How to Track Brand Mentions in ChatGPT and Other AI Assistants
To track brand mentions in AI assistants, write a fixed list of prompts your buyers would ask, run each one several times in ChatGPT, Gemini, Perplexity and the other engines your audience uses, and log every answer: whether your brand appears, its position, the sources cited and how it is described. You can do this by hand in a spreadsheet for a small audit, or with a tracking tool once you need volume, history and competitor comparison.
The hard part is not running prompts; it is running them in a way that produces numbers you can trust. This guide covers the manual method step by step, the sampling rules that make it reliable, exactly what to record, and the point at which a tool starts to pay for itself.
Why you need a method to track brand mentions in AI
Asking ChatGPT “what’s the best tool for X” once and taking a screenshot is the most common way teams check AI brand mentions, and the least reliable. Generated answers change from run to run. When SparkToro and Gumshoe had volunteers run 12 recommendation prompts 2,961 times on ChatGPT, Claude and Google’s AI features, the same list of brands came back less than 1% of the time (Search Engine Journal, January 2026). One screenshot shows one draw from a wide distribution.
The same study offers the way out: across many runs, leading brands appeared at fairly steady rates, between 55% and 77% for the top headphone brands. So the goal of tracking is a rate, not a snapshot. Every step below is designed to produce that rate consistently, so that this month’s number can be compared with last month’s. For the definitions behind the numbers, see the guide to AI visibility metrics and how to calculate them.
The manual method in seven steps
A manual audit is a good way to learn how engines talk about your category before you commit to a tool. Budget about two to four hours for a first pass with 15 prompts.
- Pick the engines. Start with the assistants your buyers use. For most B2B and consumer brands that means ChatGPT, Google AI Overviews and AI Mode, Gemini and Perplexity; add Copilot, Claude or others if your audience is there.
- Write 15 to 30 prompts. Mix category prompts (“best invoicing app for freelancers”), comparison prompts (“Brand A vs Brand B”), problem prompts (“how do I chase late invoices”) and a few branded prompts (“is Brand A any good”). Phrase them the way a person would type or speak them, not as keywords.
- Use a clean session. Log out, or use a mode that ignores history. In ChatGPT, Temporary Chat does not use or create memories, though it still follows your custom instructions if they are switched on (OpenAI Help), so turn those off for testing.
- Fix the location. Run every prompt from the same country, and note it. Google says AI Mode grounds its answers by location, and other assistants can also tailor results to where you are, so a VPN or travel can change what you see.
- Run each prompt several times. Three runs per prompt per engine is a bare minimum for a spot check; five or more is better. Open a fresh chat for every run so earlier answers do not influence later ones.
- Log every answer using the fields in the next section, including runs where you do not appear.
- Repeat on a schedule. Run the same prompts, engines and location again in two to four weeks. Without a second round you have a baseline, not tracking.
What to log for each AI answer
Log each run as one row. The fields below are the minimum needed to calculate mention rate, position, citations and sentiment later without going back to the engines.
| Field | Example | Why it matters |
|---|---|---|
| Date and time | 2026-11-03 10:40 | Engines update models and indexes; dates let you spot breaks |
| Engine and mode | ChatGPT, search on | Each engine is reported separately |
| Location and language | US, English | Answers vary by market |
| Prompt ID and exact text | P07 “best invoicing app for freelancers” | Small wording changes change answers |
| Run number | 3 of 5 | Needed to calculate rates |
| Brand mentioned (yes/no) | Yes | Base for mention rate |
| Position among brands | 2 of 5 | Shows prominence when named |
| Competitors named, in order | B, A, C, D, E | Base for share of voice and competitor gaps |
| Cited URLs | yourbrand.com/pricing, g2.com/… | Shows which pages and third parties feed the answer |
| Sentiment and key claim | Neutral, “cheaper but fewer integrations” | Catches wrong facts and weak framing |
| Full answer text | Pasted or saved | Lets you re-score later or audit a surprise result |
Merge brand variants before you count. “Brand A”, “BrandA app” and the name of your flagship product all belong to the same entity; the same goes for competitors. Count a brand once per answer even if the text repeats it.
Cited URLs are worth more attention than they usually get. They show which third-party sites, such as review platforms, forums and comparison articles, shape the answers in your category, and they are often the most direct route to change, since the pages engines cite tend to be the ones they rely on. The overview of how AI engines choose sources to cite explains why those pages win.
How many prompts and runs are enough
Sampling decides whether your numbers mean anything. The University of St. Gallen paper “Don’t Measure Once” argues that AI visibility should be measured as a distribution through repeated observation, not a single check (arXiv, April 2026). The table below translates that into practical sizes for a single engine, using the usual margin-of-error approximation for a rate near 40%.
| Prompts × runs | Answers per engine | Approx. 95% margin on a 40% mention rate | Good for |
|---|---|---|---|
| 10 × 3 | 30 | ±18 points | A first look only |
| 20 × 5 | 100 | ±10 points | Spotting large gaps between engines |
| 50 × 5 | 250 | ±6 points | Monthly tracking of one market |
| 100 × 5 | 500 | ±4 points | Detecting moderate change month to month |
Runs of the same prompt are correlated, so real uncertainty is somewhat higher than these figures. Two consequences follow. A manual audit of 30 answers can tell you that you are missing from an engine, but not that you improved by five points. And adding prompts usually helps more than adding runs, because it also covers more of the ways people actually phrase the question.
Timing matters as well. Spread runs across several days instead of firing them all in one sitting, and keep a log of engine changes. Dan Petrovic of DEJAN, for example, aggregates brand tracking weekly because daily results are too noisy (iPullRank recap, July 2025).
Manual tracking vs AI brand monitoring tools
Manual tracking is free and teaches you a lot, but it stops scaling once you track brand mentions in AI across several engines and markets: 30 prompts × 5 engines × 5 runs is 750 answers per round, each to be read and coded. Tools built as an LLM rank tracker or ChatGPT rank tracker automate the collection and scoring. The trade-offs look like this.
| Criterion | Manual spreadsheet | Tracking tool |
|---|---|---|
| Cost | Staff time only | Subscription, usually priced by prompts or answers |
| Sample size | Tens to low hundreds of answers per round | Hundreds to thousands, on a schedule |
| Consistency | Depends on the person and session | Same prompts, locations and settings every run |
| Engines covered | Whatever you have time to open | Several engines in one report |
| History | As good as your file discipline | Stored automatically, charted over time |
| Competitors and share of voice | Possible but slow to code | Calculated per prompt and engine |
| Citations | Copied by hand | Extracted and grouped by domain |
| Best for | First audits, small brands, validating a tool | Ongoing tracking, agencies, multi-market brands |
When you evaluate tools, ask how each one collects answers, how many runs it takes per prompt, whether you can set a location per prompt, and how it counts a mention. Rand Fishkin’s advice from the SparkToro study is a fair test: ask vendors to show their math. As one example, Asheard runs your prompts on a schedule across ten engines, including ChatGPT, Gemini, Claude, Perplexity, Grok, Copilot and Google AI Overviews, and logs mentions, citations, position and sentiment next to your competitors. The AI visibility tracking page describes the setup.
A sensible path for most teams is to do one manual round first, then move the same prompt set into a tool and compare a week of automated results against your hand-coded sample. If the two broadly agree, you can trust the automation.
Turning AI brand mentions into actions
Tracking is useful only if the log leads to changes. Review it with three questions:
- Where are we absent? List prompts and engines with a mention rate near zero. Read the answers: which brands appear instead, and which sources are cited for them?
- Where are we misdescribed? Collect every wrong price, missing feature or outdated claim, and trace it to the cited page. Fix your own pages first, then contact third-party sites that repeat the error.
- Which sources keep recurring? Domains cited across many prompts are the ones shaping your category. Being present and accurate on them is often worth more than another blog post on your own site.
For ChatGPT specifically, the follow-up guide on getting your brand recommended in ChatGPT covers what to change once your log shows where you stand.
Frequently asked questions
Can you track brand mentions in ChatGPT for free?
Yes. Run a fixed list of prompts in a Temporary Chat or logged-out session, repeat each several times, and record every answer in a spreadsheet: mention, position, competitors, cited URLs and sentiment. It costs only time. The drawback is sample size, since a few dozen answers can reveal big gaps but cannot reliably show small month-to-month changes.
Does ChatGPT give the same answer to everyone?
No. Answers vary between runs of the same prompt, between users with different memory and custom instructions, and between countries or cities. That is why tracking relies on repeated runs from a clean, consistent setup and reports how often a brand appears rather than a single answer.
How often should you check AI brand mentions?
For manual audits, every two to four weeks with the same prompts and settings. With a tool, weekly collection and monthly reporting is a sound default: weekly data catches sudden drops after engine updates, while monthly totals contain enough answers to separate real change from random variation. Check more often after launches or press coverage.
What is an LLM rank tracker?
An LLM rank tracker is a tool that runs your prompts in AI assistants such as ChatGPT, Gemini and Perplexity on a schedule and records whether your brand appears, where it appears and which sources are cited. Despite the name, the most useful output is not a rank but a mention rate across many runs, compared against competitors.