# How to Track Brand Mentions in ChatGPT: Manual and Tool Methods

> Track brand mentions in ChatGPT with a fixed prompt set: a manual step-by-step baseline, a log template, and daily tracking with a 3-day free trial.

- Author: Sheetal Shakshi, Founder's Office, Nurdd
- Published: 2026-11-04
- Publisher: Noma by Nurdd (https://nomahq.app)
- Original page: https://nomahq.app/blog/track-brand-mentions-in-chatgpt-manual-and-tool

You can track brand mentions in ChatGPT by running the same set of buyer questions in fresh sessions, repeating each several times, and recording whether your brand is named, how high up the answer it sits and what words come with it. Do it by hand once for a baseline, then use a tool to repeat it daily.

## Key takeaways

- Use a fixed set of 20 to 25 unbranded buyer prompts and never change the wording, or your trend line means nothing.
- Run every prompt several times in a fresh session. One run tells you very little, because ChatGPT answers vary.
- Log six fields per run: brand named (yes or no), position, sentiment wording, sources cited, rivals named and the date.
- Report mention rate as runs where you are named divided by total runs, and state both numbers.
- Treat the manual check as a baseline. Daily tracking is what shows whether your fixes worked.
- Add a referral check in your analytics for chatgpt.com visits, but keep it separate from mention data.

## What counts as a brand mention in ChatGPT?

A mention is your brand named in the text of an answer. A citation is your URL listed as a source. They are separate outcomes: ChatGPT can name you without linking to you, and it can link to your page without naming you. Track both, as [citation vs brand mention](https://nomahq.app/blog/citation-vs-brand-mention) explains, because each needs different work.

Ahrefs' study of about 37,500 AI citations, tracked daily for 20 days across five AI tools, found that brand mentions can remain in answers after website citations disappear ([Ahrefs, October 2026](https://ahrefs.com/blog/how-stable-are-ai-citations/)). Counting only links would have missed that.

Sources also appear only some of the time, because ChatGPT does not search the web for every prompt. A prompt that returns no links is not a failed run. Log it as a run with no sources, and keep it in your denominator.

## Why do ChatGPT answers change between runs?

ChatGPT generates each answer fresh, and it may or may not search the web before it does. The brand list, its order and the sources can all differ from one run to the next, even with identical wording. That is why a single check misleads.

SparkToro had 600 volunteers run 12 identical prompts through ChatGPT, Claude and Google's AI about 2,961 times in total. It found less than a 1-in-100 chance that ChatGPT or Google's AI would return the same list of brands in any two responses to one prompt ([MediaPost, as checked 3 November 2026](https://www.mediapost.com/publications/article/412364/ai-brand-recommendations-chaotic-inconsistent.html); also covered by [Search Engine Journal, as checked 3 November 2026](https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242/)). MediaPost reports that nearly every response to a repeated brand-recommendation prompt differed in its list, order or number of items.

The same coverage notes that the study did not collect data on how each brand was described or on sentiment. That is a gap your own log can fill: record the wording, not just the name.

Question type matters too. On ChatGPT and Google AI Mode, Ahrefs found that around 64 to 76% of cited pages changed from one day to the next. Questions naming a specific brand were the most stable, and "best X" questions were among the least stable ([Ahrefs, October 2026](https://ahrefs.com/blog/how-stable-are-ai-citations/)). So a factual prompt such as "what does Brand X cost" and a shortlist prompt such as "best tools for Y" will behave differently in your log. Keep them in separate groups.

The practical conclusion: you are measuring a rate, not a fact. "We were named in 12 of 60 runs" is a finding. "ChatGPT does not mention us" after one try is not.

## Why does a fixed prompt set keep results comparable?

If you change the questions, the wording or the sessions each time, you cannot tell whether a movement came from your brand or from your method. A fixed set holds everything constant except the answers. The academic paper that introduced generative engine optimisation ([arXiv, November 2023](https://arxiv.org/abs/2311.09735)) took the same approach: it defined visibility metrics and tested them on GEO-bench, a benchmark of diverse queries, so that changes could be compared.

Noma by Nurdd's [prompt tracking](https://nomahq.app/capabilities/prompt-tracking) is built on the same idea: a fixed set of prompts, put to every engine once a day from fresh sessions, so a change in the answer reflects a change in the world.

Choose prompts the way [chapter 3 of the AEO guide](https://nomahq.app/aeo-guide/choosing-prompts) describes, and see [which prompts to track](https://nomahq.app/blog/which-prompts-to-track) for the reasoning. The free [AI Prompt Finder](https://nomahq.app/tools/ai-prompt-finder) lists questions buyers ask about your category, tagged by intent, if you need a starting list.

## How do I track brand mentions in ChatGPT manually, step by step?

The manual method needs a spreadsheet and a few hours for a first baseline (a rough estimate, not a measured figure). It is the right first move because it shows you what the answers look like before you pay for anything.

1. **Write 20 to 25 unbranded buyer prompts.** Include category questions ("best [category] for [buyer type]"), comparison questions, problem questions and a few that name you directly. Unbranded prompts are the ones that test whether ChatGPT brings you up unprompted. Staying at 25 or fewer means the same list fits a 3-day trial later.
2. **Freeze the wording.** Save the list with an ID per prompt. Do not edit it later. If you must add prompts, add them as a new group and keep the old group's history separate.
3. **Use a fresh session for every run.** Open a new chat, with memory and personalisation off or from a clean logged-out browser profile, so earlier questions do not colour the answer. Use the same country and language settings each time.
4. **Run each prompt three to five times.** Paste the prompt, copy the full answer into your sheet and open the next fresh chat. Note whether ChatGPT showed sources, since it does not always.
5. **Log six fields per run.** See the template below.
6. **Calculate mention rate, average position and share of voice.** Mention rate is runs naming you divided by total runs. Share of voice is your mentions divided by all brand mentions across the set.
7. **Repeat on a schedule.** Re-run the same prompts monthly at minimum, and always after a site change, a launch or a new model release.

### The log template

| Field | What to record | Why it matters |
| --- | --- | --- |
| Brand named | Yes or no, plus the exact spelling used | Gives visibility. Misspellings and old names show up here. |
| Position | 1 for first brand named, 2 for second, and so on; blank if absent | Being named fifth in a list is a weaker result than first. |
| Wording | The phrase attached to your name, marked positive, neutral or negative | Shows how you are described, not just whether you appear. See [brand sentiment in AI answers](https://nomahq.app/blog/what-is-brand-sentiment-in-ai-answers). |
| Sources cited | Every URL or domain shown with the answer | Shows which pages ChatGPT leans on, including pages you do not own. |
| Rivals named | Each competitor and its position | Feeds share of voice and shows who gets recommended instead of you. |
| Date and run number | ISO date, run 1 to 5 | Lets you compare like with like over time. |

### A worked example with illustrative numbers

The numbers below are invented to show the arithmetic. They are not data from any real brand. Say you track 20 prompts, run each three times, and get 60 runs. You are named in 12.

- Mention rate: 12 / 60 = 20%.
- Across those 12 runs you appear 1st twice, 2nd four times and 3rd or lower six times: average position about 2.8.
- A rival is named in 33 runs: 55% mention rate.

With 60 runs, a rough margin of error on a 20% rate is about plus or minus 10 points (1.96 x the square root of 0.2 x 0.8 / 60). Treat the runs as roughly independent, which is generous. So if next month shows 26%, you cannot yet call that a gain. If it shows 45%, you can. Our [AI Visibility Measurement Standard](https://nomahq.app/resources/ai-visibility-measurement) sets out the same rule: call a change only when it exceeds the margin of error.

### Check the same prompts without the spreadsheet

For a quick first look, the free [AI Visibility Checker](https://nomahq.app/tools/ai-visibility-check) shows whether ChatGPT, Gemini, Claude and Perplexity recommend your brand, who they name instead and what to fix. It is a snapshot, so use it to see where to start, then build the fixed set above.

## Where does the manual method break down?

It works for a baseline and fails as a monitor, for four reasons.

- **Volume.** 20 prompts x 5 runs is 100 copy-and-paste jobs per round. Few teams repeat that weekly.
- **Drift in method.** Different people, accounts, locations and phrasing creep in, and each one changes the answers.
- **Slow signal.** Citations can vanish and return within days. Ahrefs found half or more of AI citations disappear by the next day ([Ahrefs, October 2026](https://ahrefs.com/blog/how-stable-are-ai-citations/)). A monthly check samples one day in thirty.
- **One engine.** Buyers also ask Gemini, Perplexity and Claude, and they often disagree. See [why engines disagree about your brand](https://nomahq.app/blog/why-engines-disagree).

## How do you monitor your brand in ChatGPT answers over time?

Automate the same loop: a fixed prompt set, fresh sessions, repeat runs, saved answers and trends. Pick the method by how often you need an answer and how many engines you need covered.

| Criterion | Manual spreadsheet | Daily tracking tool |
| --- | --- | --- |
| Best use | One-off baseline, proof of concept | Ongoing monitoring and reporting |
| Frequency | Monthly at best | Daily |
| Session hygiene | Depends on whoever runs it | Same method every run |
| Trend and margin of error | You calculate it | Calculated for you |
| Sources and rivals | Copied by hand | Listed per answer |
| Other engines | Each one repeated by hand | Covered in the same set |
| Cost | Staff time | Subscription |

## Is there a tool to track brand mentions in ChatGPT?

Yes, several vendors sell one. The test that matters is whether it holds the prompt set fixed, repeats runs, and shows the answer text and sources behind every number. [AEO tools: how to choose one](https://nomahq.app/blog/aeo-tools-how-to-choose) gives a buyer's checklist, and the [ChatGPT visibility tracker guide](https://nomahq.app/blog/chatgpt-visibility-tracker) covers what to measure and how often to run it.

Noma by Nurdd is an AEO (answer engine optimisation) and AI visibility platform. Its [ChatGPT Visibility Tracker](https://nomahq.app/features/chatgpt-visibility) asks ChatGPT your buyers' questions every day, reports visibility, position and sentiment, lists the sources each answer cites and shows which rivals are named instead. The same prompts also run across 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude. Noma launched in May 2026.

See it on your own brand: [book a demo](https://nomahq.app/demo) or [start the 3-day free trial](https://platform.nomahq.app/signup).

### Set up the tool method in five steps

1. **Start the trial and add your brand.** The 3-day free trial includes 25 tracked prompts. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged.
2. **Load your frozen prompt list.** Use the 20 to 25 prompts from your manual baseline so the two datasets line up. Add your brand's alternate spellings and your competitors.
3. **Let it run daily.** The set is asked every day, so you get a trend rather than a single reading.
4. **Read the answers, not only the score.** Open the answers for prompts where you are absent, note the cited sources, and compare against your manual log.
5. **Turn gaps into work.** Prompts where rivals appear and you do not are your backlog. Fix owned pages first, then the third-party sources engines cite. [How AI engines choose citations](https://nomahq.app/blog/how-ai-engines-choose-citations) explains the logic.

After the trial, plans start at $79 a month for Starter (50 tracked prompts, 1 project), as listed on [the Noma pricing page](https://nomahq.app/pricing). Check that page for current plans.

## How do I connect mentions to visits?

Mentions show influence inside the answer. Visits show clicks. They are different measures, so keep them in separate columns. In your analytics, filter traffic by the chatgpt.com referrer and note the landing pages. Many people read an answer and never click, so expect visits to be far lower than mentions. For the revenue link, see [AI search attribution](https://nomahq.app/blog/ai-search-attribution-guide).

## Common mistakes when tracking ChatGPT mentions

- **Only testing branded prompts.** Asking "what is [your brand]" tests recall, not recommendation.
- **Judging from one run.** Given the variation SparkToro measured, one answer is an anecdote.
- **Editing prompts between rounds.** It breaks comparability.
- **Mixing question types.** Brand-specific and "best X" prompts have different stability, so report them separately.
- **Ignoring crawler access.** OpenAI's bots documentation says OAI-SearchBot surfaces sites in ChatGPT search, and that sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links ([OpenAI, as checked 3 November 2026](https://developers.openai.com/api/docs/bots)). Run the free [AI Crawler Checker](https://nomahq.app/tools/ai-crawler-checker) to see whether OpenAI's bots can reach you, and read the [AI Crawler Policy Kit](https://nomahq.app/resources/ai-crawler-policy-kit) before changing training rules.
- **Reporting a single percentage.** State the numerator, the denominator and the date.

## How we researched this

We reviewed published research on answer variability (SparkToro's study as covered by MediaPost and Search Engine Journal, and Ahrefs' citation-stability study), OpenAI's crawler documentation, the academic paper that introduced generative engine optimisation, and Noma's own product pages and help content. Facts and prices were checked on 2026-11-03. The worked example uses invented numbers, labelled as such. We did not run original ChatGPT tests for this article, and we have not published customer results for Noma.

## What to do next

Write your 20 to 25 prompts today, run each three times in fresh sessions and fill in the log. Then decide whether you need daily tracking. To see your own numbers across 13 AI engines, [book a demo](https://nomahq.app/demo) or [start the 3-day free trial](https://platform.nomahq.app/signup).