To know if ChatGPT is citing your website, read your raw server access logs for requests from ChatGPT-User — the agent ChatGPT sends when it fetches a page to answer a live question. A citation fetch is a timestamped, verified event in your log; it does not appear in Google Analytics, which is why most sites never see it. Measured this way across three production sites over 75 days, IEO Citation Tracker recorded 952 verified citation events — and the majority of pages an AI crawler visited were never cited even once.
If ChatGPT cites your page, it first has to fetch it — and that fetch lands in your server log with a name on it. Analytics tools miss it because the fetch is made by a bot, not a browser running JavaScript. So the question "is ChatGPT citing me" is really "does a citation-class agent appear in my logs, on which pages, how often." Here is how to read it, and where the answer stops.
1. Open your raw server access log.
Download it from your hosting control panel — the raw access log, not Google Analytics. The log is the server's own record of every request that reached your site, AI agents included. Analytics only records browsers that execute its tracking script, and retrieval bots do not.
2. Find the AI retrieval agents.
Search the log for the user-agents that fetch pages to answer live questions: ChatGPT-User and OAI-SearchBot (OpenAI), PerplexityBot and Perplexity-User (Perplexity), ClaudeBot and Claude-User (Anthropic). A line from one of these is a page ChatGPT or another engine pulled while a person waited for an answer.
3. Separate a crawl from a citation.
Not every bot hit is a citation. A training crawler ingesting your whole site is not answering anyone; a citation-class agent fetching one page during a query is. And a user-agent string can be spoofed, so verify the request's IP against the agent's published range before you count it. That check is what separates a real citation fetch from a scanner wearing ChatGPT's name.
4. Count events, not prompts.
Each verified fetch by a citation-class agent is one timestamped citation event. Count events over a fixed window rather than running prompts and guessing. This is the difference between measuring what happened and sampling what might have.
Two facts only appear once you read the log this way. First, the volume is real and countable: across three unrelated production sites — a local-service business, a consumer app, and this methodology site — the instrument recorded 952 verified citation events in 75 days, each a request an identified agent actually made.
Second, and more useful: most pages an AI crawler visits are never cited. On these sites, the large majority of pages that bots crawled were never once fetched by a citation-class agent. Being crawled is not being cited — and only the log shows you which of your pages crossed that line and which did not. Method, denominators and corrections.
Reading your logs tells you, with certainty, that a citation-class agent fetched a specific page at a specific time. It does not by itself prove the answer displayed your link to the user — for Google AI features specifically, that appearance shows in Search Console's Generative AI report, which sees citations served from cache that never re-fetch your server. The complete picture is the log (all engines, precise, live fetches) crossed with Search Console (Google, including cache-served) and a captured answer (which engines actually showed you). No single source is the whole count; the log is the one you can verify yourself.
Is ChatGPT citing your website? Open your access log, find ChatGPT-User, verify the IP, and count the events. If it is there, you will see it — timestamped, on the page it pulled, in a record your analytics never kept. If it is not, no dashboard will invent it for you. IEO Citation Tracker automates exactly this read across your own logs.