How to See If My Website Is Cited in ChatGPT, One Page at a Time
Find out whether ChatGPT cites your site by testing page by page: which URLs earn citations, which are fetched but ignored, and which the bots never reach.
"Is my website cited in ChatGPT?" is a question with the wrong grain. The honest answer for most sites is "some pages, for some questions, some of the time." A yes or no hides everything you could act on.
So ask the question one page at a time. For each page you care about, there are three possible states, and each one has a different fix.
The three states of a page
Never fetched. The AI crawlers have not visited the page, or they hit an error when they did. Nothing downstream can happen. No fetch means no citation, and no amount of rewriting helps.
Fetched but not cited. The bots read it, and ChatGPT chose other sources. Here the problem is content: the page does not answer the question as directly as the pages that won, or it covers the topic at the wrong depth.
Cited. The page earned a source slot. Now the questions are for how many prompts, how stable that is, and whether a competitor page sits beside yours.
Sorting your important pages into these three buckets is the whole exercise.
Step 1: pick the pages that matter
Ignore the full site. Pick 10 to 20 URLs: top product or service pages, your best guides, and the pages that make money. Add two or three competitor pages that you suspect ChatGPT likes better.
Step 2: write the prompts each page should win
For each page, write three to five questions it exists to answer. If you cannot write them, the page may not have a job. Run those prompts in ChatGPT with search on, note the cited URLs, and repeat on another day. One run is a sample, not a verdict.
Look at the shape of the citations too. Are they mostly one domain, a forum, a review site, a competitor? That tells you the kind of source ChatGPT trusts for this topic, which may not be a blog post at all.
Step 3: check crawl access
For pages in the "never cited" pile, check whether the bots can reach them. The ones to look for in server logs are OAI-SearchBot, GPTBot and ChatGPT-User, and to treat them as three separate signals instead of one. Note that there is no bot called "ChatGPTBot", so a log search for that string will return nothing and mislead you.
Common blockers: a robots.txt rule written years ago that blocks all unknown agents, a CDN rule that challenges bots, pages that render their content only through client-side scripts, and URLs that redirect in chains. Fix access first, then wait for a recrawl before judging content.
Step 4: compare fetched pages against cited pages
Once you have fetch data, compute a simple ratio for each page: of the times a bot fetched it, how often did a citation follow? Pages with a lot of fetches and no citations are your highest-value rewrite targets, because the access problem is already solved.
Doing this without a spreadsheet
By hand, this takes real effort. You need server log access, a prompt log, and a way to join them. Most teams do it once and let it lapse.
Promptwatch was built around exactly this join. Its Page Tracker lets you add specific URLs, yours or a competitor's, and see the prompts and responses where that page appears. Agent Analytics provides the crawler side: real-time, IP-verified logs for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and the rest, with a crawl-to-citation path and a citation rate per page. Error tracking comes with failure explanations, so a 403 is not just a number.
The 360 Insights view puts crawler logs, citation clicks and AI-referred visits side by side per page. That is the three-state table above, already built. Topic Insights adds where pages underdeliver for their topic, which feeds the rewrite step.
Plan notes that affect the answer
Check the plan before you promise a workflow. Crawler logs start on Professional ($245 a month); Essential ($95) has none. Essential does include 360 Insights and site audit, but without log data the "fetched but not cited" bucket stays empty. The free Explore plan tracks 10 prompts on ChatGPT only, which is enough to test your most important pages' prompts and see whether you appear at all.
Connecting logs depends on your hosting. Promptwatch lists Cloudflare (Logpush or an auto-deployed Worker), AWS CloudFront, Fastly, Vercel, Netlify, Akamai, Google Cloud CDN, custom HTTP and a WordPress plugin as sources.
What to do with each bucket
Never fetched pages. Fix robots.txt, CDN challenges and rendering. Then give the bots time. Recheck logs after a week.
Fetched, not cited. Open the cited competitor pages from your prompt log. Compare how they answer the question in the first screen. Rewrite the opening, add the concrete detail the winners have, and consider a refresh instead of a new page. Promptwatch's Content Refresh gives diff-style suggestions for existing pages, which keeps the URL and its history.
Cited pages. Protect them. Do not rewrite a page that is working. Add internal links to it from related pages, and watch it in Citation Trends so you notice decay before it becomes a loss.
The short version
Do not ask whether the site is cited. Ask which of your twenty pages are in which bucket, and what one change each bucket needs. If you want the data without building the pipeline, start with the Page Tracker and the crawler logs in Promptwatch, and connect your CDN first, since everything else depends on knowing whether the bots can read you.