AI citation tracking: how to know if ChatGPT and Perplexity are talking about you
AI citation tracking means checking whether your brand shows up in ChatGPT, Perplexity, Claude or Google AI Overviews answers for queries related to your business. Unlike Google rank tracking, there is no stable index to query: every answer is generated on the fly, and can vary from one search to the next.
I used to run an automated scanner on this page. I removed it: it pointed at a service that was not responding, and even when working, an automated tool cannot guarantee what a real user actually sees inside ChatGPT. This page explains why, and how to do this tracking yourself, properly.
Why this is technically hard
A Google rank tracker queries a search index. The result is reproducible: the same query, at the same time, from the same location, returns the same page in position 3. A generative AI engine like ChatGPT or Perplexity does not work that way.
Every answer is produced by a probabilistic model. The exact wording of the question changes the result. Conversation history changes the result. Detected location, the model version live that day, and even plain sampling randomness change the result. Two people asking the same question ten minutes apart can receive two answers citing different sources.
Most vendors selling AI tracking tools stay quiet about this, because it complicates selling a monthly subscription. An automated scan gives a snapshot at one point in time, with real uncertainty attached. That is not useless, but it is not the same certainty as a Google ranking.
Google rank tracking versus AI citation tracking
| Criterion | Google rank tracker | AI citation tracking |
|---|---|---|
| Source queried | Fixed search index | Answer generated on every call |
| Reproducibility | High, same query gives same rank | Variable, depends on session and wording |
| Position measured | Exact numeric rank (1, 2, 3…) | Presence or absence in the answer, order of mention |
| Reliable automation | Yes, standard for years | Partial at best, sampling required |
| Most reliable method today | Automated tool (Search Console, SEMrush…) | Documented manual testing, repeated over time |
How to do it yourself, step by step
1. List 10 to 15 realistic queries
Not just your brand name alone: questions a real prospect would type. Mix generic queries (“best [your field] in [city]”) with specific queries tied to your offer. Use the wording your actual clients use, not internal jargon.
2. Choose which engines to cover
At minimum ChatGPT and Google AI Overviews, which reach the widest audience. Add Perplexity, built specifically around citing sources. Add Claude if your target is B2B or technical.
3. Ask each query from a clean session
Open a new conversation each time, with no prior history that could bias the answer. Note the exact date and time of the test.
4. Capture the answer with dated proof
A full screenshot, not a plain text copy: it proves the date, the engine used, and the interface at the time of the test. Note whether your brand appears, at what position within the answer, and whether a link to your site is included.
5. Repeat the test at a regular interval
A single measurement does not say much given how much answers vary. It is the trend across several passes, spaced a few weeks apart, that shows whether your AI visibility is improving.
6. Log results in a simple spreadsheet
Columns: date, engine, query, brand cited (yes/no), position within the answer, link included (yes/no), link to the screenshot. A basic spreadsheet is enough.
What I actually do for my clients
I do not sell an automated citation tracking tool, for the reasons above. I run documented manual tracking: I test relevant queries across several AI engines and capture each result with dated proof.
On the GolfinThai project, I manually tested brand presence across six AI engines (ChatGPT, Perplexity, Claude, Gemini, Copilot and Google AI), with screenshots dated February and then April 2026 to document progress over time. That method, not a subscription to an automated scanning tool, is what allowed me to track and prove the change.
FAQ: AI citation tracking
Several commercial tools promise this kind of automated tracking, but none can guarantee an exact match with what a real user sees, because ChatGPT, Perplexity and other AI engines do not return the same answer to two people asking the same question. An automated scan gives a sample, not an absolute truth. I would rather say that plainly than sell a precision that does not exist.
Language models generate a probabilistic response on every call, rather than looking up a fixed index. The exact wording of the question, prior conversation history, detected location and the model version live in production all influence the output. Two identical searches, minutes apart, can cite different sources.
For a set of 10 to 15 queries tested across 3 or 4 engines, expect roughly an hour for the first pass: asking the questions, capturing the answers, noting whether the brand appears and where in the response. Later passes go faster once the query list is stable.
ChatGPT and Google AI Overviews reach the widest audience. Perplexity is built specifically around citing sources, so it is particularly relevant if the goal is to get cited. Claude skews toward a technical, B2B audience. I recommend tracking these four first, and adding Gemini if your audience is more consumer-facing.
No. Google ranking and generative AI citation are two different mechanisms. A page on the first page of Google can be ignored by ChatGPT, and conversely a page with modest Google visibility can be cited by Perplexity if it answers the question directly and clearly.
No, they measure different things. Citation tracking measures an outcome: are you cited, on which queries, by which engines. A GEO audit looks at the technical and editorial causes behind that outcome, such as Schema.org markup or AI bot accessibility. The two complement each other.
Want me to run this scan for you?
I can run this manual tracking on your priority queries and send you the documented results, screenshots included. No subscription, no promise of automation: human work, honest about its limits.
Request a manual scan