To get cited in Google AI Overviews and ChatGPT, your website needs three things. It has to be in the index each tool draws its answers from: Google's normal search index for AI Overviews and AI Mode, and the dedicated crawlers of ChatGPT, Claude and Perplexity, which your robots.txt must let in. Each page has to answer its question in the opening lines, with clear facts that can be checked. And your structured data and language versions have to be marked up properly, so the tool knows who the business is and which page to show. None of this guarantees a citation, because no tool promises one to anybody. Without these basics, though, a page is ruled out before it starts.
AI Overviews use Google's normal index
AI Overviews and AI Mode don't have an index of their own. Google picks sources from pages it has already indexed and that are eligible to appear in Search with a text snippet. Google's own Search Central guidance on AI features says there are no extra requirements: no special markup and no new files.
That has a practical consequence. Whatever keeps a page out of normal search also keeps it out of AI Overviews:
- an accidental
noindex, often left over from a staging site; - pages blocked in robots.txt;
- content that only appears once JavaScript has run, which Google reads late or badly;
- duplicate pages with no canonical URL, where Google picks the wrong version;
nosnippet,data-nosnippetormax-snippetdirectives, which also limit what AI answers can use.
Search Console is still the right tool for checking what is indexed and why. A site with solid technical SEO and Google visibility already meets most of what AI Overviews ask for.
One point that often causes confusion: Google-Extended, which you can block in robots.txt, controls whether your content is used to train and ground Gemini models. It has no say over whether a page appears in Search or in AI Overviews.
ChatGPT, Claude and Perplexity have their own crawlers
The other tools don't rely on Google alone. Each runs its own named crawlers, and robots.txt decides whether they get in:
- ChatGPT uses
OAI-SearchBotfor search andChatGPT-Userwhen a user asks it to open a page, as set out in OpenAI's documentation on its bots; - Claude uses
Claude-SearchBotfor search andClaude-Userfor requests made by a user; - Perplexity uses
PerplexityBot, described in Perplexity's documentation.
These search crawlers are separate from the ones that collect training data, such as GPTBot and ClaudeBot. You can allow one group and block the other: appearing in answers doesn't mean handing over your content for training. It's a business decision, and one worth making deliberately. Requests made on a user's behalf, such as those from ChatGPT-User, may not follow robots.txt rules, as OpenAI itself notes; what decides whether you appear in answers is the search crawlers.
robots.txt isn't the only place these crawlers get turned away. Some CDN and firewall services block AI crawlers by default, often without the site owner knowing. Check your hosting settings and server logs to make sure requests from these crawlers get normal responses, not errors or verification challenges.
Then there's Bing. Microsoft Copilot relies on the Bing index, and other tools also draw on third-party search indexes. Adding your site to Bing Webmaster Tools and submitting your sitemap takes very little time and opens another door.
Writing to be cited
An AI tool cites a page when it finds a short, clear answer it can attribute to a source. How you write matters as much as the technical side.
Answer first, explain after. The first paragraph of every page should answer the question that page covers, in two or three sentences. Details, exceptions and examples follow. It's the same rule we apply to every service page on this site.
One topic per section. Headings that say what the section answers, such as "How long does delivery take?", help tools find the right passage. Vague headings like "Our approach" help nobody.
Concrete facts that can be checked. Opening hours, delivery areas, payment methods, lead times, terms. "We deliver anywhere in the country within two to three working days" can be quoted. "We're fast and efficient" can't.
Who wrote it, and when. Show the author, the publication date and the date of the last review. Update pages when the facts change. A page with stale information is a weak source for any tool.
Real frequently asked questions. An FAQ section with the questions customers actually ask, answered in two to four sentences each, gives tools passages that are ready to quote.
Text in the HTML, not in images. Key information inside images, scanned PDFs or components that only appear after several clicks is hard to read. Not every AI crawler runs JavaScript, so your main content should be in the HTML the server sends. On an older site, that sometimes means a redesign and speed-up.
Structured data and language versions
Structured data in JSON-LD tells tools, in a format they read effortlessly, who the business is and what each page is about:
OrganizationorLocalBusiness, with name, address, contact details and official profiles;Article, with author and dates, on articles;Serviceon service pages;BreadcrumbList, showing where the page sits on the site.
None of it is required for AI Overviews, and none of it guarantees anything on its own. But it backs up what the text says, and Google asks that it always matches the visible content. Structured data that contradicts the page does more harm than good.
On a site in more than one language, hreflang annotations link each page to its versions in other languages. Someone searching in Portuguese gets the Portuguese page, and someone searching in English gets the English one. Without them, tools may cite the wrong version or treat the two as duplicates.
What about llms.txt?
llms.txt is a proposed file that sits at the root of a site and summarises, in Markdown, what the site contains and where its most important pages are. The idea is to give language models a clean starting point.
Keep your expectations realistic. Google doesn't use it: its guidance on AI features says no special files are needed to appear in AI Overviews. The other tools haven't publicly committed to it either. We have one on this site because it's cheap to maintain and may help tools and agents that read it. But it's an extra. It doesn't replace indexing, permitted crawlers or well-written content.
An example
Picture a clinic with three service pages that all share the same generic text, and opening hours published as an image. Someone asks an AI tool, "Which clinic near me is open on Saturday morning?" The tool can't read the hours and finds no sentence that answers the question. The clinic is left out, even though it does open on Saturdays.
Now take the same clinic with its hours in plain text, one page per service that opens with the answer, a LocalBusiness block with the address and hours, and a robots.txt that lets the search crawlers in. The clinic now has what the answer needs. A citation still isn't guaranteed, but it has become possible.
Checklist
- Your key pages are indexed in Google, with no accidental
noindexornosnippet. - robots.txt allows
OAI-SearchBot,Claude-SearchBotandPerplexityBot. - Your CDN or firewall doesn't block those crawlers.
- The site is in Bing Webmaster Tools, with the sitemap submitted.
- Every page answers its question in the opening lines.
- The main content is in the HTML, not in images or JavaScript alone.
- Structured data matches the visible content.
- Language versions are linked with
hreflang. - Articles show an author and dates, and pages are reviewed when the facts change.
How to measure it
There's no perfect way to measure citations yet. Search Console folds AI Overviews clicks and impressions into your overall web search totals without separating them. In your site analytics, visits from ChatGPT often show up with chatgpt.com as the source, and visits from Perplexity with perplexity.ai. Combined with regular manual searches using the questions your customers ask, that gives a reasonable picture of progress.
If you'd like your site reviewed against these points, our AI search visibility service covers both the technical side and the content. Ask us for a quote: every project is quoted individually, starting from the site you already have.