Is ChatGPT citing your website? Five ways to find out, from free to proper
How to check whether ChatGPT cites your site: a free manual method, what it misses, using referral traffic and crawler logs as evidence, and when to move to a tracker.
A common claim is that ChatGPT never mentions a site, based on asking it twice. A wider set of questions can turn up the domain in answers nobody tried. The reverse happens too: a team celebrates one citation and builds a story on it.
So how do you actually find out? Here are five methods, ordered from free and rough to proper and repeatable.
Method 1: ask it directly, but ask widely
The free method is to open ChatGPT with search enabled, ask the questions your customers ask, and read the sources. It's a fine starting point if you do it with discipline.
Write down 20 questions first. Use category questions ("best option for X"), comparison questions ("A vs B"), and a few about your own brand. Run each one, and note whether your domain shows in the source list and whether the brand is named in the text.
Be aware of what this can't do. Answers vary between runs, accounts and countries. Twenty questions on one afternoon is a sample, not a measurement. Use it to form a hypothesis, nothing more.
Method 2: read the fan-outs
When ChatGPT searches, it often rewrites your question into several narrower searches. The pages it cites come from those, not from your original wording. If you can see the fan-outs, you can see why a page that ranks for your head term still isn't cited: it may not answer any of the sub-questions.
Our explainer on query fan-outs goes into this. Promptwatch lists query fan-outs per model, and its free GEO Inspector Chrome extension shows ChatGPT's fan-outs and the difference between fetched and cited sources, running locally in your browser.
Method 3: look for the referral traffic
A citation people click leaves a trace in analytics. Check your referral sources for chatgpt.com and similar. Some AI visits arrive without clear referrer data, so treat this as a floor, not a total. Our post on measuring traffic and conversions from AI search covers the caveats.
A zero here doesn't prove you aren't cited. People may read the answer and never click. But a steady stream of visits landing on a specific page is good evidence that the page is being surfaced.
Method 4: check whether ChatGPT's bots visit
Your server or CDN logs can show whether OpenAI's crawlers request your pages. There are three to know:
- GPTBot collects content that may be used in training.
- OAI-SearchBot supports ChatGPT's search index.
- ChatGPT-User fetches a page when a user's question triggers a live visit.
There is no bot called "ChatGPTBot", so if you see that string in a log, be suspicious of it.
Logs prove access, not citation. But if OAI-SearchBot never touches a page, that page is very unlikely to be cited, and you've found a fixable problem. A page that returns errors to these bots is the most common culprit we see. Verify bot IP addresses where you can, because user agent strings are easy to fake.
Method 5: use a tracker built for it
Once you move past a handful of questions, manual checking stops scaling. A tracker runs a fixed set of prompts daily, stores each answer and its sources, and gives you page-level and domain-level citation views.
Promptwatch is what we use. Its citation analytics shows which of your pages are cited, by which models, at what URL position, and for which prompts. The Page Tracker lets you follow a specific URL, so you can ask "is this exact page being cited?" and get a straight answer with the responses behind it. Agent Analytics adds IP-verified crawler logs and a citation rate per page, linking the crawl to the citation.
How to combine them
Each method catches something the others miss.
| Question | Best evidence |
|---|---|
| Is my domain ever cited? | Tracker citation view |
| Is this specific page cited? | Page Tracker |
| Why isn't it cited? | Fan-outs plus crawler logs |
| Does the citation bring visits? | Visitor analytics |
| Did something break? | Crawler error tracking |
If you want a free first pass, run Method 1 and Method 4 together. If ChatGPT never cites you and the bots never visit, fix access first. If the bots visit and you're still not cited, the content isn't answering the sub-questions, which is a content problem, not a technical one.
Traps to avoid
Logged-in personalization can skew a manual check. Using a different country changes sources. Asking the same question twice in one session can produce an answer shaped by the first. Whenever you can, run checks in a clean session and record the date.
Another trap is counting brand mentions as citations. ChatGPT may name you in its text without linking, and it may link you without naming you. They're related but separate, and a tracker that distinguishes them tells you more.
What we'd do on Monday
Run 20 manual questions and note the results. Pull a week of logs and look for OAI-SearchBot and ChatGPT-User requests to your key pages. Then set up Promptwatch with those same 20 questions as your starting panel, add the Page Tracker for your five most important URLs, and let it run for a month.
By the end of that month you'll have a real answer to the question, one that includes the pages, the prompts, the models and the trend, rather than the impression left by two questions asked on a quiet afternoon.