Knowledge base v2.13.0 · Community edition · data through 2 October 2026
Content kit
125 terms in plain English, each tied to the claims that explain it.
Google Search's AI mode, where users ask longer questions (about three times as long as a traditional search on average, a Google slide said; the opening keynote said two to three times) and follow-up questions. It uses the same crawling and index as Search and adds grounding and query fan-out when it answers.
ThingAI Mode
AI-generated summaries shown in Google results with links to supporting pages. They are built on Google's core ranking systems and use content indexed by Google Search.
ThingAI Overviews
Low-quality, mass-produced AI-generated content. Google counts it as scaled content abuse and said it is going to great lengths to keep it out of its results.
An experimental Search Console chat box that turns a natural-language request into Performance report filters, applied only after the user confirms.
The other URLs of a duplicate cluster that Google keeps tracking, which is why an old domain can still appear after a site move.
Interfering with the browser's back button, for example sending users to an auto-generated page instead of where they came from. An explicit violation of Google's malicious practices spam policy, enforced since 15 June 2026.
AI systems in Google Search: BERT understands words in the context of the sentence, RankBrain interprets the intent behind queries, and MUM is used only for specific applications, not general ranking. On Day 3 Google called BERT a predictive language model, technically an LLM.
Text that mixes right-to-left scripts, such as Persian or Arabic, with left-to-right text, which can display in a confusing order in titles, product names and URLs.
Google's term for consumers who hand AI the mental effort of a purchase decision without surrendering the choice, and feel more confident thanks to ease, assistance and suggestion.
A cognitive shortcut in which consumers use a brand's name and reputation as a proxy for product quality, reliability and value, especially in the final reassurance step.
The representative URL of a duplicate cluster, the one Google would ideally show; signals such as links are forwarded to it.
ThingCanonicalization
A chain is a canonical pointing to a URL that leads on to another URL; a loop leads back to where it started. Google can follow chains but recommends pointing straight at the final URL; a community speaker called loops a structural conflict.
A file an SEO created as a satirical take on llms.txt, showing that crawlers fetch whatever files they find. Google said such files have no importance for Google Search.
A country-code top-level domain such as .de, one of the strongest signals of which country a site targets.
A Search Console tool for telling Google that a site moved from one domain or subdomain to another; Google's site move guide pairs it with submitting the new sitemap and updating internal links.
Two spam types in Google's policies: cloaking shows different content to users and to search engines to manipulate rankings; doorways are pages made to rank for similar queries that lead users to less useful intermediate pages.
Content based on very common knowledge that anyone could have written, with little insight or added value, such as a generic top-10 buying-tips list. Non-commodity content offers a unique, expert or experienced take.
The four attributes Google's rater guidelines use for main content: the extent to which a human put effort, originality, talent or skill and accuracy into creating it.
Google crawlers that serve a specific product by agreement with the crawled site, and the only Google crawlers that may ignore robots.txt; AdsBot, for example, ignores the global user agent with the ad publisher's permission.
Merchant Center feed attributes for AI shopping experiences such as AI Mode: question and answer, documents, related products, item group title, variant option and popularity rank. Retailers worldwide can use them since May 2026.
A significant, broad change to Google's ranking systems, made several times a year and announced on the Google Search Status Dashboard. Core updates do not target specific sites and are not penalties.
D3-C259 D3-C260 D3-C262 D3-C258
ThingCore updates
How much Google crawls a site: the combination of crawl rate limit and crawl demand, the finite resources Google allocates to crawling one website. A site with fewer than a few thousand URLs likely has no crawl budget problem, Google said.
D1-C090 D1-C096 D1-C373 D1-C381
ThingCrawl budget
How much Google wants to crawl a site, driven by its quality, how often its URLs change and how popular they are. Demand from other Google products, such as Ads, counts too, a Google panelist said.
ThingCrawl demand
How much crawling a host can take, shared by all Google crawlers. It goes down when connect time, time to first byte, 429 or 5xx responses go up. It applies per host, so a site's CDN, subdomains and main www host may be counted separately. A drop is most of the time an abrupt step down, and Googlebot then opens fewer connections, a Google panelist said.
D1-C091 D1-C092 D1-C132 D1-C376 D1-C493 D1-C494
ThingCrawl capacity limit
The part of Google's crawling that hands the crawler an ordered list of URLs from the crawl queue, putting pages that change often first; Google rarely discusses it in public because the details get taken out of context.
A Search Console report, under Settings, that shows how much Google crawls a site, the responses and errors it received, the file types fetched by its crawlers and the split between discovery of new URLs and refresh of known ones.
D1-C339 D1-C340 D1-C393 D1-C396
ThingCrawl stats
A Page indexing report status: Google crawled the page but did not index it; the page does not need to be resubmitted.
The Core Web Vitals metric for visual stability. A community speaker said layout shifts, such as a button popping in after the first load, annoy users and may confuse AI agents.
A Page indexing report status: Google knows the URL but has not crawled it yet.
The effect SEO work has on other channels, such as direct visits or paid search, which a community speaker advised measuring when AI answers take clicks.
A group of pages Google considers essentially equivalent: it stores one in the index and keeps track of the other URLs.
ThingDuplicate content
Search Console reports, rolled out worldwide by 31 August 2026, that show impressions, pages, countries, devices (Search only) and dates for generative AI features, with no clicks or queries. Since 24 September 2026 they also include multimodal searches; the regular Performance report still counts AI traffic too.
Labels for generative engine optimisation and answer engine optimisation. Google's answer: don't worry about the name, good SEO is good GEO and AEO.
ThingGEO
Google's Discover surface: indexed content that meets Discover's content policies is eligible automatically, with no special tags or structured data, and Discover uses many of the same signals and systems as Search.
ThingDiscover
Google's visual search, which Google said it built after seeing young users upload pictures and expect answers about them; more than 1.5 billion people use it every month, according to Google's May 2025 announcement.
ThingGoogle Lens
A robots.txt control token, not a separate crawler, that decides whether crawled content may train future Gemini models and ground Gemini apps and Vertex AI. It does not affect inclusion or ranking in Google Search. A site can disallow it for a directory or for the whole site.
ThingGoogle-Extended
The Google crawlers that fetch images and videos; robots.txt rules for their user agent tokens control how Google indexes a site's images and videos.
An AI model making things up when it lacks information or because of issues in its training. Google said it can happen with any model and, with current training methods, cannot be eliminated; grounding reduces it.
The browser feature (pushState) that lets a single-page app use real, clean URLs instead of # fragments, so Google can follow its links.
Google's policy that no website gets an advantage in Search because it is a Google client or partner or has a personal relationship with Google; the Search team gives no site special treatment, and Google does not accept payment to rank websites in organic search.
Annotations that tell Google which pages are language or regional versions of each other, so it can show the right one. They do not set a page's language.
Thinghreflang
The rule that if page X names page Y as a language version, page Y must name page X back; Google ignores annotations that point only one way.
1xx informational, 2xx success, 3xx redirection, 4xx client error and 5xx server error. Each class affects crawling differently, and 429 is the 4xx code that tells Google to slow down instead of saying there is nothing at the URL.
The last step before Google's index: it calculates thresholds and decides which documents are kept, using the signals calculated earlier, such as quality, language, country and spam.
ThingIndex selection
Loading content only when it is needed. Triggered by an Intersection Observer it can work for Google; triggered by scroll events or clicks it does not.
ThingLazy loading
A text file some sites add for AI systems. Google says it isn't necessary: Google Search ignores such AI text files, so they neither help nor harm. On Day 1 Google said llms.txt does not matter for Google Search right now and that it does not expect it to.
Thingllms.txt
The part of a page that serves its purpose, as opposed to header, navigation and footer: any part that directly helps the page achieve its purpose, including images, video, tools and user-generated content. Google identifies it and treats it as the most important part of the page.
D2-C310 D2-C028 D2-C862 D2-C870
ThingMain content
A robots rule for the largest image preview Google may show: none, standard or large, for example in Discover.
Thingmax-image-preview
A robots rule that caps a snippet at a number of characters: 0 means no snippet and -1 means no limit. It also limits how much of the page AI features may use.
Thingmax-snippet
Google's tool through which merchants send product data feeds for Shopping ads, Search ads and organic results. Products need it to appear in the Shopping tab; product rich results do not.
ThingMerchant Center
Google's model, used since 2019, of what happens between the trigger of a purchase journey and the purchase: people loop between exploring and evaluating until they feel confident enough to buy.
A hint not to follow links or pass signals through them; as a page-level robots rule it covers every link on the page. It does not block crawling.
Thingnofollow
A robots meta rule that tells Google not to index the images on a page. Google said it also affects the page's videos, because a video's thumbnail is an image.
A robots rule that tells Google not to show a page in search results. It works only if Google can crawl the page and see the rule.
Thingnoindex
nosnippet stops text snippets and video previews for a page and keeps its content out of AI Overviews and AI Mode as a direct input; the data-nosnippet attribute hides only a marked part of the page.
Thingsdata-nosnippet nosnippet
PageRank is the link-analysis algorithm Google's founders invented; BackRub was their 1995 Stanford search engine, soon renamed Google. Google's ranking systems guide says PageRank has evolved and is still part of its core ranking systems.
ThingPageRank
Search Console properties for Instagram, TikTok, X or YouTube accounts, verified through a pop-up from the platform, that show how the account's content performs on Google Search, Discover and Google News.
ThingPlatform properties
A classic retrieval structure: for a token, the list of URLs that contain it. Google intersects the lists of a query's important words to find candidate pages.
Two existing Merchant Center attributes Google now emphasises for AI: product_highlight, short bullets about the product itself, and product_detail, specifications as section, name and value.
A community speaker's name for journalistic and content work, such as interviews and podcasts, done with a prompt-assisted workflow that optimises material already recorded rather than creating more content.
Running several related searches at once to gather more results for an AI answer; a question about lawn weeds may also search herbicides and weed prevention. In Google's AI features an LLM generates the fan-out queries, and each goes through the same query understanding as a typed query; Google said they are not added to Search Console.
D1-C053 D2-C729 D3-C058 D3-C062 D3-C063
ThingQuery fan-out
Search Console groups of queries with very similar intent written differently, computed with AI and titled with the most prominent query. Shown in Search Console Insights for properties with many queries; they do not affect ranking.
ThingQuery groups
The first serving step: Google detects a query's language, separates its words, drops stop words where they matter little, recognises entities and expands the query with synonyms before looking it up. Google said it rewrites pretty much every query.
D3-C006 D3-C012 D3-C016 D3-C028
ThingQuery understanding
The raw HTML is what the server sends; the rendered HTML is the page after its JavaScript has run. Google's testing tools show the rendered HTML, and differences between the two point to rendering problems.
A series of redirects from one URL to the next. Google's crawlers follow up to 10 hops by default; robots.txt is checked for every URL in the chain, and Search Console reports a block on the first URL, a community speaker said.
The migration document that gives every old URL an approved destination; a community speaker also recorded, per URL, what must survive, who approves and the test that proves it works.
A link element in the head of a page that names the site owner's preferred URL. Google treats it as a strong signal, not an order.
Thingrel=canonical
Where pages wait to be rendered after crawling: a few seconds, but it can take longer, and Google documents no upper limit. On Day 3 Google estimated hours in the queue and said everything in it gets rendered within weeks.
Running a page's JavaScript in a browser engine to see what users see. Google renders with Chromium and indexes the rendered HTML.
ThingRendering
The serving step that looks up the query's words and their expansions in the index's posting lists and orders the candidates, by language, country and quality among other signals, before ranking starts.
A rich result with an average star rating and often the number of reviews, from Review structured data. Self-serving reviews of a site's own business are not eligible.
The IETF standard for robots.txt, officially the Robots Exclusion Protocol (REP), supported by basically all major crawlers. It controls which automated clients may access which URLs, not how the content is used, and parsers must skip lines they cannot parse.
Search results with extra visual features, such as review stars or recipe filters, usually powered by structured data. AI Overviews and AI Mode are not rich results.
D2-C454 D2-C521 D3-C324 D3-C325
ThingRich results
Google's tool to test and preview structured data: it shows the rich result features it detected and whether the markup is valid. It tests either a URL or a pasted code snippet.
ThingRich Results Test
An HTML tag in the head of a page that gives search engines page-specific rules, such as noindex or nosnippet, for indexing and display in results.
ThingRobots meta tag
A file in the root of each host (example.com/robots.txt) that tells crawlers which URLs they may fetch. It controls crawling, not indexing, applies only to its own host and is not a security measure: anyone can read it.
D1-C084 D2-C058 D2-C209 D1-C515 D1-C514
Thingrobots.txt
A Search Console report that shows the robots.txt file as Google last fetched it, with a version history and the errors and successes of each fetch. Google said it uses Google's open-source robots.txt parser.
Thingrobots.txt report
A Google spam policy against generating many pages mainly to manipulate rankings with little value to users, however they are made, including automated translation of scraped content.
ThingScaled content abuse
The shared markup vocabulary that major search engines announced together in June 2011; Google uses only the subsets its documentation declares.
ThingSchema.org
AI Mode agents set up by typing a natural-language request, which AI Mode interprets to create an alert; Google said they launched globally a day or two before 30 September 2026.
The export of Search Console performance data to BigQuery, the most complete query dataset available, with anonymized queries excluded, a community speaker's slide said.
Elements of the results page beyond plain text results, built as the last serving step after ranking, mostly from what Google extracted at indexing. Rich results are the features that usually need structured data; AI Overviews and AI Mode are not rich results.
A Search Console property setting (inherit, include or exclude) that controls whether a site appears in AI Overviews, AI Mode and generative AI features in Discover, without affecting web rankings.
Google Trends' measure: the share of all searches for a term, on a 0 to 100 scale. It is not search volume.
ThingSearch interest
A profile publishers and creators can claim in Google Search, with name, photo, bio, links and pinned posts, which users can follow. In the US, 10,000 followers across YouTube, Instagram, X or TikTok qualify since 16 September 2026. Its verified accounts are added automatically as Search Console platform properties.
ThingSearch profiles
External reviewers who rate results for changes Google is testing, following the Search Quality Rater Guidelines. They cannot affect individual sites' rankings, and nothing in their guidelines is itself a ranking factor.
A rule language for validating structured data. Google announced downloadable SHACL rules for each feature as coming soon.
A browser feature that isolates a component's styles. Google flattens shadow DOM content when it renders a page, so the content can be indexed if it appears in the rendered HTML.
Google's ways of testing a Search change: raters compare results with and without the change side by side, and a change that does well then runs on a small slice of live traffic, usually starting at 0.1% of people.
Technically different URLs that differ in only a few characters, such as protocol or host, capitalisation, delimiters, the encoding of a space or the order of parameters; a community speaker treated them as a sign of duplicate content and other technical issues.
A JavaScript application that changes views without full page reloads. Typical search risks: links without real URLs, # routes and error views that return 200.
D2-C183 D2-C186 D2-C190 D2-C288
ThingSingle-page app
The name Google shows for a site in results, an attribution feature rather than a rich result. Google generates it automatically, and a site can indicate its preferred name with WebSite structured data on its home page.
A page that returns a success code (200) while its content looks like an error or an empty page. Google keeps it out of the index but may keep crawling it. A CDN captcha page served with 200 is a common cause.
ThingSoft 404
A notable improvement to Google's automated spam detection systems, such as SpamBrain, listed with its ranking updates. Spam updates target spam pages; to recover, a site removes the spam Google described.
ThingSpam updates
Google's AI-based spam prevention system, launched in 2018 and continuously improved since; Google said it is now built on Gemini.
ThingSpamBrain
Words such as 'a' and 'of' that Google removes from a query when it thinks they matter little; where they matter, Google tries to recognise the whole phrase, stop words included, as an entity.
Google Shopping's own crawler. It crawls more often than Googlebot, goes through product, cart and checkout pages to verify merchant data, and robots.txt rules for it affect all Google Shopping surfaces.
ThingStorebot-Google
Machine-readable markup that describes a page's content, so Google understands it more accurately and the page can become eligible for rich results.
ThingStructured data
Synonyms are words people use interchangeably, which Google adds to a query automatically even when they are not synonyms linguistically; siblings are related words in the same category that are not interchangeable, such as Canon and Nikon, so a search for one should not show the other.
Splitting text into small units. Google Search stores words (tokens) with metadata, while AI models split text into sub-word pieces mapped to numeric IDs.
ThingTokenization
The neural network architecture of the paper 'Attention Is All You Need', the T in ChatGPT, which Google said it created and published to move the industry forward, competitors included.
Searches typed in Latin letters for a language written in another script, such as Persian or Hindi, with the same intent as a search in the native script; Google says a Hindi searcher typing in Latin letters may want and get Hindi results written either way.
The Google Trends tab that shows what people are searching for on Google right now, with approximate, bucketed search volumes.
ThingTrending now
An open standard for agentic commerce, covering discovery, buying and post-purchase support, launched by Google on 11 January 2026 with retailers and platforms.
Google's 2007 launch that blended results from Images, Maps, Books, Video and News into web results, so the results page mixes result types.
Search Console's tool that shows how Google sees a URL, including the rendered HTML; its live test can differ from the indexed version.
ThingURL Inspection
In robots.txt, one or more user-agent lines followed by their allow and disallow rules. A crawler follows only the most specific group that names it, so groups do not add up; Google combines all groups that name the same user agent.
Fetchers that request a page because a user asked for it, such as a translation service. Google's documentation says they generally ignore robots.txt, and Google said AI agents acting for a user fall into the same category.
Numeric representations of parts of documents and of queries that form a vector space. Besides posting lists, Google can retrieve documents by how close their embeddings are to the query's embedding.
A sitemap or feed that tells Google which URLs carry videos, so it can visit them without checking every page; Google said it ingests video sitemaps much more often than it can process HTML pages.
A page dedicated to one video. Google said a video without one is not going to be indexed, and its Video indexing report help says only videos on a watch page are eligible for indexing.
An HTTP header that carries the same rules as the robots meta tag, also for files that are not HTML. Like the tag, Google only sees it on URLs it is allowed to crawl.
ThingX-Robots-Tag