Home / Glossary
Reference updated 2026-11-22

The AEO glossary

38 terms you'll meet when making a business visible to AI assistants. Where a standard or an operator defines a term, we link to it. Where nobody does, we say so and tell you how we use it.

Answer engine optimization · Generative engine optimization · Citation · robots.txt · User-agent token · Training crawler · Search crawler · User-triggered fetcher · Named-group trap · Policy drift · Grounding · llms.txt · Markdown copy · Agentic Resource Discovery · Agentic Browsing · WebMCP · JSON-LD · Organization schema · sameAs · FAQPage schema · Offer schema · Entity · Wikidata and P856 · Common Crawl · Soft 404 · Crawler verification · Bot challenge · Discovery question · Diagnostic question · Answer-level counting · Cited but not named · Question set · Page-level citation · AI referral · Delivery · lastmod · IndexNow · Fact Ledger

Answer engine optimization (AEO)
Making a site easy for AI assistants and answer engines to read, trust and cite: crawler access, machine-readable facts, quotable answers and a clear entity. No standards body defines it; this is how we use the term. Our usage; no standard definition.
Generative engine optimization (GEO)
A near-synonym for AEO, coined in a 2023 research paper that measured how page edits change visibility in generative search answers. Source: arxiv.org/abs/2311.09735
Citation
A link or named source an AI answer shows for a claim. Being cited is different from being mentioned: a mention names you, a citation sends the reader to your page. Our usage; no standard definition.
robots.txt
A plain-text file at /robots.txt that tells crawlers which paths they may fetch, grouped by User-agent. Standardised as RFC 9309. A crawler that matches a named group ignores the "User-agent: *" group entirely. Source: rfc-editor.org/rfc/rfc9309
User-agent token
The name a crawler matches in robots.txt. Some tokens are not crawlers: Google-Extended and Applebot-Extended are read by Googlebot and Applebot to decide whether crawled content may be used for AI training. Source: developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers
Training crawler
A crawler whose operator says the content may be used to train models, such as GPTBot, ClaudeBot or CCBot. Blocking one does not, by the operators' own accounts, remove you from their search features. Source: crawlers.html
Search crawler
A crawler that builds the index an assistant searches when it answers, such as OAI-SearchBot, Claude-SearchBot or PerplexityBot. Blocking one can stop that assistant citing you. Source: crawlers.html
User-triggered fetcher
An agent that fetches one page because a person asked an assistant to read it, such as ChatGPT-User or Claude-User. Several operators say these may not follow robots.txt, because a human requested the fetch. Source: crawlers.html
Named-group trap
A robots.txt mistake: a crawler that matches its own User-agent group ignores the "User-agent: *" group completely, so Disallow lines written only under * do not apply to it. Repeat the lines you rely on in every named group. Source: rfc-editor.org/rfc/rfc9309
Policy drift
When your robots.txt stays the same but the agents reading it change: an operator ships a new crawler or changes what an existing one does. Our usage; no standard definition.
Grounding
Retrieving live sources and basing an AI answer on them, instead of relying only on what the model learned in training. Google says the Google-Extended token also controls grounding in Gemini Apps. Source: developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers
llms.txt
A proposed Markdown file at /llms.txt: an H1 with the site name, a one-line summary in a blockquote, then sections of annotated links to the pages that matter. A proposal, not a standard: check each assistant's own documentation before assuming it reads the file. Source: llmstxt.org
Markdown copy
A plain Markdown version of a page at its URL plus .md (pricing.html.md, or index.html.md for a folder's page), which the llms.txt proposal suggests for agents that would rather not parse HTML. A page can point to its copy with a link tag (rel="alternate", type="text/markdown") or a Link header. Built from the HTML at deploy time, it can't drift from the page; a hand-kept copy can, and then agents read old facts. Source: llmstxt.org
Agentic Resource Discovery (ARD)
A proposed JSON manifest, ard.json, listing a domain's agent resources (MCP servers, A2A agents, agent skills) so agents and registries can find them. It lives at /.well-known/ard.json or wherever robots.txt (Agentmap:), a <link rel="ard"> or a Link header points; its old name was ai-catalog (/.well-known/ai-catalog.json, rel="ai-catalog"). Each entry has a urn:air: identifier, a display name, a media type and a URL or inline data. Chrome's Lighthouse checks it in its experimental Agentic Browsing category. Source: developer.chrome.com/docs/lighthouse/agentic-browsing/ard
Agentic Browsing (Lighthouse)
An experimental Lighthouse category in Chrome 150 or later that checks how well a page works for AI agents operating it: llms.txt, an ARD manifest, WebMCP tools and their schemas, an accessibility tree with names and labels, and a stable layout. It shows how many checks pass instead of a 0 to 100 score, because the standards are still emerging. Source: developer.chrome.com/docs/lighthouse/agentic-browsing/scoring
WebMCP
A proposed browser API for a page to offer its own actions to AI agents as tools. Declared in HTML, a form carries toolname and tooldescription attributes; scripts can register tools too. Experimental: Chrome's Lighthouse reports forms without the attributes, for information only. Source: developer.chrome.com/docs/lighthouse/agentic-browsing/forms-missing-declarative-webmcp
JSON-LD
A way to embed structured data as JSON inside a script tag. It is the format Google recommends for structured data, using the schema.org vocabulary. Source: developers.google.com/search/docs/appearance/structured-data/intro-structured-data
Organization schema
A schema.org Organization (or LocalBusiness) block on your homepage that states your name, URL, logo and official profiles in one place. Source: schema.org/Organization
sameAs
A schema.org property listing URLs that identify the same entity: your LinkedIn page, Wikidata item, GitHub organisation. It ties scattered profiles to one entity. Source: schema.org/sameAs
FAQPage schema
Structured data marking up questions and their answers on a page. Google stopped showing FAQ rich results in 2026 and removed their documentation, so the markup no longer changes how you look in Google; it still states your Q&A explicitly for other parsers. Source: developers.google.com/search/updates
Offer schema
A schema.org Offer states a price, its currency and what it buys, so software can read your prices without parsing the page. Markup that isn't updated when prices change can be worse than none: an assistant can quote the stale number with confidence. Our audit compares the two. Source: schema.org/Offer
Entity
A distinct thing (a company, product or person) that a knowledge graph can identify unambiguously, usually by an ID such as a Wikidata Q-number, rather than by a name that others may share. Our usage; no standard definition.
Wikidata and P856
Wikidata is the free, structured knowledge base behind Wikipedia. Property P856 records an item's official website, which is how a tool can go from your domain to your entity. Source: wikidata.org/wiki/Property:P856
Common Crawl
A non-profit that publishes a free archive of web crawls, a common source for model training datasets. Its crawler is CCBot. Source: commoncrawl.org
Soft 404
A missing page that returns 200 OK with other content, often the homepage, instead of a 404 status. A classic trap for /llms.txt: tools think the file exists and read your homepage HTML as if it were the file. Source: developers.google.com/crawling/docs/troubleshooting/http-status-codes
Crawler verification
Checking that a request claiming to be a known crawler really came from its operator, because anyone can send any user agent. Google documents reverse DNS and published IP range files; OpenAI, Bing, Perplexity and Apple also publish range files. Source: developers.google.com/search/docs/crawling-indexing/verifying-googlebot
Bot challenge
A JavaScript or CAPTCHA wall a CDN shows to suspected bots. If it is served to search crawlers and fetchers, they see the challenge page instead of your content. Our usage; no standard definition.
Discovery question
A buyer question that doesn't name you ("best CRM for small agencies?"). Answers to it show whether buyers find you. Reported apart from diagnostic questions, because a high rate on one says nothing about the other. Our usage, following the method Jam (spreadjam.com) published. Source: method.html#questions
Diagnostic question
A question that names you ("does Acme have SSO?"). Answers to it show whether what assistants say about you is right, not whether they find you. Our usage. Source: method.html#questions
Answer-level counting
Counting each AI answer once, however many times it mentions you: five mentions in one answer are one answer. Rates are answers that named you out of answers collected, with the denominator shown. Our usage, following Jam's published method. Source: method.html#count
Cited but not named
An answer links to your page but never names you. That's source visibility, not a recommendation: your page informed the answer, but a buyer reading it doesn't learn your name. Our usage. Source: method.html#count
Question set
The buyer questions you track, written down and frozen: each with an id, a version and whether it names you (diagnostic) or not (discovery). Rewording a question makes a new version, and retiring one keeps its history, so a trend never mixes two wordings. Keep one wording per buying decision. Our usage. Source: track.html#own
Page-level citation
Counting which exact pages an answer links to, not just which sites: one answer counts a page once, campaign parameters and #sections are set aside, and query strings that can change the document are kept. It tells a pricing page from an outdated tutorial on the same domain. Our usage. Source: method.html#count
AI referral
A visit that arrives from an AI assistant, recognised by the referrer (chatgpt.com, perplexity.ai…) or a utm_source the assistant adds to links. Counts are floors: visits from assistants' apps can arrive with no referrer and look like direct traffic. Source: logs.html#export
Delivery (of a fix)
Whether a change to a page actually reached the crawlers behind AI answers: live, in the HTML without JavaScript, allowed in robots.txt, indexable, in the sitemap with an accurate lastmod, and linked. Check it before reading answers, or you can't tell "the fix didn't work" from "it never arrived". Source: delivery.html
lastmod
A sitemap field giving the date a page last changed significantly. Google says it uses lastmod only when it's consistently and verifiably accurate; a sitemap that stamps its build date on every URL tells a crawler nothing about which page changed. Source: developers.google.com/search/docs/crawling-indexing/sitemaps/build-sitemap
IndexNow
A protocol for telling search engines that a URL changed: one POST with the URLs and a key you host on your site. Engines that take part share submissions with each other; Bing and Yandex are among them. Source: indexnow.org
Fact Ledger
EarnCitations' format for a signed list of a business's key facts (prices, plans, claims), each with a source and date, published at /.well-known/facts.json. Our own proposal, not a standard. Source: ledger.html
See how your site does on each of these with the free AI-readiness audit.