Area

AI and Search

How the AI features relate to classic Search, which AI systems already run inside it, how Gemini and AI agents use web content, and what that means for optimisation.

12 topics · 521 claims

Topics in this area 12

52 claims · 16 sessions
SEO vs GEO
Google's position on both days is that GEO is not a separate discipline: AI Overviews and AI Mode use the same crawling and index as Search, are built on the core ranking systems and add grounding and query fan-out at serving, so good SEO is good GEO. Day 2 repeated and extended this: the AI features are a different experience of the same content, sites need no extra work to appear in them, and they use the same index structures and token-based snippets as classic results, with fan-out queries run against the Search index. Google's structured data talk said the same processed markup feeds classic results and AI features and that raw schema.org markup is generally not put straight into a model's context (said at the event, not in Google's docs), and Google's AI features guide says no special markup is needed for generative AI search. Google added that both retrieval routes, posting lists and vector embeddings, work from page content, so the content mantra Google started about 25 years ago still stands, while Google's internationalisation talk stressed that AI answers are still language-dependent. Day 3 closed on the same line: Google's final slide said AI on Google is just SEO, since AI Overviews and AI Mode use the same processes as traditional results and have very few of their own, so no new acronym is needed, and query understanding flows into the AI features too. Google said spam techniques aimed at manipulating AI responses violate the same spam policies, and its marketing research talk also advised optimising for people to win in generative AI search. A community speaker's agency found some pages doing better in AI features and third-party LLMs than in classic search, so its answer to whether good SEO is good GEO was 'it depends'. Author’s view: that comparison mixed Google's AI features with other LLMs, so it does not refute Google's line; measure each surface separately. A second recording of Day 1 placed the 'GEO is not needed' line at the end of Gary Illyes's talk on what is new in Search; he traced 'SEO is dead' posts back to a 1997 forum post titled 'Search engine R.I.P.' and said the new AI features only create more opportunities for site owners. A community speaker argued the opposite on the label, for adopting GEO as a chance to leave SEO's bad reputation behind, while another said AI search changes the workflow but not the foundation and expands SEO's target to whether information is useful and easy to extract as a source. In the Day 1 Q&A a Google panelist said optimising for AI search can indeed hurt classic search, pointing to questionable GEO advice published online. A second recording of Day 2 confirmed that AI Overviews and AI Mode typically do not ground answers by reading pages live, unlike Gemini asked about a specific page, and that schema.org markup is first sorted, quality-checked and indexed before it is passed on as grounding context rather than put into a model's context as text.
Day 1Day 2Day 3
38 claims · 8 sessions
Query fan-out
Query fan-out expands one question into several related searches run at once, across subtopics and, for AI Mode, data sources such as the Knowledge Graph and shopping data as well as web content. Day 2 filled in the mechanism behind Day 1's grounding on the Search index: Google described the generated queries going to the index and the documents coming back with their token-based snippets, which become the source material for the AI answer (part of that passage is unclear in the recording), so a page that forbids a snippet cannot contribute one. Google's robots meta tag specification confirms that nosnippet keeps a page's content out of AI Overviews and AI Mode as a direct input and that max-snippet limits how much can be used. Google's guide warns that creating separate pages for fan-out query variations primarily to manipulate rankings or AI responses is scaled content abuse, and says its AI systems can judge relevance without an exact query match. Day 3 tied fan-out to classic query handling: Google's AI features send the query and its search results to an LLM, which returns new queries for Google to run, and Google treats those like typed queries, passing each through query understanding, synonym expansion included, while keeping them distinct for better coverage; synonym expansion belongs to traditional Search, and fan-out expands further. Google said fan-out queries are not added to Search Console (not in Google's docs, though its reports fit the statement) but can be seen in Gemini and some search APIs, and advised understanding that fan-out happens without overfocusing on individual fan-out queries, which differ from system to system. A second recording of Day 1 added Google's own framing: the keynote said Gemini lets Search understand the user's intent and fan-out then adds further queries to enrich the answer, as when a query about weeds in a lawn also looks for the best herbicide, a non-chemical solution and prevention, and Gary Illyes said fan-out is nothing new, firing, say, ten normal searches in the background on the same systems. Community speakers added that fan-out queries are often in another language than the prompt: a study (heard as Peec AI's) found 43% of prompts produced fan-out queries in another language, and a Spanish prompt about places to eat in Barcelona fanned out into English queries, so the content used to answer may be written for a different audience; one speaker said Microsoft is ahead of Google in showing site owners the grounding queries its AI uses. Author’s view: in Peec AI's published analysis of over 10 million ChatGPT prompts, 43% is the share of fan-out searches for non-English prompts that ran in English, and nearly 78% of non-English prompt runs had at least one English fan-out, so cite the 43% as a share of fan-outs, not of prompts.
Day 1Day 2Day 3
17 claims · 3 sessions
llms.txt
Google said on a slide that llms.txt is not necessary, and its documentation says such files neither harm nor help because Google Search ignores them. The Day 1 Q&A, captured by a second recording, added detail: Gary Illyes said llms.txt does not matter for Google Search now and he does not expect it to, though he has been wrong before, and that it matters to some people, which is why Lighthouse audits it (Chrome's documentation treats the file as optional and marks the audit not applicable when it is missing). A Google panelist knew of no plan to use it and said a possible future change is no reason to implement it now, since most files come from an SEO plugin checkbox; he objected that the file is designed for allegedly intelligent systems that should be able to parse a website, and compared llms.txt to meta keywords: an agent should not trust what a site says about its own authority. Crawler requests for llms.txt, or for the satirical cats.txt, in server logs do not mean a file matters; Googlebot crawls any file it finds linked, with no other effect. A community speaker cited a third-party study (heard as Ahrefs') of almost 40,000 sites over one month, in which 97% of llms.txt files had no AI agent hits and the rest got about two hits a month, and advised against markdown copies of pages for agents. Author’s view: Ahrefs' June 2026 study of 137,210 domains matches the 97% (of about 38,000 sites with the file), but the files that were fetched got around 20 requests each in May 2026, not two.
Day 1
33 claims · 6 sessions
AI-written vs human-written content
Raising on its own slide the question this topic usually prompts, whether it can tell AI text from human text, Google said its ranking systems are trained on content by humans for humans and promote natural content better. Author’s view: that is not a claim of detection, and the risk is scaled, low-value content. Day 3 answered the question more fully: Google's quality talk said quality problems are quality issues, not AI versus human ones, because much good AI-assisted content exists, and the rater guidelines say the tool is not the problem but how it was used and what for. The risk is scale and deception: Google counts AI slop, mass-produced LLM content, as scaled content abuse, a Google speaker expected scaled content abuse to replace link spam as the spam type worth discussing, and the rater guidelines rate fake author personas with AI-generated headshots as Lowest. Google also said hallucinations cannot be removed with current training methods, so publishing unchecked AI output can harm a site's standing indirectly, and its closing slide said to use AI, responsibly. A second recording of Day 1 made the answer explicit: Gary Illyes said Google is not really trying to tell AI-written from human-written content because it cares more about quality than about how content was made, as the keynote also said, though he added that the more important documents in its index tend to be human-created and that the natural content its algorithms promote includes content created, or at least edited and reviewed, by humans (said at the event). On Day 2 he said AI-generated images on a site are up to the site owner and that Google shows them when users look for them, but that image models are poor at text, so sites should check them for hallucinations; Google's image metadata guide supports IPTC and C2PA labels for AI-made images.
Day 1Day 2Day 3
50 claims · 14 sessions
AI systems already inside Search
AI in Search is not new: Day 1 slides placed statistical models (in use for over 20 years, for spam and 'Did you mean'), BERT in indexing, and RankBrain and MUM in serving. Google's ranking systems guide lists BERT among its ranking systems without mentioning indexing and says MUM is not used for general ranking, and Google's 2021 MUM announcement covered text and images in 75 languages, with video and audio named as a future step, while the slide listed audio and video as well. Day 2 named more machine learning inside the pipeline: a BERT-like model trained on page structure detects soft 404s, machine learning weights the canonical-selection criteria and the weighting changes over time, index selection is a predictive ML system, and SpamBrain is now built on Gemini and fine-tuned for spam (all said at the event, not in Google's docs). Google also said it uses more and more AI to detect spam, with high accuracy in its testing, and Google said it can retrieve documents by the distance between document and query embeddings as well as through posting lists, both consistent with Google's published descriptions. Google's structured data talk set limits: Google cannot afford complex models on every indexed page, and LLM extraction does not yet reach the very high accuracy, around 99.9% as the speaker put it, that Google needs, one reason structured data still matters (said at the event, not in Google's docs). Day 3's closing talk described AI as an umbrella of technologies: Google said it has used machine learning for probably 30 years, starting with the statistical models behind 'Did you mean' (Day 1 said over 20 years), that some ranking features are still plain machine learning models because they suit the task better than LLMs, and that BERT is technically a predictive language model, an LLM in the purest sense. Hallucinations can happen with any model, and grounding or retrieval-augmented generation reduces but cannot eliminate them (said at the event). The quality talk's slide listed MUM next to BERT and RankBrain among the ranking systems, while Google's guide says MUM is not used for general ranking, and Google said AI has made its spam updates faster and broader. A second recording of Day 1 added the history: Google said it declared itself AI-first more than a decade ago, created and published the Transformer architecture, and takes a full-stack approach from its own TPU chips through research and Gemini to its apps, and that in 2025 it delivered a decade of innovation in 12 months, with its new Gemini model at the top of all the benchmarks. Gary Illyes said 'Did you mean' launched around 2001-2002 on a statistical model, which he counts as AI, that Google began talking publicly about AI in Search around 2015-2016 with RankBrain, and that MUM helps interpret the context of query words (a hiking-shoes query that mentions Mount Everest rather than Kilimanjaro); Cherry Prommawin said a majority of Search features now use AI (said at the event). Author’s view: date these systems by Google's own posts (RankBrain 2015, MUM May 2021), not by the talk, which placed MUM three years after RankBrain.
Day 1Day 2Day 3
28 claims · 9 sessions
Gemini and Search
Gemini is not part of Search, but it uses Google's crawlers for data and grounds answers on the Search index; Google's crawler documentation defines that grounding as Search-index content given to the model at prompt time, controlled with Google-Extended. Day 2 contradicted part of Day 1's slide, which said Gemini shares tokenization with Search: Gary Illyes said Gemini's tokenization differs ('or mostly') and showed the AI-model tokenizer splitting 'robots.txt' and 'tl;dr' into pieces that the Search tokenizer keeps whole. Google also said Gemini training renders pages the same way Search does, that a Gemini user asking about a specific page triggers a live read that takes extra time, that Gemini in Chrome relies heavily on page screenshots and that SpamBrain is now built on Gemini (all said at the event; Google's docs say only, in general terms, that browser agents may analyse screenshots). Gary Illyes said Gemini's context window holds millions of tokens, and Erin Sparling urged making JavaScript content crawlable, renderable and fast, or server-side rendered, because AI systems increasingly ground answers on it. Author’s view: Google documents grounding only from the Search index, so whether a live page read follows Google-Extended or ignores robots.txt like a user-triggered fetcher is unclear. Day 3's closing talk explained the models behind it: large language models are deep learning on internet-scale data that map concepts by context in an internal vector space, and adding grounding or retrieval-augmented generation reduces hallucinations but cannot eliminate them (said at the event, not in Google's docs). A second recording of Day 1 added the keynote's framing: Google takes a full-stack approach from its own TPU chips to Gemini and its apps, and Gemini lets Search understand intent before fan-out adds further queries; an audio recording of the keynote added Google's claim that in 2025 it delivered a decade of innovation in 12 months, with its new Gemini model at the top of all the benchmarks. Gary Illyes said Google does not own third-party chatbots such as ChatGPT and has no insight into them, and that crawling for Gemini may care less about quality and more about the amount of content, because for LLMs the number of tokens matters more (said at the event). On Day 2 he put the context window at millions of tokens and then at perhaps 900,000 to a million. Author’s view: Google's long-context docs say Gemini models have context windows of one million or more tokens, so plan with about one million as the documented floor.
Day 1Day 2Day 3
125 claims · 17 sessions
Content for people
Google's repeated message is to optimise for people: it asks for unique perspectives and expertise beyond common knowledge, says content made for Search will not succeed, and calls optimising for people optimising for generative AI Search. Day 2 tied this to indexing: Google said index selection aims to keep only documents useful now or later, that page quality ultimately decides indexing, and that 'Crawled – currently not indexed' is most often a quality issue rather than a technical one (the index-selection details were said at the event, not in Google's docs). Google repeated Day 1's advice to focus on users, saying that creating content users like is a better use of time than combing documentation for signals, and Google said both retrieval methods work from page content, so the content mantra of about 25 years still stands. Google's video talk added that paying for a generative AI subscription does not make AI-generated videos succeed. Day 3 made content quality central to ranking: Google's quality talk defined it, as the rater guidelines do, by effort, originality, talent or skill and accuracy, contrasted commodity content built on common knowledge with unique, experienced takes, and said quality problems are quality issues, not AI versus human ones. Google added that pages need no synonym lists, because Google expands queries itself, and its guidance names writing to a word count or covering many topics in the hope of traffic as signs of search-engine-first content. Google's marketing research talk reached the same advice, optimise for people because a human makes the final decision, and the closing talk warned that chasing fan-out queries distracts from creating helpful content. A community case study, in which a niche article drew 28 times more clicks than a comprehensive one from far fewer impressions (no data source or period given), illustrated the point. A second recording of Day 1 added that the opening keynote closed with 'UEO', user engine optimisation, next to SEO and GEO: focus on the user and the rest will follow; it also said high-quality content may be created by humans or by AI, as long as it is made for users and not for search, and that information quality remains a core focus across Search, YouTube and Discover. Cherry Prommawin said quality is one of the most important things in ranking, and Gary Illyes said the natural content Google's algorithms understand and promote better includes content created, or at least edited and reviewed, by humans (said at the event). A Google host of the AI lightning talks advised always checking what AI tools do and focusing on what the business and its users need, and on Day 2 Gary Illyes said AI-generated images and videos must be checked for hallucinations. In Day 2's consolidation case study a community speaker kept only pages with real business potential, introduced the financial experts responsible for the content instead of an anonymous content factory, and credited the growth to quality rather than more content.
Day 1Day 2Day 3
96 claims · 12 sessions
Measuring success in the AI era
Google's answer to the myth that old metrics no longer work in the AI era was to measure success through metrics that matter to the business. For AI features, Google points to the Generative AI performance report in Search Console (launched 3 June 2026 and rolled out to all sites worldwide by 31 August 2026), which lists impressions, pages, countries, devices (Search only) and dates, with no click or query metrics. Author’s view: the report's country breakdown helps compare AI visibility across markets, and Search Console, not site: queries or Google Trends, is where to judge a migration or Discover demand. Day 3 added detail and debate. Search Console's regular Performance report still counts AI Overviews and AI Mode traffic under the Web search type, while the generative AI report counts AI-surface impressions only, by page, country and device, and Google expects it to get richer; Google said fan-out queries are not added to Search Console (not in Google's docs). Google's published position (August 2025) is that total organic clicks are relatively stable, and its guidance is to judge visits by conversions and value, not clicks alone. Community speakers argued from the practitioner side that rankings and clicks no longer equal business outcomes, that traffic after an AI recommendation may arrive direct or through paid search, and, in one case study, that a niche article converted 10 times better than a comprehensive one with far more impressions (single cases, no published data). A second recording of Day 1 added more voices. Gary Illyes said the generative AI reporting launched with the generative AI control, starting with impressions only, and may expand; in the Q&A a Google panelist expected more AI reporting to launch, without a timeline or a promise, and Google said AI reporting should not simply mirror classic results, because users get more information before they click, so Google first has to work out which data is useful and actionable. Community speakers said the industry turned everything into a metric because budgets are approved on numbers, that most AI-visibility tracking copies rank tracking with invented prompts, and that one misplaced figure in an LLM-assisted data pipeline compounds downstream; one agency replaced monthly reports that took a workday each and went unread with an automated pipeline (data in, script, AI summary, dashboard). On Day 2 migration speakers advised merging both sites' analytics, comparing daily Search Console data and crawls with the pre-migration baseline and giving moved content its own baseline, and an international speaker warned that features often launch in some languages or countries first, so cross-market comparisons should not assume a feature is live everywhere. Author’s view: after a migration, report indexing and traffic recovery separately, since they move on very different timescales.
Day 1Day 2Day 3
10 claims · 2 sessions
Chunking content for AI
Day 1's slide already answered an AI myth by saying there is no need to chop content, and on Day 2 Gary Illyes said the common SEO advice to chunk content for AI is misunderstood: chunking is real but matters at the level of a model's context window. He said Gemini's context window holds millions of tokens, so Gemini does not need content cut into pieces of 100 or 200 words; in the clearer wording of a second recording, once chunk size is thought of in millions of tokens, chunking has perhaps lost its meaning anyway. Author’s view: rewriting pages into short, self-contained chunks for AI brings nothing; structure content for readers. He then put the window at perhaps 900,000 to a million. Author’s view: Google's long-context docs give one million tokens or more, so plan with that as the floor. On Day 1 a community speaker linked the serial-position effect to the 'lost in the middle' pattern found in language models and advised placing the most important information at the beginning or end of content, where, in the speaker's experience on client sites, AI systems cited it more. Author’s view: that is a community heuristic, not something Google has said its systems do; a short summary at the top serves readers either way.
Day 1Day 2
75 claims · 13 sessions
AI crawlers and agents on your site
Google said its effort to render pages as users see them keeps its knowledge of the web current and that rendering JavaScript matters whether the client is a search crawler or an AI system, and a community speaker advised putting everything you want cited into the server-rendered HTML because some AI systems cannot render JavaScript yet. Google warned that AI agents browsing for users hit the same bot walls as scrapers and may buy elsewhere, and closed its duplication talk with 'Don't block agents'; John Mueller said AI crawlers do not really know what to do with nofollow links because they look at content (all said at the event, not in Google's docs). Google's documentation says user-triggered fetchers such as Google-Agent generally ignore robots.txt, and Google said a Gemini user asking about a specific page triggers a live read and that Gemini in Chrome relies heavily on screenshots (both said at the event; Google's docs describe neither specifically), extending Day 1's documented point that browser agents read screenshots, the DOM and the accessibility tree. Erin Sparling added that web standards such as WebMCP let a site expose tools that AI agents can operate (said at the event, not in Google's docs). Author’s view: nofollow is not an access control; at Google, Google-Extended governs Gemini training and grounding and the Search generative AI setting governs AI Overviews and AI Mode. Day 3 raised the bar for shopping data: Google said data quality matters more for AI and even more for agents acting for consumers ('garbage in, garbage out'), since an agent with wrong data might buy the wrong product (said at the event), and Google launched the Universal Commerce Protocol on 11 January 2026 as an open standard for agentic commerce, compatible with A2A, AP2 and MCP. The second recording of Day 1 added Google's view of crawlers in general. Gary Illyes said AI agents are technically the same thing as crawlers, HTTP clients acting for a user or a service; Google runs probably hundreds, if not thousands, of crawlers on one infrastructure (its Inside Googlebot post speaks of dozens of other clients, and both say only the larger ones are documented), Googlebot is the crawler for web search including Search's AI features, and the only Google-owned crawlers that ignore robots.txt are contractual ones that crawl sites by agreement (Google's docs call them special-case crawlers). Google said AI Mode sits on its existing crawling infrastructure, so it brings no extra Googlebot crawling, although sites should expect more crawling overall from other services, AI services included. In the Day 1 Q&A a Google panelist said many AI crawlers are less sophisticated than search crawlers and may simply work through a site in order, because model training mainly needs a very large number of tokens, while search crawling tracks which pages change (Gary Illyes likewise said crawling for Gemini may care more about the amount of content than its quality; said at the event). Mainstream crawlers from Google, other search engines and AI companies try to follow robots.txt, so correct rules are the control and crawlers that ignore them are a scraping problem; user-initiated fetchers, and AI agents acting on a user's request, generally do not check robots.txt, which the panelist argued makes business sense because an agent sent to buy something could not otherwise complete the purchase. Nearly all mainstream AI systems use their own user agents, so robots.txt can set a policy for each, but the panelist doubted per-crawler policies make practical sense yet, called blocking all AI crawlers while allowing search crawlers a personal decision and called a default-deny robots.txt a bad pattern. Gary Illyes added that Google now sees more 403 responses (described on stage as 'authentication required') and, more recently still, more 402 Payment Required responses (not in Google's docs), and treats both like a 404, dropping the pages from Search and its AI features. A community speaker described agent readiness in three layers: visual stability (CLS) for how agents see a page; schema, landmarks, a logical heading order and ARIA for how they understand it; and WebMCP, a proposed standard Chrome lets sites test through an origin trial, for how they act on it; the speaker advised against markdown copies of pages for agents, and another community speaker advised against blocking AI training bots in most cases. Author’s view: a logical heading order is an accessibility and agent concern rather than a Google Search requirement, structured data is not required for Google's generative AI search, and a Google user agent missing from the public lists is not proof of a fake request; verify with reverse DNS or Google's published IP ranges.
Day 1Day 2Day 3
44 claims · 4 sessions
Measuring AI visibility and brand impact
A community speaker argued that rankings and clicks no longer equal business outcomes: visibility can rise while clicks fall, and traffic that follows an AI recommendation may arrive direct, through paid search or through another channel. In one controlled test with no other marketing changes, work on a brand's AI visibility coincided with more inquiries and a 36% rise in paid search click-through rate, figures from one unpublished test of one unnamed brand. The talk leaned on third-party studies (Similarweb on AI-recommended brands, a 2018 eye-tracking study on familiar brands) that Google has not published; Google's own position is that total organic clicks are relatively stable and that sites should judge visits by conversions and value, not clicks alone. This matches Google's Day 1 advice to measure success through metrics that matter to the business, and Google's AI features guide suggests tracking conversions and time on site to value traffic from AI features, because Search Console's generative AI report shows impressions but no clicks. Day 1's first lightning session, captured by a second recording, added community methods. One speaker proposed measuring AI search as a five-stage funnel (know, association, search, retrieval, selection) with metrics such as an AI bot allow rate, a brand association rate asked with web browsing switched off, and a brand co-occurrence rate in fan-out queries, arguing that most tracking covers only selection and copies rank tracking with invented prompts of unknown search volume. Another speaker said a study linking JSON-LD to AI visibility was followed by a controlled study that found no benefit, the markup being a sign of teams doing everything else right, that listicles earn more AI mentions because LLMs check that a thing exists rather than whether the content is honest, and cited a study (heard as Peec AI's) in which 43% of prompts produced fan-out queries in another language. An agency built its own AI visibility tracker to control the method, and a community speaker reported that on client sites AI systems cited information near the end of the content. In the Q&A Google said AI query data is hard to provide because AI questions do not map back to keywords and the data would have to be grouped to protect privacy while staying useful. Author’s view: in Peec AI's published study the 43% is the share of fan-out searches for non-English prompts that ran in English, not a share of prompts. Google's AI optimisation guide says structured data is not required for generative AI search, so use markup for rich results and clear data, not as an AI-visibility lever.
Day 1Day 3
47 claims · 2 sessions
AI in SEO and content workflows
Day 1's first lightning session showed how practitioners use AI in SEO work. An agency SEO director uses AI only where it enriches work the team has already defined: people set the strategy buckets, scoring and rules, and AI pulls in the data and applies them, which cut a competitor keyword analysis from three days to a few hours; his lesson was that AI should not decide what good looks like. A second speaker showed how LLMs get search data wrong, for example calling a move in average position from 8 to 9 an improvement, and argued for receipts that trace every figure to its data request, for never letting a model check its own output and for giving it fixed metric definitions. A community speaker applied Pascal's line about false windows built for symmetry to LLMs, which sometimes are not trying to produce the most accurate figures or interpretations of them. Other talks showed an LLM loop that rewrites structured data until Google's structured-data test passes, a small model (Claude Haiku) choosing keywords under fixed rules (no brand terms, no navigation terms, at most two keywords per page), and monthly Search Console reports automated step by step with AI as a coding partner. Google's hosts closed by calling AI a tool that can be used well or badly and advised always checking what AI tools do. On Day 2 a community speaker described 'prompt journalism': a prompt-assisted workflow that cleans, maps and restructures recorded interviews into an article and social posts, to make more of material that already exists rather than to create more content. Author’s view: a fix loop that stops when Google's test passes proves only that the markup is valid, not that its values are true, and the Search Console API itself is free, so the cost check before automating is about usage limits and the paid parts of the pipeline.
Day 1Day 2

Across days 103

Stage D2-C325 Day 2 · Understanding what's on a page
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'.
contradicts
Slide D1-C039 Day 1 · How Search works and where's AI?
Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization and deduping, and grounds on the Search index.
Analysis D2-C067 Day 2 · Controlling indexing
nofollow is a link-graph hint for search engines, not an access control: to keep AI crawlers away from content, use robots.txt rules for the specific crawlers; at Google, Google-Extended covers Gemini training and grounding, and the Search generative AI control covers AI Overviews and AI Mode.
extends
Docs D1-C127 Day 1 · How Google thinks about crawl budget
Google treats nofollow as a hint for crawling, not a block: nofollow links will generally not be followed, but the linked pages may still be crawled if Google finds them through sitemaps or other links. To stop Google fetching URLs on your own site, use a robots.txt disallow rule.
Analysis D2-C067 Day 2 · Controlling indexing
nofollow is a link-graph hint for search engines, not an access control: to keep AI crawlers away from content, use robots.txt rules for the specific crawlers; at Google, Google-Extended covers Gemini training and grounding, and the Search generative AI control covers AI Overviews and AI Mode.
extends
Stage D1-C482 Day 1 · Q&A
Nearly all mainstream AI crawlers and AI systems use their own user agents, so robots.txt can set a separate policy for each; at the event OpenAI was said to have three or four, Anthropic a few and Microsoft some.
Stage D2-C073 Day 2 · Controlling indexing
John Mueller said Google uses the snippet as a way of building AI Overviews and AI Mode answers, so if a page forbids a snippet, Google cannot use that snippet for them.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Docs D2-C074 Day 2 · Controlling indexing
Google's AI features guide says a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional technical requirements.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Stage D2-C118 Day 2 · Lightning session D: Rendering and JavaScript
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.
extends
Slide D1-C039 Day 1 · How Search works and where's AI?
Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization and deduping, and grounds on the Search index.
Stage D2-C120 Day 2 · Lightning session D: Rendering and JavaScript
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.
extends
Slide D1-C039 Day 1 · How Search works and where's AI?
Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization and deduping, and grounds on the Search index.
Docs D2-C121 Day 2 · Lightning session D: Rendering and JavaScript
Google's crawler documentation defines grounding in Gemini Apps and in Grounding with Google Search on Vertex AI as providing content from the Google Search index to the model at prompt time, and sites manage whether their content is used for it with the Google-Extended robots.txt token.
extends
Docs D1-C086 Day 1 · How Google interprets robots.txt
Google-Extended is a control token, not a crawler with its own user agent string. It decides whether crawled content may be used to train future Gemini models and to ground Gemini apps and Vertex AI.
Docs D2-C122 Day 2 · Lightning session D: Rendering and JavaScript
Google says user-triggered fetchers, which fetch a URL because a user asked for it in a Google product (for example Gemini Notebook fetching URLs users add as sources, or Google-Agent acting on a user's request), generally ignore robots.txt rules.
extends
Stage D1-C434 Day 1 · Q&A
User-initiated fetchers, such as a translation service fetching a page a user asked to translate, are a different category from crawlers and generally do not check robots.txt, because a user asked for the fetch.
Stage D2-C124 Day 2 · Lightning session D: Rendering and JavaScript
Google's simple solution for the extra time of live page reads is to make JavaScript-generated content render quickly and efficiently, or to use server-side rendering.
extends
Slide D1-C056 Day 1 · How Search works and where's AI?
Myth: your website is no longer relevant. Google's answer: keep content crawlable, well structured, fast and easy to read, for readers and for AI tools.
Stage D2-C134 Day 2 · Lightning session D: Rendering and JavaScript
Google urged sites to make sure their JavaScript content can be crawled, rendered and indexed, calling this important today and also tomorrow, as AI systems increasingly ground answers to user requests.
extends
Slide D1-C056 Day 1 · How Search works and where's AI?
Myth: your website is no longer relevant. Google's answer: keep content crawlable, well structured, fast and easy to read, for readers and for AI tools.
Stage D2-C139 Day 2 · Lightning session D: Rendering and JavaScript
A community speaker strongly advised putting everything you want cited into the raw, server-side rendered HTML, especially for AI systems that cannot render JavaScript yet.
extends
Slide D1-C056 Day 1 · How Search works and where's AI?
Myth: your website is no longer relevant. Google's answer: keep content crawlable, well structured, fast and easy to read, for readers and for AI tools.
Stage D2-C303 Day 2 · What is Google friendly JavaScript
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.
extends
Stage D1-C281 Day 1 · Lightning session A: Automation and AI
A community speaker presented WebMCP (Web Model Context Protocol) as a way for a site to declare the actions it offers to AI agents, for example booking an appointment in a calendar, so an agent does not have to scrape the interface; the speaker called it faster, easier and cheaper.
Stage D2-C330 Day 2 · Understanding what's on a page
Gary Illyes said the common SEO advice to chunk content for AI systems is misunderstood: chunking is real, but it matters at the level of an AI model's context window.
extends
Slide D1-C054 Day 1 · How Search works and where's AI?
Myth: optimise for AI over readers. Google's answer: optimise for people, with no need to obsess over precise keywords or AI phrasing, no need to chop content, and no need for llms.txt.
Stage D2-C332 Day 2 · Understanding what's on a page
Gemini does not need content cut into small chunks of 100 or 200 words, Gary Illyes said, since a smaller book fits in its context window.
extends
Slide D1-C054 Day 1 · How Search works and where's AI?
Myth: optimise for AI over readers. Google's answer: optimise for people, with no need to obsess over precise keywords or AI phrasing, no need to chop content, and no need for llms.txt.
Stage D2-C338 Day 2 · Understanding what's on a page
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.
extends
Slide D1-C042 Day 1 · How Search works and where's AI?
BERT is used in indexing to understand each word in the context of the whole sentence rather than one word at a time.
Stage D2-C344 Day 2 · Handling web duplication
Google deduplicates pages because many sites have very many pages and Google's index does not have room for everything.
extends
Slide D1-C037 Day 1 · How Search works and where's AI?
For classic Search, crawling means Googlebot, scheduling and robots.txt, with AI used in parts such as scheduling. Indexing is one big but not limitless index that calculates signals and understands more than words, with AI such as BERT. Serving uses hundreds of signals tailored to the moment, with AI such as RankBrain.
Stage D2-C377 Day 2 · Handling web duplication
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.
extends
Stage D1-C436 Day 1 · Q&A
A Google panelist argued that it makes business sense for agents not to follow robots.txt: an agent sent to buy something that obeyed a robots.txt block could not complete the purchase, a bad experience for the user and lost revenue for the shop.
Slide D2-C397 Day 2 · Handling web duplication
Google's duplication talk closed with the advice not to block agents, which the speaker said are sometimes really cool.
extends
Docs D1-C131 Day 1 · session not recorded
Google's guide for generative AI features says browser agents may read a site through screenshots, the DOM structure and the accessibility tree, and recommends semantic HTML because it helps users such as screen reader users navigate a page.
Docs D2-C462 Day 2 · What is Structured Data and why we need it on the internet.
Google's guidance on generative AI content warns that AI output can contain hallucinations and says AI-generated metadata, including structured data and image alt text, should be fact-checked before publishing and the markup validated.
extends
Stage D1-C250 Day 1 · Lightning session A: Automation and AI
In a community demo, a fix loop gave a large LLM (Claude Opus) the current markup, the errors Google's test reported and Google's documentation, had it write new JSON-LD and re-ran the test, with at most three attempts; the demo's page passed on the second.
Stage D2-C465 Day 2 · What is Structured Data and why we need it on the internet.
Gemini in Chrome relies heavily on the screenshot it takes of a page.
extends
Docs D1-C131 Day 1 · session not recorded
Google's guide for generative AI features says browser agents may read a site through screenshots, the DOM structure and the accessibility tree, and recommends semantic HTML because it helps users such as screen reader users navigate a page.
Stage D2-C476 Day 2 · What is Structured Data and why we need it on the internet.
The structured data Google processes is not fed very differently to AI Overviews and AI Mode: after cleaning and quality work, the same data goes to both the classic results page and the AI features.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Stage D2-C477 Day 2 · What is Structured Data and why we need it on the internet.
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.
extends
Docs D1-C061 Day 1 · How Search works and where's AI?
Google's guide says new machine-readable files, AI text files or special markup are not needed to appear in Search, and that such files neither harm nor help visibility because Google Search ignores them.
Docs D2-C480 Day 2 · What is Structured Data and why we need it on the internet.
Google's guide to optimizing for generative AI features lists 'overfocusing on structured data' among the things site owners don't need to do: structured data is not required for generative AI search and no special schema.org markup is needed, though it remains worth using because it helps pages become eligible for rich results.
extends
Docs D1-C061 Day 1 · How Search works and where's AI?
Google's guide says new machine-readable files, AI text files or special markup are not needed to appear in Search, and that such files neither harm nor help visibility because Google Search ignores them.
Stage D2-C603 Day 2 · Focusing on Internationalisation and Localisation
Users often assume AI is all-knowing and borderless, but AI is still language-dependent: if an AI answer is synthesized from the top results, a query in another language draws on a totally different set of data.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Analysis D2-C607 Day 2 · Focusing on Internationalisation and Localisation
Check AI Overviews and AI Mode with native-language queries in each target market, not with translated English keywords, and compare with the country breakdown of Search Console's generative AI performance report; topics where competitors are cited and you are not point to missing or weak local content.
extends
Docs D1-C124 Day 1 · How Search works and where's AI?
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Analysis D2-C612 Day 2 · Focusing on Internationalisation and Localisation
Treat machine translation as a first draft: have a native speaker review it and adapt dates, calendars and units before publishing, because unreviewed bulk translation that adds little value can also fall under Google's scaled content abuse policy.
extends
Analysis D1-C048 Day 1 · How Search works and where's AI?
The answer is not a claim that Google detects AI text. It says ranking favours text that reads as natural to people. The risk with AI content is scale without value, which falls under Google's scaled content abuse policy, not the tool itself.
Stage D2-C648 Day 2 · Calculating (some) signals
Among the many signals Google calculates during indexing, the ones singled out as having large effects on search results were country, language, freshness, SafeSearch and spam.
extends
Slide D1-C037 Day 1 · How Search works and where's AI?
For classic Search, crawling means Googlebot, scheduling and robots.txt, with AI used in parts such as scheduling. Indexing is one big but not limitless index that calculates signals and understands more than words, with AI such as BERT. Serving uses hundreds of signals tailored to the moment, with AI such as RankBrain.
Stage D2-C668 Day 2 · Calculating (some) signals
Google uses more and more AI to detect spam, and Google's testing shows that this AI-based detection is highly accurate.
extends
Slide D1-C041 Day 1 · How Search works and where's AI?
Statistical models have been used at Google for over 20 years, for catching spam and originally for the 'Did you mean' feature.
Stage D2-C669 Day 2 · Calculating (some) signals
SpamBrain, Google's AI-based spam detection system, is nowadays built on Gemini and fine-tuned specifically for finding spam.
extends
Slide D1-C039 Day 1 · How Search works and where's AI?
Gemini is not part of Search, but it uses crawlers for data, shares some technologies such as tokenization and deduping, and grounds on the Search index.
Stage D2-C669 Day 2 · Calculating (some) signals
SpamBrain, Google's AI-based spam detection system, is nowadays built on Gemini and fine-tuned specifically for finding spam.
extends
Slide D1-C041 Day 1 · How Search works and where's AI?
Statistical models have been used at Google for over 20 years, for catching spam and originally for the 'Did you mean' feature.
Stage D2-C684 Day 2 · Deciding what goes in the index?
Index selection is a predictive AI system that relies heavily on machine learning.
extends
Slide D1-C037 Day 1 · How Search works and where's AI?
For classic Search, crawling means Googlebot, scheduling and robots.txt, with AI used in parts such as scheduling. Indexing is one big but not limitless index that calculates signals and understands more than words, with AI such as BERT. Serving uses hundreds of signals tailored to the moment, with AI such as RankBrain.
Stage D2-C695 Day 2 · Deciding what goes in the index?
Page quality is ultimately what decides whether a document is indexed, so focusing on quality is the most reliable way to get pages into Google's index.
extends
Slide D1-C093 Day 1 · How Google thinks about crawl budget
Crawl demand is driven by the quality of the site, the change frequency of its URLs and their popularity on the internet.
Stage D2-C714 Day 2 · Deciding what goes in the index?
'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.
extends
Analysis D1-C098 Day 1 · How Google thinks about crawl budget
On smaller sites, slow indexing is almost always a demand problem, meaning quality, not a capacity problem.
Analysis D2-C716 Day 2 · Deciding what goes in the index?
Treat 'Crawled – currently not indexed' as a quality audit list: compare those URLs with indexed pages of the same type for thin, duplicate or soft-404-like content and for template differences before looking for technical faults.
extends
Analysis D1-C098 Day 1 · How Google thinks about crawl budget
On smaller sites, slow indexing is almost always a demand problem, meaning quality, not a capacity problem.
Stage D2-C722 Day 2 · How does the index look like?
Each document in Google's index has pretty much all the signals calculated for it attached, for example quality signals plus the page's country and language, according to an illustration the speaker called an approximation of the real structure.
extends
Slide D1-C037 Day 1 · How Search works and where's AI?
For classic Search, crawling means Googlebot, scheduling and robots.txt, with AI used in parts such as scheduling. Indexing is one big but not limitless index that calculates signals and understands more than words, with AI such as BERT. Serving uses hundreds of signals tailored to the moment, with AI such as RankBrain.
Stage D2-C726 Day 2 · How does the index look like?
AI Overviews and AI Mode use the same index structures and token-based snippets as classic web results, a point Google called important but not obvious.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Stage D2-C727 Day 2 · How does the index look like?
Google said that when AI Overviews or AI Mode run a query fan-out, the generated queries are sent to Google's Search index and documents come back with their snippets, which then feed the AI-generated answer (part of this passage is unclear in the recording).
extends
Slide D1-C051 Day 1 · How Search works and where's AI?
Three reasons were given: generative AI features are built directly on the core ranking systems, query fan-out expands the original query to find related information, and generative AI features highlight content indexed by Google Search.
Stage D2-C727 Day 2 · How does the index look like?
Google said that when AI Overviews or AI Mode run a query fan-out, the generated queries are sent to Google's Search index and documents come back with their snippets, which then feed the AI-generated answer (part of this passage is unclear in the recording).
extends
Docs D1-C053 Day 1 · How Search works and where's AI?
Query fan-out means running several related searches at once to gather more results; a question about lawn weeds may also search herbicides and weed prevention.
Stage D2-C727 Day 2 · How does the index look like?
Google said that when AI Overviews or AI Mode run a query fan-out, the generated queries are sent to Google's Search index and documents come back with their snippets, which then feed the AI-generated answer (part of this passage is unclear in the recording).
extends
Stage D1-C172 Day 1 · Welcome and opening keynotes
Google said the Gemini model lets Search understand the user's intent, and query fan-out then adds further queries to the first one to enrich the quality of the answer.
Docs D2-C729 Day 2 · How does the index look like?
Google says query fan-out in AI Overviews and AI Mode issues related searches across subtopics and several data sources, which for AI Mode include the Knowledge Graph and shopping data as well as web content (AI Mode launch post, March 2025).
extends
Docs D1-C053 Day 1 · How Search works and where's AI?
Query fan-out means running several related searches at once to gather more results; a question about lawn weeds may also search herbicides and weed prevention.
Stage D2-C741 Day 2 · How does the index look like?
In embedding-based retrieval, the distance between the embeddings of documents and the embedding of the user's query decides which documents are returned.
extends
Docs D1-C129 Day 1 · How Search works and where's AI?
Google's guide says creating separate content for every variation of how people might search, including fan-out queries, primarily to manipulate rankings or AI responses violates its scaled content abuse policy. It adds that its AI systems can understand a page's relevance even without an exact match to the query.
Stage D2-C770 Day 2 · Google Trends
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.
extends
Stage D1-C162 Day 1 · Welcome and opening keynotes
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.
Slide D3-C057 Day 3 · Making sense of users' queries
Google's slide on how LLM features with grounding generally work showed a query going to both the search engine and an LLM, the search engine's results going to the LLM, the LLM generating fan-out queries that go back to the search engine, and the LLM returning answers with links.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Stage D3-C059 Day 3 · Making sense of users' queries
Google treats fan-out queries generated by the LLM the same way as queries typed by users, so understanding how normal queries work explains fan-out queries too.
extends
Stage D1-C172 Day 1 · Welcome and opening keynotes
Google said the Gemini model lets Search understand the user's intent, and query fan-out then adds further queries to the first one to enrich the quality of the answer.
Stage D3-C059 Day 3 · Making sense of users' queries
Google treats fan-out queries generated by the LLM the same way as queries typed by users, so understanding how normal queries work explains fan-out queries too.
extends
Stage D2-C727 Day 2 · How does the index look like?
Google said that when AI Overviews or AI Mode run a query fan-out, the generated queries are sent to Google's Search index and documents come back with their snippets, which then feed the AI-generated answer (part of this passage is unclear in the recording).
Stage D3-C061 Day 3 · Making sense of users' queries
Google tries to make fan-out queries distinct from each other for better coverage, avoiding asking the same question several times, which would return the same answers.
extends
Stage D1-C225 Day 1 · How Search works and where's AI?
Google said query fan-out is nothing new: it fires, for example, ten different searches in the background, each a normal search on the same systems.
Stage D3-C061 Day 3 · Making sense of users' queries
Google tries to make fan-out queries distinct from each other for better coverage, avoiding asking the same question several times, which would return the same answers.
extends
Docs D2-C729 Day 2 · How does the index look like?
Google says query fan-out in AI Overviews and AI Mode issues related searches across subtopics and several data sources, which for AI Mode include the Knowledge Graph and shopping data as well as web content (AI Mode launch post, March 2025).
Stage D3-C063 Day 3 · Making sense of users' queries
Google said fan-out queries are not added to Search Console, because Google considers them part of its infrastructure.
extends
Docs D1-C124 Day 1 · How Search works and where's AI?
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Stage D3-C065 Day 3 · Making sense of users' queries
Because every system runs fan-out differently, Google advised understanding that fan-out happens but not overfocusing on individual fan-out queries or on how to rank for them.
extends
Docs D1-C129 Day 1 · How Search works and where's AI?
Google's guide says creating separate content for every variation of how people might search, including fan-out queries, primarily to manipulate rankings or AI responses violates its scaled content abuse policy. It adds that its AI systems can understand a page's relevance even without an exact match to the query.
Stage D3-C077 Day 3 · Making sense of users' queries
The first condition for retrieving a document is that the query's words, or its concepts in the case of vectors or embeddings, are in the document or related to it.
extends
Stage D2-C740 Day 2 · How does the index look like?
Besides posting lists, Google can retrieve documents through vector embeddings: parts of documents are associated with embeddings, which form a vector space used for retrieval.
Stage D3-C104 Day 3 · Lightning session K: Facets of quality
To choose what to cover in depth, a community speaker's team used Google Trends to find what people in their market were interested in and then gave them that content.
extends
Stage D2-C789 Day 2 · Google Trends
Google presented three uses of Google Trends for content creation, marketing and SEO: keyword selection, scheduling and ideation.
Analysis D3-C135 Day 3 · How Google thinks about Quality
The quality talk's slide listed MUM among Google's ranking systems, while Google's ranking systems guide says MUM is not currently used for general ranking in Search; read the slide as a list of systems Google runs, not as proof that each one ranks every query.
extends
Docs D1-C130 Day 1 · How Search works and where's AI?
Google's ranking systems guide says MUM is not currently used for general ranking in Search, only for specific applications such as COVID-19 vaccine searches and featured snippet callouts.
Stage D3-C170 Day 3 · How Google thinks about Quality
Google's quality talk pointed to page 21 of the Search Quality Rater Guidelines for its definition of content quality by effort, originality, talent or skill and accuracy, noting that the document is updated from time to time.
extends
Stage D2-C311 Day 2 · Understanding what's on a page
Gary Illyes pointed to Google's Search Quality Rater Guidelines as the detailed source on how Google thinks about the main content of a page.
Stage D3-C179 Day 3 · How Google thinks about Quality
Google's quality talk said non-commodity content offers a unique, experienced take, and advised writing content that readers will find very helpful and reliable.
extends
Slide D1-C055 Day 1 · How Search works and where's AI?
Myth: build content for every possible consumer need. Google's answer: prioritise unique perspectives, expertise and in-depth experience that go beyond common knowledge.
Stage D3-C205 Day 3 · How Google thinks about Quality
Google's quality talk said AI has fundamentally changed how Google builds spam updates, letting it evaluate many more candidates and drastically increasing its velocity, so it launches faster with more impact.
extends
Stage D2-C668 Day 2 · Calculating (some) signals
Google uses more and more AI to detect spam, and Google's testing shows that this AI-based detection is highly accurate.
Stage D3-C206 Day 3 · How Google thinks about Quality
Google's quality talk said AI lets Google catch new types of spam and catch more of it.
extends
Stage D2-C668 Day 2 · Calculating (some) signals
Google uses more and more AI to detect spam, and Google's testing shows that this AI-based detection is highly accurate.
Stage D3-C212 Day 3 · How Google thinks about Quality
Google's quality talk said Google clarified that traditional spam techniques aimed at manipulating AI responses also violate its spam policies, and that Google can take action against them.
extends
Docs D1-C129 Day 1 · How Search works and where's AI?
Google's guide says creating separate content for every variation of how people might search, including fan-out queries, primarily to manipulate rankings or AI responses violates its scaled content abuse policy. It adds that its AI systems can understand a page's relevance even without an exact match to the query.
Stage D3-C311 Day 3 · How Search results are born
Google handles an image used as a search query much like a text query interpreted as an embedding: the image is broken down into vectors (embeddings) that are then searched for in the index.
extends
Stage D2-C740 Day 2 · How does the index look like?
Besides posting lists, Google can retrieve documents through vector embeddings: parts of documents are associated with embeddings, which form a vector space used for retrieval.
Stage D3-C325 Day 3 · How Search results are born
AI Mode and AI Overviews are not rich results but standard search features: they need no structured data to function and work with the normal text results from Google's index.
extends
Stage D2-C726 Day 2 · How does the index look like?
AI Overviews and AI Mode use the same index structures and token-based snippets as classic web results, a point Google called important but not obvious.
Stage D3-C440 Day 3 · Inside Search Console: What’s New & How to Use It
Google called AI reporting in Search Console an evolving space and expects the generative AI report to get richer, with more information in the future.
extends
Stage D1-C449 Day 1 · Q&A
A Google panelist said he expected more AI reporting to launch in Search Console, without a timeline or a promise, because many people at Google care about giving site owners the information to understand and adapt to what happens in Search.
Stage D3-C442 Day 3 · Inside Search Console: What’s New & How to Use It
Search Console launched platform properties, which bring social platforms into Search Console, about three or four months before the October 2026 event.
extends
Stage D2-C960 Day 2 · Lightning session F: Media
A community speaker said publishers can also verify their social accounts in Search Console, which makes it easier to track the performance of content repurposed for social platforms.
Docs D3-C468 Day 3 · Inside Search Console: What’s New & How to Use It
Google's social and video performance guide says that if you already claimed your Search profile, all of its verified accounts are added automatically as platform properties in Search Console.
extends
Stage D2-C960 Day 2 · Lightning session F: Media
A community speaker said publishers can also verify their social accounts in Search Console, which makes it easier to track the performance of content repurposed for social platforms.
Stage D3-C469 Day 3 · Lightning session L: Understanding SERPs and your users
In a community lightning talk, agency founder Nik Vujic presented how his agency feeds Search Console and GA4 data to an LLM agent to support decisions for its clients.
extends
Stage D1-C255 Day 1 · Lightning session A: Automation and AI
A community speaker who is not a developer automated monthly SEO reports with Python in Visual Studio Code, using Claude and ChatGPT as coding partners and Google Cloud for access to the Search Console API.
Slide D3-C477 Day 3 · Lightning session L: Understanding SERPs and your users
The community workflow's third step feeds the exported Search Console dataset into the agency's LLM agent, which queries the dataset and returns the evidence rather than a sample.
extends
Stage D1-C256 Day 1 · Lightning session A: Automation and AI
An agency's automated monthly SEO report follows four simple steps: the data comes in, a script processes it, AI summarises it and the result goes onto a dashboard.
Slide D3-C478 Day 3 · Lightning session L: Understanding SERPs and your users
Nik Vujic's slide said no agent is needed to start: pull the Search Console data, import it into any LLM and query it in conversation; his agency built its agent to have everything in one place.
extends
Stage D1-C256 Day 1 · Lightning session A: Automation and AI
An agency's automated monthly SEO report follows four simple steps: the data comes in, a script processes it, AI summarises it and the result goes onto a dashboard.
Analysis D3-C487 Day 3 · Lightning session L: Understanding SERPs and your users
Google's ranking systems guide describes 'query deserves freshness' systems that show fresher content where it would be expected, which is narrower than a general preference for fresh content; refresh pages whose queries expect current information, and judge other refreshes by quality.
extends
Stage D2-C656 Day 2 · Calculating (some) signals
Freshness is a signal for queries that deserve fresh results ('query deserves freshness'): when a breaking event hits a city, such as possible closure of Barcelona's airport, users want really fresh results, not results from two weeks ago.
Stage D3-C495 Day 3 · Lightning session L: Understanding SERPs and your users
Nik Vujic said he hopes Search Console's generative AI report will get query data.
extends
Docs D1-C124 Day 1 · How Search works and where's AI?
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Analysis D3-C497 Day 3 · Lightning session L: Understanding SERPs and your users
Pages that do better in AI answers than in classic results do not refute Google's line that optimising for people is optimising for generative AI Search, since the comparison mixed Google's AI features with third-party LLMs; measure each surface separately, per page, before deciding what to change.
extends
Slide D1-C058 Day 1 · How Search works and where's AI?
Optimising for people is optimising for generative AI Search.
Stage D3-C518 Day 3 · Lightning session L: Understanding SERPs and your users
A community speaker concluded that rankings and clicks no longer equal real business outcomes.
extends
Slide D1-C057 Day 1 · How Search works and where's AI?
Myth: the old metrics don't work in the AI era. Google's answer: measure success through metrics that matter to your business.
Docs D3-C535 Day 3 · Lightning session L: Understanding SERPs and your users
Beyond Search Console, Google's AI features guide suggests tracking conversions and time spent on the site in tools such as Google Analytics to understand the value of traffic from AI features.
extends
Docs D1-C062 Day 1 · How Search works and where's AI?
Google's guide for generative AI features recommends the Generative AI performance report in Search Console for measuring how content performs in generative AI features on Google Search and Discover.
Docs D3-C536 Day 3 · Lightning session L: Understanding SERPs and your users
A May 2025 Search Central blog post advised site owners to look at the overall value of visits from Search rather than focusing too much on clicks, using indicators of conversion such as sales, sign-ups, a more engaged audience or information lookups about the business.
extends
Slide D1-C057 Day 1 · How Search works and where's AI?
Myth: the old metrics don't work in the AI era. Google's answer: measure success through metrics that matter to your business.
Analysis D3-C542 Day 3 · Lightning session L: Understanding SERPs and your users
Google's own line on Day 1 was to measure success by metrics that matter to the business, and its Generative AI performance report shows impressions but no clicks; together with the community talk this argues for a report that puts organic next to direct, branded paid search, conversions and revenue.
extends
Slide D1-C057 Day 1 · How Search works and where's AI?
Myth: the old metrics don't work in the AI era. Google's answer: measure success through metrics that matter to your business.
Analysis D3-C542 Day 3 · Lightning session L: Understanding SERPs and your users
Google's own line on Day 1 was to measure success by metrics that matter to the business, and its Generative AI performance report shows impressions but no clicks; together with the community talk this argues for a report that puts organic next to direct, branded paid search, conversions and revenue.
extends
Docs D1-C124 Day 1 · How Search works and where's AI?
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Stage D3-C680 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Gary Illyes said machine learning is about 50 years old and that Google has been using it for 'probably 30 years', starting with statistical models that made the 'Did you mean' feature possible.
extends
Slide D1-C041 Day 1 · How Search works and where's AI?
Statistical models have been used at Google for over 20 years, for catching spam and originally for the 'Did you mean' feature.
Stage D3-C684 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Predictive language models have been used a lot in Search, and BERT is one: technically, in the purest sense, an LLM.
extends
Slide D1-C042 Day 1 · How Search works and where's AI?
BERT is used in indexing to understand each word in the context of the whole sentence rather than one word at a time.
Stage D3-C686 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Hallucinations can happen with any AI model, and with current training methods there is no way to get rid of them.
extends
Docs D2-C462 Day 2 · What is Structured Data and why we need it on the internet.
Google's guidance on generative AI content warns that AI output can contain hallucinations and says AI-generated metadata, including structured data and image alt text, should be fact-checked before publishing and the markup validated.
Stage D3-C688 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Generative models, including image diffusion models, make things up when they lack information or because of issues in their training.
extends
Stage D2-C939 Day 2 · Using images to your advantage and Engaging Search users with videos
The diffusion models that generate images were built to generate images, not text, so they are typically poor at rendering text inside an image.
Stage D3-C695 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
The cheaper tokens become, the more AI slop is created, and Google counts AI slop as scaled content abuse.
extends
Docs D2-C611 Day 2 · Focusing on Internationalisation and Localisation
Google's spam policies define scaled content abuse as generating many pages mainly to manipulate rankings, with little or no value to users, no matter how they are created, and list automated translating of scraped content among the examples.
Stage D3-C697 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Google's Search Quality Rater Guidelines point out that the tool used to create content is not the problem, but how it was used and what for.
extends
Docs D2-C611 Day 2 · Focusing on Internationalisation and Localisation
Google's spam policies define scaled content abuse as generating many pages mainly to manipulate rankings, with little or no value to users, no matter how they are created, and list automated translating of scraped content among the examples.
Slide D3-C700 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Google's closing slide said to use AI responsibly because AI hallucinates, and, especially when creating content briefs with AI, to make sure not to add to the sea of AI slop already flooding the internet.
extends
Stage D2-C941 Day 2 · Using images to your advantage and Engaging Search users with videos
Sites that use AI-generated images or videos should make sure they work for users, check them for hallucinations and regenerate them where needed.
Slide D3-C701 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Google's closing slide said AI on Google is just SEO: AI features on Google Search use exactly the same processes as traditional results, so no new acronym is needed, as none was for mobile-first indexing or structured data.
extends
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Stage D2-C116 Day 2 · Lightning session D: Rendering and JavaScript
AI Overviews and AI Mode are built on top of Search results: they are a different experience of the same content Google already has.
repeats
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Docs D2-C122 Day 2 · Lightning session D: Rendering and JavaScript
Google says user-triggered fetchers, which fetch a URL because a user asked for it in a Google product (for example Gemini Notebook fetching URLs users add as sources, or Google-Agent acting on a user's request), generally ignore robots.txt rules.
repeats
Stage D1-C273 Day 1 · Lightning session A: Automation and AI
A community speaker pointed out that, according to Google's documentation, user-triggered fetchers, which fetch a page because a user asked for it, generally ignore robots.txt rules.
Stage D2-C601 Day 2 · Focusing on Internationalisation and Localisation
Sites need no extra work to appear in AI Overviews and AI Mode: both work on top of the existing web results, so what works for web results also works for the AI features.
repeats
Slide D1-C050 Day 1 · How Search works and where's AI?
Google's answer to 'SEO is dead, long live GEO' is not to worry about the name: good SEO is good GEO and AEO.
Stage D2-C601 Day 2 · Focusing on Internationalisation and Localisation
Sites need no extra work to appear in AI Overviews and AI Mode: both work on top of the existing web results, so what works for web results also works for the AI features.
repeats
Slide D1-C051 Day 1 · How Search works and where's AI?
Three reasons were given: generative AI features are built directly on the core ranking systems, query fan-out expands the original query to find related information, and generative AI features highlight content indexed by Google Search.
Stage D2-C679 Day 2 · Calculating (some) signals
Combing Google's documentation for signals is not the best use of an SEO's time; creating content that users will like is a better one.
repeats
Stage D1-C059 Day 1 · Welcome and opening keynotes
The opening keynote closed with the advice to think about UEO, user engine optimisation, next to SEO and GEO: focus on the user and the rest will follow.
Stage D2-C680 Day 2 · Deciding what goes in the index?
Google's index is immense but finite, so Google cannot index every URL it finds on a web with a practically infinite number of URLs.
repeats
Slide D1-C037 Day 1 · How Search works and where's AI?
For classic Search, crawling means Googlebot, scheduling and robots.txt, with AI used in parts such as scheduling. Indexing is one big but not limitless index that calculates signals and understands more than words, with AI such as BERT. Serving uses hundreds of signals tailored to the moment, with AI such as RankBrain.
Stage D3-C056 Day 3 · Making sense of users' queries
Google's generative AI features in Search build on the traditional ways of searching, so query understanding also flows into AI Overviews and AI Mode.
repeats
Slide D1-C051 Day 1 · How Search works and where's AI?
Three reasons were given: generative AI features are built directly on the core ranking systems, query fan-out expands the original query to find related information, and generative AI features highlight content indexed by Google Search.
Stage D3-C058 Day 3 · Making sense of users' queries
In Google's AI features, the query and the search results are additionally sent to an LLM, which returns new queries for Google to run.
repeats
Docs D1-C053 Day 1 · How Search works and where's AI?
Query fan-out means running several related searches at once to gather more results; a question about lawn weeds may also search herbicides and weed prevention.
Stage D3-C116 Day 3 · Lightning session K: Facets of quality
Google's long-standing advice to write for people and give them what they want has become true in practice because Googlebot has become more and more human, a community speaker argued.
repeats
Slide D1-C047 Day 1 · How Search works and where's AI?
Google's answer was that ML-based ranking systems are trained on content written by humans for humans, so they understand and promote natural content better.
Stage D3-C129 Day 3 · Lightning session K: Facets of quality
If you cannot say what problems your pages solve for users, or the answer is 'it depends', you are optimising for the search engine rather than for people, and that will not work for long, a community speaker warned.
repeats
Slide D1-C013 Day 1 · Welcome and opening keynotes
Ecosystem principle 4, incentivise high-quality content: content made for Search will not be successful.
Stage D3-C190 Day 3 · How Google thinks about Quality
Google's quality talk said quality problems should be treated as quality issues, not as AI versus human content, because a lot of good AI-assisted or AI-written content exists.
repeats
Stage D1-C210 Day 1 · How Search works and where's AI?
Google said it is not really trying to tell AI-written from human-written content, because it cares more about the quality of content than about how it was created.
Analysis D3-C435 Day 3 · Inside Search Console: What’s New & How to Use It
Review the filters and regexes that Search Console's AI-powered configuration proposes before trusting the numbers: Google's launch post calls the feature experimental, warns that AI can misinterpret requests, and limits it to the Search results Performance report (not Discover or News) and to configuration, not sorting or exporting.
repeats
Stage D1-C243 Day 1 · Lightning session A: Automation and AI
The first rule for using LLMs on search data, according to a community speaker, is never to let the model check its own output.
Stage D3-C437 Day 3 · Inside Search Console: What’s New & How to Use It
Search Console added reporting of the impressions a site's content gets in the AI surfaces of Google Search, found in the left navigation nested under Search results.
repeats
Docs D1-C124 Day 1 · How Search works and where's AI?
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Stage D3-C439 Day 3 · Inside Search Console: What’s New & How to Use It
Search Console's generative AI report shows impressions broken down by pages, countries and devices.
repeats
Docs D1-C124 Day 1 · How Search works and where's AI?
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Stage D3-C440 Day 3 · Inside Search Console: What’s New & How to Use It
Google called AI reporting in Search Console an evolving space and expects the generative AI report to get richer, with more information in the future.
repeats
Stage D1-C196 Day 1 · What's new in the world of Search
Google said Search Console's generative AI reporting launched alongside the generative AI control, starting with impressions only, and may expand later.
Slide D3-C586 Day 3 · Mastering the messy middle
Google's marketing research talk reached the same advice as Search: optimise for people to win in generative AI search.
repeats
Slide D1-C058 Day 1 · How Search works and where's AI?
Optimising for people is optimising for generative AI Search.
Slide D3-C588 Day 3 · Mastering the messy middle
Google's slide advised producing unique, helpful, human-centric content.
repeats
Slide D1-C055 Day 1 · How Search works and where's AI?
Myth: build content for every possible consumer need. Google's answer: prioritise unique perspectives, expertise and in-depth experience that go beyond common knowledge.
Stage D3-C680 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Gary Illyes said machine learning is about 50 years old and that Google has been using it for 'probably 30 years', starting with statistical models that made the 'Did you mean' feature possible.
repeats
Stage D1-C209 Day 1 · How Search works and where's AI?
Google launched the 'Did you mean' feature around 2001-2002 using a statistical model, which Gary Illyes counted as AI because it is a form of machine learning.
Slide D3-C701 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
Google's closing slide said AI on Google is just SEO: AI features on Google Search use exactly the same processes as traditional results, so no new acronym is needed, as none was for mobile-first indexing or structured data.
repeats
Slide D1-C050 Day 1 · How Search works and where's AI?
Google's answer to 'SEO is dead, long live GEO' is not to worry about the name: good SEO is good GEO and AEO.
Stage D3-C702 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
AI Overviews and AI Mode are built on the Search infrastructure Google has used for 25 to 30 years and have very few processes of their own.
repeats
Slide D1-C038 Day 1 · How Search works and where's AI?
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Stage D3-C702 Day 3 · Wrapping all up: AI, Search, and making sense of everything.
AI Overviews and AI Mode are built on the Search infrastructure Google has used for 25 to 30 years and have very few processes of their own.
repeats
Stage D2-C726 Day 2 · How does the index look like?
AI Overviews and AI Mode use the same index structures and token-based snippets as classic web results, a point Google called important but not obvious.