Short answer: there is no official dashboard, so you check manually — ask ChatGPT questions your buyers would ask, see whether it names you, and whether it links your pages as sources. Do it across a fixed set of prompts, record the results, and repeat monthly so you can see the trend rather than a snapshot.
Here is the exact process, including the two traps that make most people’s results meaningless.
Why you cannot just look it up
Google Search Console reports impressions and clicks from Google Search. It reports nothing about ChatGPT. OpenAI publishes no equivalent tool for site owners. So the only way to know whether ChatGPT cites you is to query it the way your customers do and record what comes back.
That makes this a sampling exercise, not a lookup. Which means method matters — the difference between a useful measurement and a misleading one is entirely in how you set it up.
Step 1: Write a prompt set, not a question
Asking ChatGPT “do you know [my brand]?” tells you almost nothing useful. It tests recall of your name, not whether you get recommended to someone with a buying problem.
Instead write 15–30 prompts a real buyer would type, across four types:
- Recommendation: “best project management tool for a five-person design agency”
- Comparison: “Asana vs Monday for small teams”
- Problem: “how do I stop losing track of client feedback”
- Brand: “is [your brand] any good?” and “[competitor] alternatives”
Write them once and keep them fixed. Changing the prompts between checks makes the results incomparable, which is the most common mistake we see.
Step 2: Query in a clean session
This is the trap that invalidates most self-checks. ChatGPT personalises answers using memory and conversation history. If you have discussed your own company in that account, it is far more likely to name you — and you will conclude you are doing great when you are invisible to everyone else.
Before checking: start a brand-new chat, turn off memory or use a temporary chat, and ideally log out or use a separate account. You want the answer a stranger would get.
Step 3: Record three things per prompt
For each prompt, note:
- Mention — were you named at all? Yes or no.
- Citation — was one of your pages linked as a source? This is where referral traffic comes from, and it is separate from being mentioned.
- Framing and competitors — were you the lead recommendation, one option among several, or mentioned with a caveat? Which brands appeared alongside you?
A simple spreadsheet with one row per prompt is enough. Your mention rate is the percentage of prompts where you were named; your share of voice is your mentions relative to each competitor’s across the same set.
Step 4: Check the other engines too
ChatGPT is the largest assistant, but it is not where most of your search demand sits. Google AI Overviews appear directly on the searches you already compete for, so they usually matter more commercially. Run the same prompt set through Google AI Mode, Perplexity and Gemini.
Our free AI Visibility Checker runs a standardised prompt set across engines and reports mention rate and competitor share, and the AI Overview Checker tells you which of your keywords trigger an Overview and who it cites.
Step 5: Confirm with your analytics
Filter your analytics for referral traffic from chatgpt.com, perplexity.ai and gemini.google.com. Volumes are small for most sites, but these visitors arrive mid-decision and convert unusually well. Rising AI referrals are the clearest confirmation that citations are actually happening.
Why ChatGPT’s answers change between runs
If you run the same prompt twice and get different sources, nothing is broken. This is the single biggest reason people misread their own results, so it is worth understanding before you draw conclusions.
A 2026 review of 45 studies on generative engine optimization reported day-to-day source overlap for identical queries running at a Jaccard similarity of roughly 0.34 to 0.42. Roughly a third to two fifths of cited sources persist between runs. The same review found an audit in which 57.8% of repeated ChatGPT queries failed to trigger a web search at all — the answer came from the model’s existing knowledge rather than live retrieval, meaning no on-page change could have affected it.
Two adjustments follow. Run each prompt at least three times and record how often you appear, not whether you appeared once. And judge on a rolling three-month view, because a single month of movement is well within normal noise.
It also helps to know how the retrieval works. Per OpenAI’s own documentation, ChatGPT search rewrites your question into one or more targeted queries and sends them to third-party search providers. OpenAI’s stated requirement for publishers is direct: allow OAI-SearchBot to crawl your site, and make sure your host or CDN permits traffic from its published IP ranges. If either is blocked, no amount of content work will get you cited — see our guide to AI crawlers.
What to do if you are not cited
Work through the causes in order of how often they are the culprit:
- You are blocking the crawlers. Check robots.txt for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Blocked bots cannot cite you.
- Your answer is buried. Assistants extract passages, not pages. Put the direct answer in the first two sentences under a question-style heading.
- You are absent from the sources models trust. Assistants lean heavily on comparison articles, review platforms and forum threads. If every “best X” list omits you, so will the model.
- You do not rank. Retrieval-based answers pull from strong organic results, so traditional SEO is the supply line.
Each of these is covered in detail in our LLM SEO guide, and you can score any single page against all 20 citation-readiness checks with the free GEO Audit.
How often to re-check
Monthly is right for most brands, plus an extra check after any major model release. Engine behaviour can shift overnight, and a monthly cadence catches the change while you can still explain it. If you would rather not run it by hand, the ChatGPT Rank Tracker re-runs your prompt set weekly and charts the trend.
What the citation data says about your odds
Two figures worth knowing before you interpret your results.
Semrush’s analysis of 126 million AI search prompts found ChatGPT citing an average of 15 sources per response, against just 3 for Gemini. ChatGPT is comparatively generous with citation slots — which means failing to appear there is more informative than failing to appear on Gemini.
And Profound’s analysis of 680 million citations found Wikipedia taking 7.8% of ChatGPT citations and Reddit 1.8%, with 80.41% of ChatGPT citations going to .com domains. If a reference site or community thread dominates your category, being mentioned on it may be more achievable than displacing it.
Common questions
Can I see ChatGPT referral traffic in Google Analytics?
Partly. Clicks from ChatGPT typically arrive as referral traffic from chatgpt.com or chat.openai.com, so you can segment them. What you cannot see is the far larger number of occasions where you were named or cited and nobody clicked — which is exactly why the manual prompt method above is necessary.
Does being cited by ChatGPT send meaningful traffic?
Usually modest in absolute terms today, and growing fast. Treat citation primarily as a visibility and credibility metric rather than a traffic channel, and measure it as one — see what AI visibility is and how to measure it.
Why does ChatGPT name my competitor but not me?
Most often because they appear in more third-party sources than you do. Branded web mentions correlate with AI visibility roughly three times as strongly as raw backlink count. The fix is usually off-site — see how to get more brand mentions in AI search results.
Should I check other AI assistants too?
Yes. Cross-engine source overlap has been measured as low as a Jaccard similarity of 0.11 to 0.18, so being cited in ChatGPT tells you very little about Gemini or Perplexity. Use one prompt set and score it separately per engine.
How do I turn this into an ongoing programme?
Map your prompt set to the subtopics you intend to own, then score each one monthly. That artefact is a topical authority map, and it turns a one-off check into a plan.
