Citation check
A ChatGPT citation check that never opens ChatGPT
A ChatGPT citation check for one public URL: a 0–100 score and about ten fixes. On-page signals plus an optional LLM review of the HTML — not a live ChatGPT login.
3 October 2026 · 8 min read
People search “ChatGPT citation check” when a buyer’s answer named a competitor and skipped the product. The honest version of that check is narrow. It asks whether the public page gives a model a definition, a name, and a fact it can lift — and whether the bots that fetch pages for training are blocked. It does not ask ChatGPT a prompt and screenshot the reply.
CiteScore is that check, sold as a $39 GEO audit of one URL. You get a 0–100 score, a per-engine snapshot that includes a ChatGPT line, and about ten fixes in a PDF. When a language-model key is configured, the model rewrites the summary and the fix list from the HTML we fetched. That review is optional infrastructure, disclosed on Privacy. It is not a ChatGPT account with your URL typed into the box.
What the ChatGPT row is allowed to say
The snapshot status is one of four words: Cited, Mentioned, Absent, or Competitor cited. For ChatGPT, the baseline is conservative. Poor AI-crawler access lands on Absent. A page with a definitional lead and something quotable — a number, a year, a named method — can move to Mentioned. We do not mint Cited because the homepage says the product is “loved by teams.” Cited is for when the HTML itself shows a third-party citation. The language-model pass may adjust wording and scores within a tight band. It is instructed not to invent a live citation.
The fictional Northbound sample shows the shape: ChatGPT “Mentioned,” with a sentence that the product shows up as a category example and almost never as the source. Northbound is not a customer. The format is the one you buy.
The signals behind that row
The score is weighted across eight dimensions. Three of them do most of the work for the ChatGPT line:
- Answer fitness. A self-contained “X is a Y that helps Z” in the first screen, plus question-shaped headings. A slogan under eight words with no period is treated as a weak lead.
- Citable evidence. Dates, counts, and research language. A comparison or alternatives URL on the same host helps. Off-site mention share is not crawled — the “mention share” dimension only sees comparison language on the site you submitted.
- AI-crawler access. We fetch
robots.txtand record allow, disallow, or unspecified for GPTBot, PerplexityBot, and Google-Extended. We also fetch/llms.txt. Detail lives on the crawler page and the llms.txt page.
Entity markup, FAQPage JSON-LD, author attribution, and visible dates still move the overall 0–100. They matter more for the AI Overviews and Perplexity rows than they do for the ChatGPT threshold, which is why a single score can look “fine” while one engine line says Absent.
What a citation check cannot see
ChatGPT answers move with the prompt, the memory on that account, whether browsing is on, and the day the model was willing to fetch. A check of your homepage cannot reproduce that. If you need “for these 40 prompts, did ChatGPT mention us this week?”, that is a monitoring product. Otterly and Profound sell that job. CiteScore sells the prior job: make the page worth mentioning before you pay to watch prompts.
We also do not parse every OpenAI user-agent. GPTBot is the token on the crawler dimension. ChatGPT-User and OAI-SearchBot — the agents OpenAI documents for user-triggered browsing and search — are not read. Allowing those two while disallowing GPTBot is a real configuration. Our ChatGPT row will still treat the GPTBot disallow as a negative crawler signal. If that disallow is deliberate (you opted out of training crawls), read the crawler dimension with that policy in mind.
Checklist, then the paid check
The citation checklist is the unpaid version: twelve homepage checks, including the definition, the FAQ, the dated fact, and the robots.txt own-goal. Use it if you want to edit before you buy. Buy the check when you want those checks scored against the URL you actually ship, with the fixes ordered and written down.
Same product as the GEO explainer describes. The URL is different because the search is different. One charge, one PDF, no account.
Questions the check gets
Do you open ChatGPT and search my brand?
No. The ChatGPT row is inferred from on-page signals — definitions, entities, evidence, schema, freshness, and crawler access — plus an optional language-model review of the public HTML we fetched. The report says so.
Can the status say Cited?
The status set is Cited, Mentioned, Absent, or Competitor cited. We prefer Mentioned or Absent. Cited is reserved for when the HTML itself shows a third-party citation. Marketing copy alone does not count.
How is this different from the citation checklist?
The checklist is twelve things you can inspect yourself. This check is the scored PDF: a 0–100 CiteScore, a ChatGPT snapshot, and about ten ranked fixes for the URL you paid for.