How to get cited by ChatGPT, Perplexity and Google's AI surfaces

How to get cited by ChatGPT, Perplexity, Google's AI surfaces or Copilot starts with a fact that is easy to miss: none of them publish how sources are ranked. What every one of them does publish is the plumbing — which crawler feeds which surface, and which control governs it. That plumbing is checkable in an afternoon, and getting it wrong is the reason most pages never enter the pool in the first place.

Read this first

This chapter is the hub for a cluster. It covers what is common to all four engines and hands off to the specifics: which crawlers belong to which company is in one company, several crawlers, and the directive that governs how much of your page may be quoted is in nosnippet is the real AI lever.

Before any of it helps, your pages have to be fetchable and have to contain text in the delivered HTML. If you have not confirmed that, start there instead — nothing below matters to a page that returns an empty shell.

What is published, and what is not

The distinction is worth stating in one line because most writing on this topic blurs it. Crawler names, user agents, IP ranges and opt-out mechanisms are documented by each vendor. Selection and ranking logic is documented by nobody.

So there is no honest answer to "what makes ChatGPT pick source A over source B". There is a very concrete answer to "what has to be true before your page is even a candidate", and that is what this chapter is about.

Which crawler feeds which surface

This is the table to keep. Every row comes from the vendor's own documentation, checked 2026-08-17.

The column that does the work is the middle one. Four surfaces, and no two of them are fed by the same agent, so a permission granted in one place buys you nothing in another.

Two consequences fall straight out of it. Blocking GPTBot does not remove you from ChatGPT's search results, because that surface is fed by a different agent — OpenAI's documentation states the settings are independent. And there is no separate crawler to admit for Google's AI surfaces or for Copilot; if you are in the underlying index, you are already in the pool.

Perplexity says its search crawler is not a training crawler

Worth quoting because it removes a common worry. Perplexity's bot documentation says PerplexityBot "is designed to surface and link websites in search results on Perplexity. It is not used to crawl content for AI foundation models."

That means the trade people usually assume they are making — visibility in exchange for training data — is not the trade on offer here. Whether you believe a vendor's statement about its own use of your content is a judgement call, but the statement is on the record and dated.

The deliverable: what to get cited by ChatGPT and the rest actually requires

Run this once per site. Every line is verifiable from outside with curl or a browser, which is the point — nothing here depends on a vendor telling you how you are doing.

1. Fetch your page with curl and confirm the answer text is in the delivered HTML, not injected later.
2. Check robots.txt for OAI-SearchBot and PerplexityBot separately from GPTBot. Confirm the first two are allowed if you want to appear in those search surfaces.
3. Confirm Googlebot and Bingbot are not blocked anywhere — including at the CDN, which is a separate decision from robots.txt.
4. Check your meta robots and X-Robots-Tag for nosnippet. It removes you from Google's AI surfaces even while you rank.
5. Confirm each page states its own claim in the first 80 words under a heading, so an extractor has a bounded passage to lift.

Anything that fails one of the first four is a plumbing problem and gets fixed once. Line five is the ongoing work, and it is the only line where more effort keeps paying.

Order matters here more than it looks. Every one of the first four can silently cancel the ones after it, so working top to bottom saves you from tuning a paragraph on a page that no agent is allowed to fetch. Run the whole list on one representative page before rolling anything out across the site.

Two controls people mix up

Both are separate from robots.txt, and both are the reason a site that looks fully open still misses AI surfaces.

Google-Extended in particular gets set defensively and then blamed for traffic changes it cannot cause. The full accounting of what it does and does not cost is in Google-Extended opt out.

Three ways this goes wrong

All three are configuration failures that look like content failures, which is why they survive so long — the site owner goes off and rewrites a page that was never the problem.

1. Blocking one company's training bot and assuming its search surface went with it. The settings are documented as independent, and the result is an invisible half-block.
2. Fixing robots.txt while the CDN still challenges the same agents. Your file says yes and the edge says no, and the edge wins.
3. Optimising the writing before checking the plumbing. A page nobody can fetch does not benefit from a better first paragraph.

Where this chapter stops

We cannot tell you why one page gets cited and a similar one does not. Nobody outside these companies can, and any source that gives you a ranked list of factors is describing a belief, not a specification. We also cannot tell you how often a given engine re-fetches, because that is not published either.

What this chapter does is bound the problem. Everything above is checkable, and once it is all true, the remaining variable is the content itself. If the plumbing is right and you are still not being cited, the next thing to change is the page, not the configuration — and that is a different chapter.

Put plainly: how to get cited by ChatGPT is two questions wearing one coat, and only the first has a documented answer. Keeping all five checks true across a whole site, rather than on the one page you happened to test, is what QueryWin is being built to do.

Common questions

Does blocking GPTBot stop ChatGPT citing me?

No. OpenAI documents GPTBot and OAI-SearchBot as independent settings, and the search surface is fed by the latter. Blocking the training bot leaves the search path intact.

Is there a crawler to allow for AI Overviews?

No separate one. AI Overviews draw on Google's regular index, so the relevant permission is the one you already gave Googlebot. The lever that does affect appearance there is max-snippet.

Should I allow ChatGPT-User and Perplexity-User?

Those two are the user-triggered agents: they fetch because someone in the product asked about your page specifically. Blocking them means the one visitor who explicitly went looking for you gets an apology instead of your content. It is a different decision from the indexing bots and usually an easier one.

How do I know if an engine actually visited?

Server logs, matched against the published IP ranges each vendor lists. User agent strings alone prove nothing, since anyone can send one.

Do these answers change?

Yes, and that is why the crawler names in this chapter carry a date. Each vendor adds and renames agents on its own schedule; re-read the source documents before acting on a list older than a few months.