What “ChatGPT brand monitoring” should mean
It should mean a dated record of what a defined set of buying questions produced: whether the brand was mentioned, whether it was recommended, where it first appeared, which competitors were in the same answer, and which sources the answer cited. It should not mean a single “rank” number with no answer behind it.
An assistant does not expose a stable leaderboard across users. The same prompt can produce a different order, a different source list and a different recommendation on the next run. Monitoring that ignores the variance is monitoring noise.
What to monitor
- Mention rate: the share of sampled answers where the brand appears at all.
- Recommendation rate: the share where it is put forward as a choice, not just listed.
- First-mention position: where it appears relative to named competitors.
- Citation mix: the domains and page types the answer used.
- Variance: how much the answer changed between runs.
Five numbers, each with a date and a raw answer behind it. If a number cannot be traced to an answer, it does not belong in the report.
How to set up a useful baseline
- Pick ten buying questions, including best-of, alternatives, how-to-choose and pricing.
- Name three competitors you actually lose to.
- Log out. Use one location and one language; write them down.
- Run each question three times over several days, not back-to-back.
- Keep the raw answer, the citation list and the date.
- Re-run the identical set at day 60. Do not add questions to make the trend look better.
How often to monitor
Weekly is enough for a category that moves; monthly is enough for a stable one. Daily sampling produces more noise, not more signal, unless you are watching a launch window. The useful cadence is the one you can keep for a quarter.
Do not treat a one-day drop as a crisis. Compare the same question set across runs, and look for a change that lines up with a change in the inputs. A change in the answer without a change in the inputs is noise.
Common mistakes
- Running one sample and calling it a rank.
- Using a logged-in session with memory and personalisation.
- Tracking only the brand name and not the recommendation sentence.
- Ignoring citations, which are usually the most actionable part.
- Changing the question set between months, which makes the trend unreadable.
- Promising a citation or a position. No one controls the model’s answer.
Tracker or diagnostic?
A tracker gives you a recurring number and a dashboard; it is useful after you know which questions matter. A diagnostic gives you a baseline, a source audit and a plan. Most teams need the diagnostic first, because a trend without a baseline is just a screenshot with a subscription.
The free check on this site is three questions and one run. The diagnostic is ten questions, thirty dated answers, three competitors, a source audit and a day-60 re-run. See the method and AI search metrics.
Questions
Can I monitor ChatGPT without an account?
You can sample logged-out sessions, which is what the diagnostic does. A logged-in session adds memory and personalisation, so it is a different surface and should be labelled separately.
Is a daily number useful?
Only if you keep the raw answers and the dates. A daily “rank” without variance hides the thing that matters: the answer is not stable.
What should I do with the data?
Start with the questions where a competitor appears and you do not. Read the cited sources and fix the page that is actually doing the work. Re-run the same questions at day 60.
Run the free three-question check · See the $850 diagnostic · Rank tracking vs AI answers