GEO glossary: AI search terms in plain words

AEO is the work of getting answer-first search features, such as the AI-written summary above the results, to use your page when they answer a question. It is close to a synonym for GEO: the AEO name stresses the direct answer and the GEO name stresses that an AI writes it, and Google's guide to generative AI search, updated in July 2026, explains the two terms side by side and says that from Google Search's point of view this work is "still SEO".

More in What is GEO, and how is it different from SEO?

Agentic Storefronts is the Shopify feature that sends a store's product data to AI channels including ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta; Shopify's help center says it is on by default for eligible stores, under Sales channels > Agentic in the admin. It runs on a different track from GEO: it carries product listings (title, description, images, price, availability), while whether your buying guides and answer pages get read and cited still depends on the pages themselves. Shopify's own documentation keeps the two apart too: product data goes mainly through Shopify Catalog, and AI crawlers that visit your pages on the open web are what robots.txt controls.

More in Shopify Agentic Storefronts: what store owners should check.

agents.md is a text file that Shopify serves automatically at yourstore.com/agents.md to tell AI agents how the store works; Shopify's help center says that by default it gives the store name, store URL, sitemap link, store policies and endpoints for other store data. It differs from llms.txt mainly in role: Shopify calls /agents.md the "canonical agent discovery URL" and keeps /llms.txt and /llms-full.txt for older AI crawlers that look for those names, and by default all three return the same content.

Not to be confused with AGENTS.md in code repositories, a file of instructions for AI coding assistants. More in Does llms.txt matter for an online store?

AI poisoning is planting false or inflated content where AI systems will read it, so they repeat it as fact: tampering with training data, faking authoritative data, or posting the same claim from many accounts and sites until it looks like a consensus. It differs from honest off-site work in whether the voices are real: answering questions people actually asked, under your own name, is fine; manufacturing the look of many people saying something is not. Google's spam policies count "attempting to manipulate generative AI responses" as spam, Bing's guidelines list content designed to manipulate language models among the abuses, and in April 2026 China's Cyberspace Administration named malicious marketing with GEO techniques as one way of poisoning AI data.

More in White-hat GEO: what we won't do, and why and Reddit, YouTube and AI answers: how to take part honestly.

AI search optimization is the plain-English name for GEO: getting AI search tools to find and cite your pages when they answer. It differs from GEO only in the label, and people who say "AI SEO" usually mean the same thing.

The definition and the work are under GEO; see also What is GEO, and how is it different from SEO?

AI visibility is a loose term for whether AI answer engines can find, read, trust and cite a website; the wording follows how the geo-score rubric defines AIV. It differs from citation rate in scope: citation rate is one measured outcome, while AI visibility can mean either how ready your site is or how often AI actually cites you, so ask which one someone means before you compare numbers.

Readiness can be scored, see AIV score; outcomes are under citation rate.

The AIV score is a readiness score out of 100 that geo-score, our open-source checker (MIT license), gives a website from the outside using the public rubric v1.1: 21 scored checks plus 4 bonus checks, each backed by evidence someone else can reproduce (rubric text). It differs from citation rate in what it measures: the AIV score covers things you can change and comes out the same whoever runs it, while whether AI actually cites you is reported separately and never folded in, because asking the same question twice can give different answers.

Try it on your own store: Free GEO audit.

An answer capsule is the first paragraph under a page's title: its first sentence answers the question the page is about, names the subject in full, and still makes sense when quoted on its own. It differs from an ordinary intro in that an intro says what the article will cover, while a capsule gives the answer; the paragraph at the top of this page is one. We write capsules at 50 to 80 words in English (80 to 160 characters in Chinese), which is our own house rule, not any engine's requirement: Bing's guidelines only ask for key information near the top of the page, and Google says you can ignore advice to "chunk" content for AI.

More in How to write a store page that AI answers will quote.

Citation rate is the share of AI answers that cite your website when you ask a fixed set of shopper questions (say, 20) across several assistants on a regular schedule. It differs from the AIV score in being an outcome: AI answers change from one run to the next, so you read the trend over weeks rather than any single screenshot, while the AIV score measures readiness and comes out the same whoever runs it. We track citation rate, but we don't set targets for it or use it for sign-off.

More in How to check whether AI recommends your store.

A citation is a link to your website in an AI answer's source list or footnotes; a mention is your brand or domain named in the answer text itself. The two are worth keeping apart because they often don't happen together: an assistant may name your store without linking to it, or list your page as a source without naming you in the answer, so log them separately.

How to log them: How to check whether AI recommends your store.

AI companies run three kinds of bots: search crawlers build the index an AI search tool pulls sources from when it answers (OAI-SearchBot, PerplexityBot, Claude-SearchBot); training crawlers collect data for training models (GPTBot, ClaudeBot); user-triggered fetchers open a page on the spot because someone asked a question in a chat (ChatGPT-User, Perplexity-User, Claude-User). The difference decides what you allow: to show up in AI answers, the search crawlers must get in, while blocking a training crawler, by each company's own description, keeps your content out of training. For user-triggered fetchers, OpenAI says robots.txt rules may not apply, Perplexity and Google say these fetchers generally ignore robots.txt, and Anthropic says Claude-User can be controlled through robots.txt.

Cloudflare's AI bot settings use the same three groups (Search, Agent, Training). How to check your store: Can AI crawlers reach your store? How to check.

Doorway pages are pages made to rank for a set of similar searches that pass visitors on to some other page that is actually useful; Google's examples include pages targeted at specific regions or cities that funnel users to one page. They differ from scaled content abuse in what gives them away: doorways are in-between pages that all lead to the same place, while scaled content abuse is about volume with little value. Both are in Google's spam policies.

Our test: a page that carries its own facts (local prices, specs, cases) and answers a question that group of buyers really asks is not a doorway; pages where only the city name or keyword changes, we don't build. See what we won't do.

Gate checks are three checks in the public geo-score rubric v1.1: whether search crawlers are blocked in robots.txt (g.robots), whether a request with those crawlers' user agents gets the real page (g.reachable), and whether the main text is in the HTML the server sends, without needing JavaScript (g.ssr). They differ from other lost points in that a zero on any one of them means crawlers cannot get the content at all, so the normalized percentage is capped at 40%; a middle score is a deduction, not a cap (rubric text).

Gate problems usually take days to fix. More in Can AI crawlers reach your store? How to check.

GEO is the work of making your web pages easy for generative AI tools such as ChatGPT, Google AI Overviews and Perplexity to find, read and cite when they answer a question. It differs from SEO in what it competes for: SEO is about your position in the list of links, GEO about whether the AI-written answer includes you. Bing's webmaster guidelines say GEO "focuses on content eligibility for grounding and reference in AI responses", and the same passage says GEO does not guarantee citations.

Most of the work is shared with SEO. More in What is GEO, and how is it different from SEO?

Google-Extended is a name you can use in robots.txt, not a separate crawler: it controls whether content Google crawls may be used to train future Gemini models and to ground answers in products such as the Gemini apps. It differs from Googlebot in reach: Google says blocking Google-Extended does not affect inclusion or ranking in Google Search, though the Gemini apps can no longer use your pages to ground their answers. AI Overviews and AI Mode are part of Search, and Google says the control for them is your robots.txt rules for Googlebot.

More in Can AI crawlers reach your store? How to check.

Grounding is when an AI system first retrieves relevant pages from a search index and then writes its answer from them, linking to the sources; Google says the technique is also known as retrieval-augmented generation, or RAG. It differs from training: training is what a model learned in advance, updated slowly and out of your view, while grounding happens at answer time, so a page that is indexed and readable has a chance to be used. Bing's AI Performance report has a "grounding queries" view: the phrases the AI used when it retrieved content.

More in How AI assistants decide which stores to mention.

A head term is a short search made of just a product category with no conditions attached, such as "best running shoes" or "power station". It is the opposite of a long-tail question: many people search it, but big brands and established review sites covered it long ago, so AI has plenty of sources to choose from, and competing for it with your own site is measured in months, often six or more.

We don't sell results on head terms and don't use them for sign-off. How the timing breaks down: How long until a new page shows up in Google and AI answers?

Indexing is a search engine analyzing a page it has crawled and storing it in its database, which is called the index; a page that is not in the index cannot be found through search, and AI search cannot use it either. It differs from crawling: crawling is a bot downloading the page, indexing is the next step, and a crawled page may still be left out, since Google says it does not guarantee indexing. Google also says a page has to be indexed and eligible to show with a snippet before it can appear in AI Overviews or AI Mode.

More in How long until a new page shows up in Google and AI answers?

IndexNow is a protocol a website uses to tell participating search engines that a URL was added, changed or deleted, instead of waiting for them to notice. It differs from a sitemap in direction: a sitemap is a list of URLs that search engines come and read, while IndexNow pushes each change to them. The engines on the official IndexNow list are Bing, Yandex, Seznam, Naver, Yep, the Internet Archive and Amazonbot; Google is not one of them. Bing's guidelines ask site owners to use it for additions, updates and deletions; OpenAI has not said whether it helps ChatGPT.

More in How long until a new page shows up in Google and AI answers?

LLMO is optimizing so that large language models such as ChatGPT, Claude and Gemini mention you and describe you correctly when they answer; it names roughly the same work as GEO and AEO, and we have found no official definition from any search or AI company. It differs from GEO in emphasis: "model" in the name suggests influencing training data, but what a store can actually influence is what the AI finds when it searches the web at answer time. On the training side, about all you control is a switch that keeps your content out of training, such as blocking OpenAI's GPTBot.

The term does not appear in the Google and Bing guidelines we checked. The work is the same as for GEO.

llms.txt is a Markdown file, usually at a site's root, that introduces the site and lists links to its important pages, so AI tools can get the picture quickly; it comes from a public proposal at llmstxt.org, not a standard set by any search engine. It differs from robots.txt in force: robots.txt says which crawlers may fetch which URLs, and the major AI search crawlers say they follow it, while llms.txt is only a description. As of 2026-10-08 we found no major AI search company saying officially that it reads llms.txt when it answers, and Google says that for Google Search you can ignore creating AI text files like llms.txt.

More in Does llms.txt matter for an online store?

A long-tail question is one a shopper asks with specific conditions attached, such as budget, use, model, size or a worry; an illustrative example is "portable power station under $500 for van life". It is the opposite of a head term: each long-tail question is asked by few people, but they are closer to buying, and "long-tail keyword" is the older SEO name for the same idea. Google's guide to generative AI search says you don't have to worry about having enough long-tail keyword variations, because its AI understands synonyms, so the job is to answer real questions fully, not to collect phrasings.

A page that answers one long-tail question well usually takes weeks to get indexed and start appearing in AI answers. More in Long-tail questions: why small stores can win them.

OAI-SearchBot is the crawler OpenAI uses for ChatGPT search, to surface websites in its search results. It differs from GPTBot, which collects training data: blocking GPTBot only keeps your content out of training, while if you block OAI-SearchBot, OpenAI says your site will not be shown in ChatGPT search answers, though it can still appear as a navigational link.

OpenAI recommends allowing it in robots.txt and allowing requests from its published IP ranges. How to check: Can AI crawlers reach your store? How to check.

Query fan-out is when an AI splits one question into several related searches, runs them at the same time, and combines what it finds into one answer. It differs from a long-tail question in who writes it: a shopper asks the long-tail question, while the AI generates the fan-out queries itself, and Google warns that creating a page for each fan-out query, primarily to manipulate rankings or AI answers, violates its scaled content abuse policy. Google's own example: for "how to fix a lawn that's full of weeds", fan-out queries might include "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn". OpenAI likewise says ChatGPT search typically rewrites a question into one or more targeted queries.

More in How AI assistants decide which stores to mention.

Scaled content abuse is a rule in Google's spam policies: generating many pages mainly to manipulate search rankings rather than help people, typically unoriginal content of little value, no matter whether people, AI or scripts made it. It differs from doorway pages in what it looks at: doorways are judged by whether pages only catch searches and pass people on, scaled content abuse by volume without value. Google's generative AI guide names one case directly: separate content for every variation of a query or every fan-out query, made primarily to manipulate rankings or AI answers.

More in Long-tail questions: why small stores can win them.

The generative AI report in Google Search Console shows how many times links to your site were shown in AI Overviews and AI Mode, broken down by page, country, device and date; Google's help page says it rolled out to all websites worldwide on August 31, 2026. It differs from the AI Performance report in Bing Webmaster Tools: Google's report counts impressions and does not appear until your site has enough of them, while Bing's shows how often you were cited in Microsoft Copilot, Bing's AI-generated summaries and some partner integrations, which pages were cited, and the grounding queries behind them.

More in How to check whether AI recommends your store.

Structured data is machine-readable information in a page's code, such as a product's price, stock status and rating, written with the schema.org vocabulary; Google recommends the JSON-LD format. It differs from the page text in who it is for: the visible text is for people and the markup is for search engines, and the two must match. Google says no special schema.org markup is needed to appear in AI Overviews or AI Mode, and Bing says structured data "may support clearer grounding" but does not guarantee visibility.

More in Does structured data help with AI answers?

White-hat GEO is GEO done only with methods that search engines and platforms allow: letting search crawlers in, answering shoppers' questions clearly on your own site, joining discussions under your real identity, and labeling paid content as ads. It differs from black-hat and gray-hat GEO in methods: buying and selling links, mass pages that only swap keywords, sock-puppet accounts, undisclosed paid posts and AI poisoning all appear in Google's spam policies or the abuse list in Bing's guidelines, and the penalty lands on your domain.

The rules behind each item: White-hat GEO: what we won't do, and why.