AEO glossary: answer engine and AI search terms

A Google Search feature that answers a question with a detailed AI-written response and links to supporting websites. Google says it is "particularly helpful for queries where further exploration, reasoning, or complex comparisons are needed" (opens in a new tab) (official documentation).

Why it matters. Google says AI Mode and AI Overviews "may use different models and techniques, so the set of responses and links they show will vary" (opens in a new tab), so a page can appear in one feature and not the other.

See: How to rank in Google AI Overviews and AI Mode

AI-written summaries that Google shows in some search results, with links to supporting pages. Google says they "help people get to the gist of a complicated topic or question more quickly" (opens in a new tab) and are shown only when its systems judge them "additive to classic Search, and as such, often don't trigger" (opens in a new tab) (official documentation).

Why it matters. People click through less when one appears: in browsing data from 900 U.S. adults (March 2025), users clicked a traditional search result on 8% of visits when an AI summary appeared, compared with 15% when none appeared (opens in a new tab) (browsing-panel study).

See: How to rank in Google AI Overviews and AI Mode

The practice of making a company's pages and facts easy for AI answer engines to find, cite, and recommend. An answer engine is a system that answers a question in its own words and names or links its sources, such as ChatGPT, Google's AI features, Perplexity, Claude, and Microsoft Copilot. Generative engine optimization (GEO) is another name for the same work, and Google's guidance for site owners refers to third-party advice about "AI experiences (sometimes called AEO for 'answer engine optimization' or GEO for 'generative engine optimization')" (opens in a new tab) (official documentation).

Why it matters. Many buyers now begin with an assistant: in a March 2026 survey of 1,076 B2B software buyers, 51% said they now begin software research with an AI chatbot more often than with Google (opens in a new tab) (vendor survey).

See: What is answer engine optimization (AEO)?

The three web robots that Anthropic documents for Claude. According to Anthropic, ClaudeBot collects web content that "could potentially contribute to" model training, Claude-SearchBot "navigates the web to improve search result quality for users," and Claude-User may visit a site when a person asks Claude a question (opens in a new tab) (official documentation). Anthropic says its bots honor robots.txt, including for Claude-User's user-initiated requests (opens in a new tab).

Why it matters. Each can be allowed or blocked separately, and Anthropic says blocking Claude-SearchBot "may reduce your site's visibility and accuracy in user search results" (opens in a new tab), while blocking ClaudeBot signals that the site's future content should be left out of training.

See also: How to rank in Claude

A source that an AI answer names or links to support what it says. A citation is not the same as a recommendation, which is a company an answer suggests as an option, and AEO HQ measures the two separately. Citations do not always support the text beside them: in a 2023 audit of four generative search engines, only 51.5% of generated sentences were fully supported by citations, and 74.5% of citations supported the sentence they were attached to (opens in a new tab) (peer-reviewed).

Why it matters. Citations are the part of an AI answer that a site owner can count: Bing Webmaster Tools' AI Performance report shows citations, cited pages, and grounding queries for Copilot and Bing's AI summaries (opens in a new tab), though not clicks (official documentation).

See also: How to measure AI visibility

A crawler visits pages automatically, for example to build a search index; Google calls crawlers "automated programs" (opens in a new tab) that download text, images, and videos from pages they find (official documentation). A user-initiated fetcher visits a page only when a person's request, such as a question to an assistant, needs it. Google says its user-triggered fetchers "generally ignore robots.txt rules" (opens in a new tab) because a user requested the fetch (official documentation).

Why it matters. The two need different controls: robots.txt governs crawlers, but OpenAI says robots.txt rules "may not apply" to its user-initiated ChatGPT-User agent (opens in a new tab), and Perplexity says its Perplexity-User fetcher "generally ignores robots.txt rules" (opens in a new tab) (official documentation).

See also: How ChatGPT, Gemini, Claude, Perplexity, and Copilot find and cite sources

A distinct thing that search systems and language models can identify and keep apart from others, such as a company, a person, a product, or a place. Keeping entities apart is hard when they share a name: language models often yield ambiguous answers or incorrectly merge information belonging to different entities that share a name (opens in a new tab) (peer-reviewed).

Why it matters. An assistant that merges two companies or two people with the same name can describe one with the other's facts, which is why AEO HQ recommends using the same name, description, and profile links everywhere a company or founder appears.

See also: Entity and brand consistency checklist

Another name for work that aims to get content retrieved, cited, and used in AI-written answers. The term comes from a research paper first posted to arXiv on 16 November 2023 (opens in a new tab) and published at the KDD 2024 conference (opens in a new tab) (peer-reviewed). Bing's webmaster guidelines say GEO "focuses on content eligibility for grounding and reference in AI responses" (opens in a new tab) (official documentation).

Why it matters. Much of the academic research on the topic uses this label: a 2026 critical survey reviewed 45 studies of it (opens in a new tab) (preprint).

See: What is generative engine optimization (GEO)?

A robots.txt product token, not a separate crawler, that controls whether content Google crawls may be used to train future Gemini models and for grounding in Gemini Apps and on Vertex AI (opens in a new tab) (official documentation). Google says it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search" (opens in a new tab).

Why it matters. Because the token also covers grounding, which Google describes as providing content from the Google Search index to the model at prompt time (opens in a new tab), blocking it also keeps the site's content out of grounding in Gemini Apps, not only out of training.

See also: How ChatGPT, Gemini, Claude, Perplexity, and Copilot find and cite sources

Basing an AI answer on source material retrieved for the question, rather than on the model's memory alone. Bing says that in AI-generated answers, grounding "refers to the source material and web evidence the system uses to support and cite its response" (opens in a new tab), and Google says retrieval-augmented generation is "also known as grounding" (opens in a new tab) (both official documentation).

Why it matters. Grounded answers draw on specific pages, and Bing shows site owners the grounding queries behind their citations, which it describes as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers" (opens in a new tab) (official documentation).

A protocol that lets a website tell participating search engines when a URL is added, updated, or removed. The IndexNow FAQ lists Amazon, Bing, Naver, Seznam.cz, Yandex, and Yep, but not Google, as participating engines, shares each submission with all of them, and says "Submitting a URL does not guarantee immediate indexing" (opens in a new tab) (protocol documentation).

Why it matters. Bing asks site owners to use IndexNow when URLs are added, content is updated, or URLs are removed, and says timely notifications "reduce outdated or incorrect URL references in Copilot responses and grounding results" (opens in a new tab) (official documentation).

See also: Technical SEO checklist for AI search

A search tool's estimate of how hard it is to reach the first page of Google for a query. Ahrefs, whose data AEO HQ uses, measures it from 0 to 100, bases it on the number of referring domains (other websites linking in) that the top 10 organic results have, and says it does not take on-page SEO factors into account (opens in a new tab) (vendor documentation).

Why it matters. It helps compare queries with one another, but it measures links to the pages that rank in Google's organic results, not how hard it is to be cited in an AI answer.

See also: How AEO HQ measures AI visibility, and how our research is done

The date after which a language model has little or no information from its training data. Anthropic lists two such dates for each Claude model, a reliable knowledge cutoff and a training data cutoff, and they can differ: for Claude Haiku 4.5 they are February 2025 and July 2025 (opens in a new tab) (official documentation).

Why it matters. Anything newer than the cutoff, such as a new company or a changed price, can reach an assistant's answer only through search or through what the user supplies; Anthropic says Claude searches when a request depends on information that is "current, changing, or outside its training data" (opens in a new tab) (official documentation).

A proposed Markdown file, placed at a site's root (/llms.txt) or under a path, that gives language models a short summary of a site and links to its key pages. The proposal, first written in 2024, describes a file that "offers brief background information, guidance, and links to detailed markdown files" (opens in a new tab) (proposal document).

Why it matters. Evidence of use by AI search systems is thin: of about 38,000 domains that published a valid llms.txt file, 97% received no requests for it in May 2026 (opens in a new tab) (vendor log study), and Google says Google Search does not use AI text files such as llms.txt, including for its generative AI features (opens in a new tab) (official documentation).

See: llms.txt: what it is and whether it matters

The web agents OpenAI documents for ChatGPT and for model training. OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features," and GPTBot crawls content that "may be used in training" OpenAI's foundation models (opens in a new tab) (official documentation). ChatGPT-User visits a page when a user's question in ChatGPT or a custom GPT calls for it and does not crawl automatically, and a fourth agent, OAI-AdsBot, checks pages submitted as ads (opens in a new tab).

Why it matters. OpenAI says each setting "is independent of the others," and that sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers" (opens in a new tab), so a site can block GPTBot's training crawl and still appear in ChatGPT search (official documentation).

See also: ChatGPT SEO: how to rank in ChatGPT search

What a language model learned during training and can recall without searching. A 2023 study showed that a model's ability to answer a fact-based question relates to how many documents about that question it saw during pre-training, and that retrieval can reduce that dependence (opens in a new tab) (peer-reviewed; a background study from before AEO HQ's September 2023 evidence window).

Why it matters. A new or rarely discussed company is unlikely to be in a model's parametric knowledge: among 112 startups from the 2025 Product Hunt leaderboard, a model without web access named 5.4% of them at least once in discovery-style questions, while a search-based model named 27.7% (opens in a new tab) (preprint; a master's thesis).

See also: How to get cited and recommended by ChatGPT

The two web agents Perplexity documents. PerplexityBot "is designed to surface and link websites in search results on Perplexity" and "is not used to crawl content for AI foundation models"; Perplexity-User visits a page when a user asks Perplexity a question and, because a user requested the fetch, "generally ignores robots.txt rules" (opens in a new tab) (official documentation).

Why it matters. To appear in Perplexity's search results, Perplexity recommends allowing PerplexityBot in robots.txt and permitting requests from its published IP ranges, and explains how to allow its bots through web application firewalls (opens in a new tab), so firewall rules matter as well as robots.txt (official documentation).

See also: How to rank in Perplexity

Splitting one question into several related searches and writing one answer from all the results. Google defines it as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query" (opens in a new tab); in its example, a question about fixing a lawn full of weeds fans out to searches such as "best herbicides for lawns" and "remove weeds without chemicals" (official documentation). OpenAI says ChatGPT search likewise "rewrites your query into one or more targeted queries" (opens in a new tab) (official documentation).

Why it matters. An answer can cite pages that rank for the sub-questions rather than for the question as typed: in a 2025 audit of 4,706 queries, 53% of the domains that AI Overviews consulted were outside Google's organic top 10 (opens in a new tab) (preprint).

The step in which an answer engine runs searches and pulls candidate pages into the model's working context before it writes. Google says its AI features rely on its "core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index" (opens in a new tab) (official documentation).

Why it matters. A page that is not retrieved cannot be cited, and plain wording helps at this step: in a 2026 laboratory pipeline, rewrites that added technical terms or unique words caused the largest drops in retrieval, because their words no longer matched the query's (opens in a new tab) (peer-reviewed).

See also: Answer engine optimization (AEO): the complete guide

A system design in which a language model first retrieves documents relevant to a question and then writes its answer from them. Google says its AI features in Search use RAG, which it describes as a technique used "to improve the quality, accuracy, and freshness of AI responses" (opens in a new tab) (official documentation).

Why it matters. Engines do not retrieve for every question: in a 2025 study, GPT-4o with a search tool retrieved a median of zero links and answered static questions from its internal knowledge (opens in a new tab) (preprint), so the same assistant can answer one question from retrieved pages and another from memory.

A plain-text file at the root of a website that tells automated crawlers which parts of the site they may visit. It follows the Robots Exclusion Protocol, first defined in 1994 and specified by the IETF as RFC 9309 in 2022, whose rules "are not a form of access authorization" (opens in a new tab) (Internet standard). Under the standard, a crawler obeys the group of rules that names it and uses the "*" group only when no group matches (opens in a new tab).

Why it matters. AI companies use separate robots.txt names for search, training, and user-requested visits, so one rule can remove a site from an assistant's answers while another only opts it out of training; OpenAI, for example, says "each setting is independent of the others" (opens in a new tab) (official documentation).

See also: Technical SEO checklist for AI search

A schema.org property, used in structured data, that lists the web addresses of other pages about the same person or organization, such as its profiles on other sites. Google's documentation describes it as "the URL to other external profiles or home pages for the profile" (opens in a new tab) (official documentation).

Why it matters. It helps search systems tell apart people and companies that share a name: Google's article markup documentation asks for an author URL that "uniquely identifies the author" and accepts sameAs links to the author's other profiles (opens in a new tab) (official documentation).

See also: Entity and brand consistency checklist

A search engine's database of the pages it has crawled and processed, which it searches when someone enters a query. Google describes its index as "a large database hosted on thousands of computers" and says "Indexing isn't guaranteed; not every page that Google processes will be indexed" (opens in a new tab) (official documentation).

Why it matters. A page missing from the index an engine uses cannot be retrieved or cited by that engine; Google, for example, says a page must be "indexed and eligible to be shown in Google Search with a snippet" (opens in a new tab) to appear as a supporting link in AI Overviews or AI Mode (official documentation).

See also: How ChatGPT, Gemini, Claude, Perplexity, and Copilot find and cite sources

AEO HQ's name for its approach to AI search optimization: search engine optimization (SEO) plus four additions, which are third-party corroboration, specific and verifiable facts, measurement that treats AI answers as distributions, and offers that software can read.

Why it matters. The name reflects that search basics come first: Google says "the best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems" (opens in a new tab) (official documentation).

See: AI search optimization: SEO plus corroboration, facts, and measurement

A company's mentions as a share of all mentions of a named set of competitors, counted across a fixed set of prompts and runs. For example, if answers name the companies in a five-company set 100 times in total and name one of them 20 times, that company's share of voice is 20%. Bing Webmaster Tools reports a related metric, Citation Share, which it calculates as "the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query" (opens in a new tab) (official documentation).

Why it matters. It shows a company's position against competitors, which a mention rate alone does not, but it changes with the competitors chosen, so the competitor set should be named alongside the number.

See also: AEO metrics and KPIs: definitions

Whether Google may show a page in its results with a snippet, the short text preview of the page. Site owners can limit snippets with the nosnippet, data-nosnippet, and max-snippet controls, or keep a page out of results with noindex (opens in a new tab) (official documentation).

Why it matters. Google says a page must be indexed and eligible to be shown with a snippet (opens in a new tab) to appear as a supporting link in AI Overviews or AI Mode, and that the site must also be included in Search Console's Search generative AI setting, which is the default (opens in a new tab) (official documentation).

See also: How to rank in Google AI Overviews and AI Mode

Code in a page, usually JSON-LD that uses the schema.org vocabulary, that labels the page's facts for software, such as a product's price or an article's author. It is also called schema markup. Google describes it as "a standardized format for providing information about a page and classifying the page content" (opens in a new tab) (official documentation).

Why it matters. Google says structured data isn't required for its generative AI features and there is no special schema.org markup to add, though it helps pages qualify for rich results (opens in a new tab) (official documentation), and in a matched study of 1,885 pages that added it, AI citations did not rise (opens in a new tab) (vendor study).

See: Does schema markup help AEO? What the evidence says