How to Get Your Website Cited by ChatGPT and Other AI Search Engines

We tested real buyer questions in ChatGPT, logged every source it cited, and read our own server logs to find out what it takes for a website to be quoted by AI search.

• To get cited by ChatGPT, let its search crawler read your site, then publish pages that answer one specific question each, lead with the answer and rest on data only you have
• AI companies run separate bots for training, search and live user requests, and blocking the wrong one can remove you from AI search
• In our test of 13 buyer questions, seven answers carried sources, 14 links in all, and none pointed to us
• The pages that were cited were primary sources with a clear answer, named sources and often a visible update date

To get your website cited by ChatGPT, let OpenAI's search crawler read your pages, then publish content that answers one specific question each, puts the answer first, and rests on first-hand data. The same approach applies to Perplexity, Claude and other AI search tools. None of the AI companies has published a ranking formula, so nobody can promise a citation. What you control is whether your page can be reached and whether it is worth quoting.

Instead of repeating what the SEO vendors say, we measured this on our own site. Some of the results are encouraging and some are not, including the fact that we were not cited once.

What is AI search optimisation, and what are AEO and GEO? #

AI search optimisation is the work of making your pages easy for AI tools to find, read and quote when someone asks them a question. You will see it called answer engine optimisation (AEO) or generative engine optimisation (GEO). The names differ, but the job is the same: when a person asks ChatGPT, Perplexity or Claude a question and the tool looks at the live web, your page should be one it can reach, understand and trust enough to cite.

It overlaps with ordinary SEO, since many AI tools search the web before they answer. But it is not identical, and the clearest difference is who is doing the reading. A search engine ranks your page. An AI tool reads it, takes what it needs, and may or may not show where it came from.

Exa says its new search engine for AI agents beats every rival on eight benchmarks Exa Labs, backed by a16z at a $2.2 billion valuation, unveiled a cheaper Exa Lite tier at $4 per 1,000 requests alongside the October 18 release, part of a broader push to undercut Perplexity on agent search costs. - AI search engine beats Google for agent queries - Exa 3.0 embedding model outperforms all rival benchmarks

Which AI crawlers visit your website, and what does each one do? #

The big AI companies run several bots, and they do different jobs. Mixing them up is the easiest way to lock yourself out of AI search by accident. According to the companies' own documentation, each one splits its bots into three kinds: a bot that collects data for training, a bot that builds a search index, and a bot that fetches a page when a person asks for it.

Two things stand out. First, training crawlers barely visited us, while the bots answering live questions dominated. Second, the user-initiated bots are not true crawlers, and the companies say so: OpenAI notes that robots.txt rules may not apply to ChatGPT-User, and Perplexity says the same of Perplexity-User. You can block training and still appear in search. The bot names can be faked, and we did not verify the IP ranges, so read the counts as the traffic that identifies itself that way.

How do you check that ChatGPT can crawl your website? #

Start with robots.txt. If it carries a blanket disallow for AI bots, or a rule you copied without reading, you may be blocking more than you meant to. OpenAI's crawler documentation says a change to robots.txt can take about 24 hours to be picked up, and that you can allow search while disallowing training.

Then check that robots.txt itself loads from the outside. We found ours was being answered by our CDN with a cached redirect that pointed back to the same address, so every crawler asking for it got a loop and no file. The server behind it was serving the file correctly, which is why nothing looked wrong from inside. Test it from a network that is not your own, with and without a query string on the URL.

Finally, read your server logs. They tell you more than any plugin. Across 25 and 26 October, the ChatGPT-User agent made about 126,800 requests to 6,421 different URLs on our site, and 99.7% returned a normal page. Almost all of it landed on timely news posts. On 26 October, 877 of its 65,080 requests (1.3%) went to our long-form guides, even though guides make up roughly 5% of what we publish.

Reachable is not the same as chosen. A page can be fetched all day and never appear as a source. Your logs show whether the door is open, not whether anyone is walking through it.

What does ChatGPT actually cite when people ask buyer questions? #

We wrote 48 questions that a founder or small business owner might really type, and ran the first 13 through a logged-in ChatGPT account one at a time, recording every source it cited.

Nvidia Is in Talks to Value Perplexity at More Than $30 Billion Perplexity's valuation has climbed from $20 billion to $23 billion to a proposed $30 billion-plus in under a year, pricing the AI search startup at roughly 40 times its $750 million revenue run rate. - nvidia invests in perplexity AI valuation deal - perplexity search startup funding more than 30 billion

• Seven of the 13 answers cited at least one source. Six cited nothing and answered from memory.
• Across all 13 answers there were 14 cited links, and none pointed to our site.
• In a separate run where we added "cite sources" to the question, one answer carried eight cited links among 20 links in total. Asking for sources changes the result.

We then fetched the 14 cited pages. Twelve loaded properly. Apart from one 54,000-word report, they ranged from 528 words to about 5,600. Three carried FAQ schema, three had a table, and six showed a visible modified date, including pages updated this month. There was no single format. What they shared was that each page was the primary source on its question: an organisation explaining its own product, a company documenting its own tool, a data team publishing its own quarterly numbers.

That is a small sample from one account on one day, and ChatGPT's answers vary between runs, so treat it as a pointer and not a rule. We stopped at 13 questions because the result was predictable: we had not built our content to be cited, and running the other 35 would only have confirmed it.

How do you get cited by ChatGPT, Perplexity and Claude? #

Nobody outside the AI companies knows the selection rules. SEO vendors argue that these tools favour passages that are easy to lift and match to the sub-questions they break a query into. That is an observation, not something any AI company has confirmed. Given that, treat what follows as hypotheses built from what the cited pages had in common.

• One question per page. A page built around a single, specific question beats a general overview. The cited pages were narrow.
• The answer in the first 100 words. Say the conclusion first, then support it.
• Something only you have. Your own numbers, your own test, a decision you made and what happened. A generic how-to competes with every large organisation that publishes on the subject, and loses.
• Named sources with links. Every claim that is not yours should point to where it came from.
• A visible date and a reason to trust it. Put an updated date on the page and actually update it.
• Plain structure. Question-style headings, a table where you are comparing things, and a short FAQ with schema. These are cheap, but the cited pages show they are not what earns the citation.

How do you track traffic and citations from AI search? #

Three signals, none of them complete. First, ChatGPT-User, OAI-SearchBot and PerplexityBot hits to the page in your server logs. Second, referrals in your analytics: links ChatGPT shows to users carry a utm_source=chatgpt.com parameter, which we saw on the cited links in our own test. Third, a fixed list of questions that you rerun on the same schedule and record the result, which is the only way to see a change over time.

Do not trust a single run. Independent testing of ChatGPT search has found numerous inaccurate citations, so a citation, or the lack of one, is a noisy measurement.

What we are doing about our own zero #

We are rebuilding our long-form guides around the list above, starting with narrow questions where we have first-hand data, and we will rerun the same questions afterwards. If the numbers move, we will say so here, and if they do not, we will say that too.

Frequently asked questions #

Does blocking GPTBot hide my website from ChatGPT?

Not by itself. OpenAI says GPTBot relates to training, while OAI-SearchBot controls search appearance. Blocking training and allowing search is a supported combination.

Can I pay to be cited by an AI search tool?

We found no sign of a paid route for ordinary websites, and the documentation we read describes eligibility in terms of crawler access, not payment.

How long until a new page can be cited?

OpenAI says robots.txt changes take about 24 hours to be picked up. No AI company has published how quickly a new page is indexed for search, so treat any specific promise with suspicion.

Updated 27 October 2026. The test described here was run on 27 October 2026 from a single ChatGPT account, and the log figures cover 25 and 26 October.

This article is posted in Guides News, check it out for more related stories.

Join the discussion #

Open in the community → Almost there. Sign in and your reply posts straight away.