How to track brand mentions in ChatGPT
Three ways to track whether ChatGPT mentions your brand, from manual checks to daily API runs, and what each one costs you in time and accuracy.
- Asking ChatGPT once and screenshotting the answer measures nothing: the same prompt gives different answers on different runs.
- For each prompt record five things: mention yes or no, position in the list, tone, cited sources, and the date.
- Manual checks work for a handful of prompts, a spreadsheet routine for a few weeks, and past about 15 prompts you want scheduled runs through the API.
- Daily runs turn a noisy answer stream into a number you can defend, with “based on N answers” next to it.
Asking once tells you nothing
One ChatGPT answer is a sample of one from a system that varies run to run. Screenshot it and you have documented a moment, not a pattern.
Ask the same question in the morning and at night and the wording, the order of the recommendations, even the set of brands can shift. That is not a bug you can wait out; it is how these models work. The practical consequence: any claim about your visibility needs a count behind it, like “mentioned in 9 of 30 answers this month”, never “ChatGPT recommends us”.
This is the reason tracking exists. You are not watching one answer, you are building a small sample per prompt and reading the average.
What to record for every prompt
Five fields per answer are enough to answer every question a manager will ask: mention, position, tone, cited sources, and when it was run.
| Field | What it captures | Why it matters |
|---|---|---|
| Mention | Was your brand named, yes or no | The raw input to Visibility |
| Position | Where in the list you appear, 1 being the top | Distinguishes “mentioned” from “recommended first” |
| Tone | How the answer talks about you, on a 0 to 100 scale where 50 is neutral | Catches praise that slid into apology language |
| Sources | The domains the answer cited | Tells you which pages to influence next |
| Date and engine | When and where the answer was collected | Turns screenshots into a series you can chart |
Keep the original text of every answer as well. Parsers improve, categories change, and the ability to re-analyse history is what turns a reporting tool into evidence you can revisit.
Three ways to track mentions
Manual checks, a spreadsheet routine, or scheduled runs through the API. Pick by prompt count, not by ambition.
- Manual checks. Open ChatGPT, ask your questions, note the results. Zero setup, zero cost, and it dies the first busy week. Defensible for up to about five prompts and one engine.
- Spreadsheet with a recurring reminder. Same questions every Monday, results pasted into rows, counts computed by hand. Works for a few weeks and one person; breaks when a second person touches it or when you add engines.
- Scheduled runs through the provider APIs. A script or tool asks each prompt on each engine daily, stores the answers and parses the fields above. This is the only version that produces sample sizes instead of anecdotes, and the only one that survives staff changes.
Option three sounds expensive and usually is not: runs call the official APIs, and API tokens are billed by your provider at their own prices. In advisible every run uses your workspace's own key (bring your own key), so a daily habit costs what the tokens cost and nothing on top.
How often to run your prompts
Daily for prompts you actively manage, weekly for the long tail. Daily is what makes month-over-month numbers comparable, because it gives every prompt the same number of chances to appear.
Run on demand too. When you publish a page or fix a claim, a manual run right after tells you whether the next answers picked it up, without waiting for tomorrow's schedule.
What good tracking looks like
Every number declares its sample size, every engine is broken out instead of blended, and the raw answers are one click away from the chart that summarises them.
- Visibility and Share of voice shown with “based on N answers” under each figure.
- Per-engine views, because a blended average hides “we are invisible in Claude”.
- An answer archive, so a skeptical stakeholder can read the text behind a metric.
- Estimates marked est. and demo figures marked Sample data, so nothing masquerades as measured.
That is the shape advisible ships: daily runs on ChatGPT, Perplexity, Gemini, Grok and Claude, the five fields above parsed out of every answer, and an overview that will not show a number it cannot count. The setup wizard suggests prompts to start from if your list is blank. Next, choose the prompts carefully: which prompts to track.
Common questions
Can I track ChatGPT brand mentions for free?
Yes. Ask your prompt list once a week and log the results in a spreadsheet. It is honest data and it scales badly: past roughly 15 prompts the manual routine quietly stops happening.
How many prompts should I track?
Start with 15 to 25 questions your buyers actually ask: category, comparison, best-for and a few branded ones. Track branded prompts separately so they do not inflate your visibility.
Does ChatGPT show citations for its answers?
In search mode it can link the sources it used; in plain chat it usually shows none. Perplexity always cites, so citation metrics are strongest there and weakest in default ChatGPT and Claude chats.
Should I track other engines too?
If your buyers use them, yes. ChatGPT, Perplexity, Gemini, Grok and Claude give materially different answers for the same prompt, and a brand strong on one can be absent from the others.
What is AI search visibility?
AI search visibility is how often your brand shows up when buyers ask ChatGPT, Perplexity or Gemini about your category. What it means and how to measure it.
Which prompts to track for AI visibility
How to build an AI visibility prompt set from real buyer questions, how many prompts to start with, and why branded prompts belong on their own line.
How many answers before you can trust a number?
What a sample line like based on 18 answers licenses you to say, our rule of thumb for ignoring noise, and how a confidence interval reads without the math.
Track this for your own brand
advisible runs your prompts on ChatGPT, Perplexity, Gemini, Grok and Claude every day, and every number shows how many answers it is based on.