How to track brand mentions in ChatGPT

You can’t search ChatGPT the way you search Twitter. There is no feed of everyone’s conversations, and OpenAI doesn’t tell brands when they are mentioned. Tracking brand mentions in ChatGPT means asking it the questions your buyers ask, the same way every time, and recording what it says.

This guide gives you two ways to do that. The manual method costs nothing and takes about an hour a week for a small prompt set. The automated method does the same thing on a schedule. Both use the same logic, so you can start by hand and switch later without losing comparability.

What you are tracking

For each answer, you want four facts:

• Mentioned: does the answer name your brand, under any of its names?
• Competitors: which of the competitors you track are named?
• Citations: which sources does ChatGPT attribute the answer to, and is any of them on your domain?
• Description: how does the answer describe you? Old pricing, wrong audience and missing features all show up here.

Counted over many answers, these become a mention rate, an owned citation rate and a tracked share of voice.

The manual method, step by step

1. 1

   Pick 10 to 20 prompts.

   Use questions a buyer would type before they know your name: "best product analytics tool for a seed-stage startup", "Mixpanel alternatives for a small team", "PostHog vs Amplitude for B2B SaaS". Our guide on choosing prompts explains how to pick them. Write them in a sheet and never edit the wording casually.

2. 2

   List your brand names and competitors.

   For your brand, write every alias: product name, company name, domain and common spellings. Add three to five competitors with their aliases too. "Acme", "Acme Analytics" and "acme.io" should all count as one brand.

3. 3

   Open a temporary chat.

   In ChatGPT, start a new chat and turn on temporary chat. Temporary chats don’t use or create memories and don’t appear in your history, so your past research doesn’t color the answer. Use a new temporary chat for every prompt.

4. 4

   Paste the prompt exactly.

   Copy it from your sheet. Small wording changes produce different answers, so consistency matters more than phrasing.

5. 5

   Wait for the full answer and open the sources.

   If ChatGPT searched the web, it shows source links. Open the sources list and copy each URL.

6. 6

   Record one row per answer.

   Date, prompt, mentioned yes or no, competitors named, cited URLs, whether any cited URL is on your domain, and a short note on how you were described.

7. 7

   Mark failures separately.

   If ChatGPT errored, asked you to log in or hit a limit, write "failed" and leave that row out of your rates. A failed check is not a missed mention.

8. 8

   Repeat weekly.

   Same prompts, same day of the week if you can, same setup. If you reword a prompt, treat it as a new prompt with a new baseline.

A sample tracking table

Here is what one week looks like for a fictional product, Acme Analytics, tracked against Mixpanel, PostHog and Amplitude.

From this week: 5 answers captured (the failed one is excluded), Acme mentioned in 3, so the mention rate is 3 ÷ 5 = 60%. Acme’s own domain was cited in 2 of 5, an owned citation rate of 40%. The two answers without Acme are gaps: PostHog appears in both, and their cited sources show where to look next.

Reading the results

• Look at rates over weeks, not single answers. One answer can change on the next run. Five weeks of the same 20 prompts tells you something.
• Group by prompt type. You might be named in every "simple analytics" prompt and none of the comparison prompts. That points to the pages you are missing.
• Count cited domains. Across all answers, which domains come up most? Those sources describe your category. If you are absent from the top five, start there.
• Watch the description. A wrong fact repeated across answers usually traces back to one outdated page.

Where the manual method breaks down

The spreadsheet works well for a first baseline. It gets hard when:

• You want more than one assistant. Twenty prompts across ChatGPT, Claude, Perplexity and Google AI Overviews is 80 answers per round.
• You want daily data. More answers smooth out run-to-run variation.
• You lose track of which wording was used when, and trends start comparing different prompts.
• Copying sources by hand becomes the slowest part of the job.

The automated method with AEOmeter

AEOmeter runs the same loop for you. You set it up once:

1. 1

   Add your brand and competitors.

   Your brand, its aliases and owned domains, then the competitors you want to compare against.

2. 2

   Add prompts.

   Start from the suggested prompts, paste your own, or import a CSV or text list with a preview before saving. Tag them by type. See prompt tracking.

3. 3

   Connect the Chrome extension.

   It runs checks in background tabs of your own Chrome, signed into your own ChatGPT account, with temporary chat turned on where it can.

4. 4

   Set a daily schedule.

   Checks are queued at the time you choose and run when your Chrome is open and connected.

Each answer is saved with its cited sources. Mentions are counted with exact, case-insensitive alias matching, so the counts are repeatable. Failed and blocked checks are shown and kept out of the rates. Editing a prompt starts a new baseline, so trends compare like with like. The ChatGPT engine page covers ChatGPT specifics, and citation tracking shows how sources are grouped by domain and page.

One trade-off to know up front: collection happens in your browser, so your Chrome has to be open for scheduled checks to run.

FAQ