AI SEO: how to get cited by ChatGPT, Claude, Gemini and Copilot

This is for the owner or marketer of a small or medium-sized business who hears that customers now ask ChatGPT, Claude, Gemini and Copilot instead of searching, and wants to know what it takes to be named in those answers. Each tool reaches your site through a search index or a crawler of its own, so your page has to be indexed, open to those bots, and written so that a short answer can be lifted from it and credited to you. No markup buys a citation. Below is what each of the four says it reads, in its own documentation, followed by the checklist we run on every page we build.

AI SEO, GEO and AEO: three names for one job

AI SEO, generative engine optimization (GEO) and answer engine optimization (AEO) all describe the same work: making a site easy for AI tools to find, read and quote. Google’s guide to generative AI search, last updated in July 2026, addresses the labels directly: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” What changes in practice is how a page is written. An AI answer quotes a passage, so that passage has to make sense on its own.

People also ask how to rank in ChatGPT. There is no list to climb the way there is on Google. The tool picks its sources for each question, and the same page can be cited in one conversation and missing from the next.

What each tool reads, in its own words

• OpenAI runs a separate search bot for ChatGPT, OAI-SearchBot, which OpenAI’s list of crawlers says is “used to surface websites in search results in ChatGPT’s search features”. Sites that opt out of it “will not be shown in ChatGPT search answers”. The training crawler, GPTBot, has a setting of its own, so you can block training and keep search.
• Anthropic lists two bots for Claude: Claude-SearchBot, which “navigates the web to improve search result quality for users”, and Claude-User, which fetches a page when someone asks Claude a question. Blocking either reduces your visibility in Claude’s answers.
• Gemini answers from Google Search. Google calls the process grounding: the model receives content from the Google Search index at the moment of the question and cites its sources. A site controls that use with the Google-Extended token in robots.txt, and Google’s crawler documentation states that the token “does not impact a site’s inclusion in Google Search nor is it used as a ranking signal”.
• Microsoft’s documentation for Copilot Chat says the service is “grounded in data from the public web from the Bing search service”, and that it sends Bing “web search queries consisting of a few words generated from the prompt”. A page Bing has not indexed cannot be in that answer.

So your robots.txt has to let the search bots through, whatever you decide about training bots, and you need to be indexed by both Google and Bing, because the four tools lean on different indexes.

How to appear in AI Overviews: what Google says

Google’s documentation on AI Overviews and AI Mode is short and worth reading in full. “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” A page needs to be “indexed and eligible to be shown in Google Search with a snippet”. There is “no special schema.org structured data that you need to add”, and no need for “new machine readable files, AI text files, or markup”. Its one piece of advice on structured data is that it should match the visible text on the page.

Since 31 August 2026, Search Console also has a Search generative AI setting, under Settings, where a site chooses whether its content can appear in AI Overviews, AI Mode and the AI features in Discover. Inclusion is the default.

If a vendor offers you an AI markup package, that documentation is your reply.

The checklist we run on every page

This is what we did on our own site, in two languages, and what we do on the sites we build.

• The answer sits in the first paragraph: two or three sentences that could stand on their own, then the explanation, then the proof. That paragraph is what a model can lift and credit.
• One page answers one question. A page that tries to answer six things gives a model six weak passages instead of one strong one.
• Every figure has a source you can open. A claim a reader can check is a claim a model can cite, and we do not publish statistics we have not read at their source.
• Subheadings carry the question the way someone would ask it, and AI tools quote them back as section titles.
• Structured data says only what the page visibly says. We add Organization, Service or Article, and BreadcrumbList to every page.
• Both languages are marked properly. For a site in English and Serbian, each version carries hreflang links naming itself and the other, plus x-default. Google’s documentation treats the three methods (HTML tags, HTTP headers or a sitemap) as equivalent; what matters is that the set is complete.
• Nothing already indexed gets lost. On the new site we built for a tile showroom with 2,160 products, every product keeps the address it has today, so Google loses none of the pages it already knows when the site switches over.
• Pages are fast on a phone. Most visits come from phones, and we test every page on a phone before handover.
• Bing Webmaster Tools and IndexNow are set up from day one, because Bing feeds Copilot. IndexNow lets a site announce new and changed pages itself, and Bing, Naver, Seznam.cz, Yandex and Yep support it. Google does not, so the sitemap and Search Console still do that job there.
• We keep an llms.txt file and promise nothing from it. The format is a proposal made by Jeremy Howard in September 2024 for a file that describes a site to language models, and the proposal page names no AI search product that reads it. Google’s guide says Google Search ignores such files, so they “neither harm nor help” a site there. Ours costs nothing, and nothing in our reports depends on it.
• We do not write a separate page for every way someone might phrase a question. Google’s guide warns that doing so “primarily to manipulate rankings or generative AI responses in Google Search violates Google’s scaled content abuse spam policy”.

What nobody can promise

A citation is the tool’s choice, made per question and per day. Google’s own advice on hiring SEO help applies here word for word: “No one can guarantee a #1 ranking on Google”, and if someone guarantees you first place, “find someone else”. The same goes for AI SEO services: an agency that guarantees you a ChatGPT citation is promising a decision that only ChatGPT makes. We promise the work above, visible on the page, and a check you can repeat.

How we track AI visibility

Google Search Console and Bing Webmaster Tools tell you what Google and Bing do with each page, and Search Console now has a Generative AI performance report for Google’s own AI features. We read them by page, not by query. Google’s documentation explains that queries not issued by more than a few dozen users over two to three months are anonymised: they are left out of the query table but still counted in the totals. The smaller your site, the larger that hidden share.

ChatGPT, Claude and Copilot have no such console, so we check all four tools by hand. Once a month, the same twenty questions your customers ask go into ChatGPT, Claude, Gemini and Copilot, and we record who is named, whether you are among them, and who is named instead of you. The first month is the baseline. Because the questions never change, the record shows after a few months whether anything moved.

What you can do this week

• Check your robots.txt for blanket rules on AI bots, and keep decisions about training bots separate from those about search bots.
• In Search Console, open Settings and check that Search generative AI still includes your site.
• Verify your site in Bing Webmaster Tools if you have only ever used Google.
• Open your three most important pages and read the first paragraph of each. If it does not answer the question in the title, rewrite it until it does.
• Write down twenty questions customers ask you, put them to all four tools, and keep the answers. That is your baseline, whomever you hire.

If you would rather go through it together, ask for a free website check or write to us. We look at what you have first, and only then suggest what should change.