To track your brand in ChatGPT, Gemini and Perplexity, run a fixed set of buying-intent prompts against each engine on a schedule, and record three things: whether you are mentioned, where you rank among the recommendations, and which sources the engine cites. You can do this manually with a spreadsheet in about an hour a month, or automate it.
This guide covers the exact method, the four metrics that matter, and the mistakes that make spot-checks worthless. It draws on our own dataset: 6,678 real AI answers for 215 businesses across 417 cities, the same data behind our AI Visibility Benchmark 2026.
One distinction before we start: a one-time look tells you where you stand today, and if you have never done even that, begin with our quick guide to checking your AI visibility engine by engine. This article is about the next step: turning that one-time check into ongoing tracking, because a single snapshot goes stale the moment the engines update.
Why tracking matters: most brands are invisible and don't know it
When we ran buying-intent questions through the three major engines for 215 businesses, the results were not close:
Two things follow from those numbers. First, if you have never checked, the base rate says AI is probably recommending someone else to your customers right now. Second, the upside is real: when an engine does name a brand, it makes that brand the top recommendation more than half the time. Being tracked is the first step to being named.
One more finding worth knowing before you start: in our citation data, only 10.8% of the sources AI engines cited were the brand's own website. Another 22% pointed at Google, usually the brand's own listing, and the rest were third-party pages: directories, review sites, roundups and articles. Tracking shows you which third-party sources actually drive your answers, which is exactly where to focus your effort. (More on how engines pick brands in our data study on how AI engines choose brands.)
The four metrics that matter
| Metric | What it tells you |
|---|---|
| Mention rate | Of your tracked prompts, how many answers name your brand at all. This is your headline visibility number. |
| Rank | When you are named, where you sit in the list. #1 converts very differently from #6. |
| Description accuracy | What the engine actually says about you. Wrong hours, old pricing or a stale positioning line costs trust silently. |
| Citations | Which pages the engine used to build the answer. These are the sources you need to be present on. |
The manual method, step by step
1. Build a fixed prompt set
Write 8 to 12 questions a real customer would ask, and freeze them. Cover three types: best-in-category ("best jewellery store in Bangalore", "best local SEO software for small business"), comparisons ("BrandX vs BrandY"), and direct trust checks ("is BrandX reliable?"). If you serve multiple cities, pick your top 2 or 3; the combinations grow fast.
2. Ask each engine in a clean session
Use incognito or a logged-out session. Your own account carries memory and personalization, so it will show you a flattering, unrepresentative answer. Ask the same prompt 2 or 3 times per engine, because answers vary run to run; you are tracking the pattern, not one response.
3. Record it
A simple sheet is enough: prompt, engine, date, mentioned (yes/no), rank, what was said, cited sources. Thirty minutes of copy-paste per monthly run once your prompt set is frozen.
4. Repeat monthly, same prompts
The value is in the trend line. Engines retrain, retrieval sources change, and competitors publish. A brand that was invisible in June can be #2 in September, and the reverse happens too. Changing your prompts between runs destroys the comparison, so resist the urge to rewrite them.
5. Act on the citation gaps
The cited-sources column is your to-do list. If engines keep citing a directory you are not listed on, get listed. If a competitor owns a "best of" roundup that gets quoted, pitch yourself into it. This is the practical loop: track, find the sources, earn your place in them, measure again.
Where the manual method runs out
Honest limits: manual tracking works well for one brand, one or two cities, and three engines. It gets painful when you need geography (the same question asked from Delhi and Bangalore returns different answers), when you manage many locations, or when you want week-level trend data instead of monthly snapshots. The combinations multiply: 10 prompts × 3 engines × 5 cities × 3 runs is 450 answers to read every cycle.
That scale problem is what automated AI visibility tracking exists for: the same fixed-prompt methodology, run continuously across engines and locations, with mention rate, rank, sentiment and citations turned into a dashboard instead of a spreadsheet.
See what AI says about your brand right now
Run a free AI visibility check: your mention rate, how the engines describe you, and which sources they cite, in under a minute. Then track it continuously with BizLoc8.
Check my AI visibility freeFrequently Asked Questions
How often should I check my brand in ChatGPT and other AI engines?
Monthly at minimum. Engines update models and retrieval sources continuously, so answers drift. A monthly run of a fixed prompt set catches shifts while they are still actionable; weekly is better once AI search sends you meaningful traffic.
Why do AI answers about my brand change every time I ask?
Models are not fully deterministic, and live-retrieval engines pull different sources per run. That is why single spot-checks mislead. Run each prompt 2 or 3 times and track the pattern rather than any single answer.
Can I just ask ChatGPT about my brand from my own account?
Avoid it. Logged-in sessions carry memory and personalization, so the engine may mention you because it knows you, not because a customer would see it. Use a clean session or a neutral tracking tool.
What prompts should I use?
The questions a buying customer would ask: best-in-category, comparisons, and direct trust checks, across your main categories and top locations. 8 to 12 frozen prompts is enough to see the pattern.
Sources
BizLoc8 AI Visibility Benchmark 2026 (6,678 AI answers, 215 businesses, 417 cities, pulled 10 Sep 2026): bizloc8.ai/ai-visibility-benchmark-2026
How AI engines choose which brands to recommend (BizLoc8 data study): bizloc8.ai/blog/how-ai-engines-choose-brands
Google Search Central, AI features documentation: developers.google.com
Related Articles
