language code in the URL.

Without language or country input to index selection, Google's index would become an English index, because so many documents on the web are in English.

“if you just let the index decide what to rank without language or country input, then you would have an English index”

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Page language and language versions

Google also takes country into account in index selection, so that countries producing less content in a shared language, such as the UK for English or Switzerland for German, are not disadvantaged against the US or Germany, which produce most of it.

Speaker GoogleEvidence transcript

• Country targeting
• Index selection: deciding what gets indexed

Freshness plays no big role in index selection: whether a page was published 30 years ago or today makes no difference to its chance of being indexed.

“It doesn't play a big role in index selection.”

Speaker GoogleEvidence transcript

• Signals calculated during indexing
• Index selection: deciding what gets indexed

Whether a page's content is explicit does not matter for indexing: judged on content type alone, an explicit page has the same chance of being indexed as a major news site's homepage.

Speaker GoogleEvidence transcript

• Signals calculated during indexing
• Index selection: deciding what gets indexed

Google calculates SafeSearch signals during indexing rather than in ranking, because ranking happens online with no time for such calculations, while indexing has the processing power.

“In ranking, everything has to happen online, and there's just not enough time to calculate those things.”

Speaker GoogleEvidence transcript

• Signals calculated during indexing

SpamBrain, Google's AI-based spam detection system, is nowadays built on Gemini and fine-tuned specifically for finding spam.

“built on top of, well, nowadays, Gemini, and fine-tuned to that specific purpose of finding spam”

Speaker GoogleEvidence transcript

• Spam detection and SpamBrain
• AI systems already inside Search
• Gemini and Search

Used byglossary term SpamBrain

• Extends D1-C041 Day 1: Statistical models have been used at Google for over 20 years, for catching spam and originally for the 'Did…
• Extends D1-C039 Day 1: Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization…

SpamBrain was said to detect 5 times more spam sites than the spam algorithms Google had launched before it.

Speaker GoogleEvidence transcript

• Spam detection and SpamBrain

Google gives two reasons for not indexing every URL it knows: most of them would not be useful to users, and including URLs in the index that users would never see would be an immense investment.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed

Index selection aims to index only documents that are useful to users now or potentially in the future.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Content for people

Index selection is a predictive AI system that relies heavily on machine learning.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• AI systems already inside Search

• Extends D1-C037 Day 1: For classic Search, crawling means Googlebot, scheduling and robots.txt, with AI used in parts such as…

Index selection uses what Google already knows about a site: if the site, or even a section of it, satisfies users' needs well, new URLs from it are treated more forgivingly.

“the index selection system is going to be more forgiving when it sees a new URL from your site”

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Crawl demand inherited by path

Used byrequirement DEV-URL-10

• Extends D1-C094 Day 1: If the quality or popularity of a URL is unknown, the aggregate quality or popularity of its parent path is…

Index selection uses the signals calculated earlier in indexing for each document it has to select or discard.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Signals calculated during indexing

Used byglossary term Index selection

• Extends D1-C205 Day 1: Signals calculated for a page during indexing are stored in the index and used both to decide whether the…

Index selection is the last step before documents enter Google's index.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Processing: from fetched page to index

Used byglossary term Index selection

• Extends D2-C026 Day 2: A Google pipeline slide placed processing between the crawler and the index and listed six processing steps…
• Extends D1-C211 Day 1: Index selection runs after signals are collected and duplicates are dropped, and decides what goes into…

When Google's coverage of a country is limited, index selection becomes more likely to select documents relevant to that country, even if other signals would suggest otherwise.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Country targeting

When Google lacks content in a language, such as Basque, index selection becomes more likely to select lower-quality documents in that language.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Page language and language versions

Under-served languages were presented as an opportunity: where Google's index holds a lot of spam in a language such as Basque, a site that starts publishing in that language can very likely replace that spam with its own content and rank for those keywords (one clause of the reasoning was inaudible).

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed
• Page language and language versions
• Localisation beyond translation

The overall importance of a document or site plays a significant role in index selection.

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed

News sites were given as the example of importance at work in index selection: they are generally very important on the web and their pages usually get indexed very fast ('indexed' is a likely but not certain reading of the recording).

Speaker GoogleEvidence transcript

• Index selection: deciding what gets indexed

'Discovered – currently not indexed' in Search Console is a crawl-scheduling state: Google knows the URL exists but does not want to crawl it yet. Of the two not-indexed statuses discussed, it was called the 'kind of nastier' one.

“The first one is kind of nastier.”

Speaker GoogleEvidence transcript

• Indexing statuses in Search Console
• Crawl budget

Used byrequirement DEV-MON-03glossary term Discovered – currently not indexed

• Extends D1-C097 Day 1: Google's crawl budget guide is written for sites with over 1 million unique pages that change weekly, over…
• Extends D1-C065 Day 1: The scheduler is shared infrastructure that decides what to fetch and when and sends URLs to the crawler.…

'Crawled – currently not indexed' is most of the time a quality issue rather than a technical one: the pages are usually low quality or useless for the index, for example duplicates or soft 404s.

“most of the time it is actually a quality issue”

Speaker GoogleEvidence transcript

• Indexing statuses in Search Console
• Index selection: deciding what gets indexed
• Content for people

Used byrequirement DEV-MON-03

• Extends D1-C098 Day 1: On smaller sites, slow indexing is almost always a demand problem, meaning quality, not a capacity problem.

The first thing to check for pages reported as 'Crawled – currently not indexed' is whether their quality is on par with the parts of the site that Google does index.

Speaker GoogleEvidence transcript

• Indexing statuses in Search Console
• Content for people

Google's Search index stores the tokens produced by tokenizing each page, as they are, together with the metadata attached to the tokens during tokenization.

Speaker GoogleEvidence transcript

• What the index looks like
• Tokenization: how text is stored

Used byrequirement DEV-HTM-07

• Repeats D2-C323 Day 2: When tokenizing for Search, Google attaches metadata to each token for use in ranking, such as whether the…

Google's Search index does not hold the full content of pages; Google said storing full pages and pulling them out at serving time would be a very inefficient way of doing search.

“we don't have the full content of the page in our index”

Speaker GoogleEvidence transcript

• What the index looks like
• Tokenization: how text is stored

• Repeats D2-C318 Day 2: Google does not store the complete sentences or the full HTML of a page in the Search index, because large…

The snippet shown for a web result is reconstructed from the tokens stored in Google's index: Google knows the position of each token in the document and rebuilds the snippet from those positions.

“the snippet that you see was reconstructed from these tokens”

Speaker GoogleEvidence transcript

• What the index looks like
• Snippets and preview controls
• Tokenization: how text is stored

• Extended by D3-C316 Day 3: Google generates the parts of a text result, such as title link and snippet, from its understanding of the…

Google said posting lists, which Google's serving system uses to find the pages that contain a query's words, are not new: they are at least 60 years old (as of 2026).

Speaker GoogleEvidence transcript

• What the index looks like

• Extends D2-C732 Day 2: To find relevant pages, Google's serving system relies on posting lists, a long-established information…

For most of the tokens Google finds on the web, though not every single one, the index keeps a posting list of the URLs that contain that token.

Speaker GoogleEvidence transcript

• What the index looks like
• Tokenization: how text is stored

Used byglossary term Posting list

• Repeated by D3-C074 Day 3: Google's index uses posting lists: for each word, a list of the URLs associated with that word.

In posting-list retrieval, the posting lists of the query's words are intersected, which yields an unranked list of candidate URLs.

Speaker GoogleEvidence transcript

• What the index looks like

Used byglossary term Posting list

• Extended by D3-C075 Day 3: At retrieval, Google splits the query into words, applies query understanding and expansion, and matches the…

A search query is broken into words with the same segmenter or tokenizer that Google used to build the index.

Speaker GoogleEvidence transcript

• Tokenization: how text is stored
• What the index looks like

• Repeats D2-C321 Day 2: For languages written without spaces, such as Thai and Chinese, Google uses exactly the same word…
• Extended by D3-C013 Day 3: Google's query processing deliberately mirrors indexing: a query is transformed into something that can be…

At retrieval, Google looks up the posting lists of the query words that are actually important rather than of every word in the query.

Speaker GoogleEvidence transcript

• What the index looks like
• Tokenization: how text is stored

• Extended by D3-C075 Day 3: At retrieval, Google splits the query into words, applies query understanding and expansion, and matches the…

Google said a vector space also holds embeddings for associations the web makes with a page, such as what is known about its author; most of them sit far from typical queries, and a query that names the association may move closer to them.

Speaker GoogleEvidence transcript

• What the index looks like
• AI systems already inside Search

The legacy Google Trends Explore page is still available but will be retired: Google is adding its features to the new Explore page until everyone can move to the new one (no date was given).

“the legacy Explore page, which is still available, but not for a long time”

Speaker Omri WeismanEvidence transcript

• Google Trends tools and features

A Google Trends Explore chart presented on stage put worldwide search interest in Gemini above ChatGPT for the first time ever in September 2025, during a sharp surge for Gemini.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data

Google attributed the September 2025 surge in worldwide search interest for Gemini to the viral launch of Nano Banana (the image editing model in the Gemini app), as people searched for both Gemini and Nano Banana.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data

• Extends D1-C162 Day 1: Google said that after the launch of Nano Banana, its image generation model, the Gemini app became the…

After the Nano Banana surge faded, worldwide baseline search interest in Gemini stayed much higher than before, which Google read as a sustained gain in brand recognition.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data
• Using Google Trends for SEO and content

A Google Trends chart of worldwide search interest in AI from 2004 to mid-2025 showed a huge rise in the two to three years before mid-2025.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data
• How Search is evolving

Google said worldwide search interest in AI has climbed far higher since mid-2025, beyond the range of the 2004 to mid-2025 chart shown in the talk.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data
• How Search is evolving

An early-2012 worldwide spike in Google Trends search interest for 'ai' was caused by the global hit song 'Ai Se Eu Te Pego', whose title contains the Portuguese word 'ai', not by interest in artificial intelligence.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data

A Google Trends engineer found the cause of the 2012 'ai' spike by zooming in on that time range on the Explore page and reading the related queries.

Speaker Omri WeismanEvidence transcript

• Google Trends tools and features
• Using Google Trends for SEO and content

Worldwide Google Trends search interest for ski since 2004 shows strong seasonality, peaking every winter.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data

Worldwide Google Trends search interest for ski also shows a mostly steady decline since 2004, which can be misread as falling search volume.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data

Google said its internal numbers show the actual volume of ski-related searches is going up, even though ski's share of all searches in Google Trends has declined since 2004.

“we can tell you from our own internal numbers that we're seeing on Google Trends: the actual volume is, in fact, going up.”

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data

A Google Trends chart of US search interest over five years showed umbrella and sunscreen correlated, both peaking in June, against the common assumption that umbrella searches peak in winter.

Speaker Omri WeismanEvidence transcript

• Reading Google Trends data
• Using Google Trends for SEO and content

Google said journalists use Google Trends' Trending now every morning to find content gaps: topics people are searching for that their publication may need to cover.

Speaker Omri WeismanEvidence transcript

• Using Google Trends for SEO and content

Google said the search terms returned by Google Trends' Suggest search terms are ranked by search interest.

Speaker Omri WeismanEvidence transcript

• Google Trends tools and features

Google said colour maps of the geographic breakdown of search interest (regions within a country, or countries worldwide) had launched in the new Google Trends Explore page about a month before the October 2026 talk.

Speaker Omri WeismanEvidence transcript

• Google Trends tools and features

• Google Trends Gains Explore Maps & Regional Breakdown Press Search Engine Roundtable (19 August 2026) · checked 2 October 2026

The category filter of the legacy Google Trends Explore page has been reintroduced in the new Explore page.

Speaker Omri WeismanEvidence transcript

• Google Trends tools and features

• Google Trends Explore Page Adds Category Filtering Press Search Engine Roundtable (2 September 2026) · checked 2 October 2026

Google named Thai as a language that makes query understanding more complex because it does not separate words with spaces; the speaker added, hedging with 'apparently', that Thai uses spaces to separate sentences.

Speaker John MuellerEvidence transcript

• Non-Latin scripts and word segmentation
• Query understanding

• Extends D2-C320 Day 2: Text in languages written without spaces, such as Thai and Chinese, would end up in the index as long strings…

After detecting the query language and separating the words, Google removes words it thinks matter little to the query, such as 'a' and 'of', known as stop words.

Speaker John MuellerEvidence transcript

• Query understanding

Used byglossary terms Query understanding, Stop words

Google's query processing deliberately mirrors indexing: a query is transformed into something that can be matched against the index, and stop word removal is part of that transformation.

Speaker John MuellerEvidence transcript

• Query understanding
• Tokenization: how text is stored

• Extends D2-C321 Day 2: For languages written without spaces, such as Thai and Chinese, Google uses exactly the same word…
• Extends D2-C737 Day 2: A search query is broken into words with the same segmenter or tokenizer that Google used to build the index.

For some searches the stop words are important, and Google then tries to recognise the whole phrase, stop words included, as an entity; in indexing, such words are indexed together.

Speaker John MuellerEvidence transcript

• Query understanding

Used byglossary term Stop words

When a query names an entity, Google treats it as a request for that entity rather than as a collection of separate words.

Speaker John MuellerEvidence transcript

• Query understanding

Used byglossary term Query understanding

In ranking, Google can match a query's words or its entity, and, the speaker said with a 'probably', mixes both to some degree.

Speaker John MuellerEvidence transcript

• Query understanding

In Google's example, the query word 'photograph' could be expanded to 'image', 'picture' or 'photo', but one German candidate had to be dropped because the similar German word means 'photographer', so expansions are language-specific.

Speaker John MuellerEvidence transcript

• Query understanding

For technical terms, Google's synonym swapping can return either a technical page or a simplified page for the same query.

Speaker John MuellerEvidence transcript

• Query understanding

Google suggested a test: a search that lists a term's synonyms joined with OR will probably return results very similar to the plain query, because Google adds the synonyms itself (part of the sentence is unclear in the recording).

Speaker John MuellerEvidence transcript

• Query understanding

Internally, a short query can turn into a much longer rewritten query, because Google adds entities, synonyms and other information before looking it up.

Speaker John MuellerEvidence transcript

• Query understanding

In Google's rewrite example, [fried chicken place in Barcelona] keeps 'fried' and 'chicken' as two words, may add an entity for fried chicken, and replaces 'place' with alternatives such as 'area', 'location' or 'restaurant'.

Speaker John MuellerEvidence transcript

• Query understanding

Some alternatives in a rewritten query make no sense, such as 'fried chicken area in Barcelona', which is harmless because few indexed pages match them.

Speaker John MuellerEvidence transcript

• Query understanding

In a rewritten query, a place name such as Barcelona can stay a word or be swapped for an entity, possibly a more specific location.

Speaker John MuellerEvidence transcript

• Query understanding

Some synonyms are contextual and depend on the rest of the query: 'GM' probably means General Motors in [GM car], general manager in [GM restaurants] and genetically modified in [GM barley].

Speaker John MuellerEvidence transcript

• Query understanding

Google's synonyms need not be synonyms linguistically: words people use interchangeably are treated as synonyms, because the aim is to find the right content in the index.

“we don't need to be technically accurate”

Speaker John MuellerEvidence transcript

• Query understanding

Used byglossary term Synonyms and siblings

When Google highlights a word in its results that looks like the wrong synonym, the reason is probably that many people use the two words interchangeably.

Speaker John MuellerEvidence transcript

• Query understanding

Besides synonyms, Google detects 'siblings', words in the same category that are related but not interchangeable, such as Canon and Nikon.

Speaker John MuellerEvidence transcript

• Query understanding

Used byglossary term Synonyms and siblings

Because Canon and Nikon are siblings rather than synonyms, a search for [Canon camera] should not show Nikon cameras.

Speaker John MuellerEvidence transcript

• Query understanding

Used byglossary term Synonyms and siblings

Google learns synonyms and siblings from search behaviour: words people search with in the same way become synonyms, while frequent comparison queries mark words as not interchangeable (the end of the sentence is unclear in the recording).

Speaker John MuellerEvidence transcript

• Query understanding

To check whether Google treats two terms as the same, search for each: [Iberico ham] and [jamón ibérico] both brought up the same entity, so Google understands them as the same thing.

Speaker John MuellerEvidence transcript

• Query understanding

At the time of the talk, a search for [Spanish cured ham] brought up mainly a Wikipedia page, showing that Google does not automatically equate that phrase with Iberico ham.

Speaker John MuellerEvidence transcript

• Query understanding

Google sometimes gets diacritic variants wrong, so it is worth searching to see whether Google understands a variant; if it does, pick one spelling and use it.

Speaker John MuellerEvidence transcript

• Non-Latin scripts and word segmentation
• Query understanding

Users expect content written the way they search: in some languages they search in Latin characters, in others in the local script, and Hindi users, for example, search both in Hindi and in Latin letters.

Speaker John MuellerEvidence transcript

• Non-Latin scripts and word segmentation

Used byrequirement DEV-INT-11

• Extends D2-C599 Day 2: Google usually understands non-English words typed 'in English' (probably meaning romanised, Latin-letter…

Google said a short video showing how one query is expanded into several fan-out queries has been in its documentation for a while.

Speaker John MuellerEvidence transcript

• Query fan-out

Google said fan-out queries are not added to Search Console, because Google considers them part of its infrastructure.

“Fan-out queries are not added in Search Console, because they're basically a part of our infrastructure.”

Speaker John MuellerEvidence transcript

• Query fan-out
• Measuring success in the AI era

Used byrequirement DEV-AIF-03glossary term Query fan-out

• Extends D1-C124 Day 1: Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to…

At retrieval, Google splits the query into words, applies query understanding and expansion, and matches the words and their expansions against the posting lists.

Speaker Gary IllyesEvidence transcript

• Retrieval: finding candidates in the index
• Query understanding

Used byglossary term Retrieval

• Extends D2-C736 Day 2: In posting-list retrieval, the posting lists of the query's words are intersected, which yields an unranked…
• Extends D2-C738 Day 2: At retrieval, Google looks up the posting lists of the query words that are actually important rather than of…

Because a query like [best fried chicken ever] can match millions of pages, Google already orders the candidates during retrieval, before ranking starts.

Speaker Gary IllyesEvidence transcript

• Retrieval: finding candidates in the index

Google called quality the most important of the signals used to order candidates at retrieval: a URL of high quality is more likely to be retrieved from the index for specific queries.

“if the quality of a URL is high, then it's more likely to be retrieved from the index for specific queries.”

Speaker Gary IllyesEvidence transcript

• Retrieval: finding candidates in the index
• Ranking systems and signals

Used byglossary term Retrieval

• Extends D2-C722 Day 2: Each document in Google's index has pretty much all the signals calculated for it attached, for example…

A community speaker recalled that around 2009 writers bent titles and phrases to include a keyword because their CMS counted keyword density, and said that hitting the ratio really worked at the time.

Speaker Community speakersEvidence transcript

• Content for people
• Spam detection and SpamBrain

A community speaker compared Googlebot to an inspector who cannot enter the shop and only looks through the window, seeing the page as ones and zeros, which the speaker linked to tokenization.

Speaker Community speakersEvidence transcript

• Tokenization: how text is stored
• Content for people

Google's quality talk described quality as a ranking signal in its own right: a number calculated from many different parts (the word 'quality' is a repaired speech-to-text reading).

Speaker GoogleEvidence transcript

• Content for people
• Ranking systems and signals

Google's quality talk called PageRank the speaker's favourite ranking system, elegant in its time, but said Google does not really use it so much anymore.

“My favorite is probably PageRank, even though we don't really use them so much anymore.”

Speaker GoogleEvidence transcript

• Ranking systems and signals

• A guide to Google Search ranking systems Google Search Central · checked 3 October 2026

Google's quality talk said search quality raters can be employed by other search engines too, not just by Google.

Speaker GoogleEvidence transcript

• Testing changes and quality raters

Google's quality talk said a big portion of the new pages Google discovers every day is spam, adding that the speaker did not know of Google ever publishing the percentage.

Speaker GoogleEvidence transcript

• Spam detection and SpamBrain

The speaker opened the talk on why Search changes by noting that Google's logo has changed only a handful of times in about 30 years.

Speaker GoogleEvidence transcript

• How Search is evolving