Mention Clock

How to track brand mentions in ChatGPT (5 steps)

A fixed prompt list, a clean session, and a row for every answer. One reply is a sample. The repeated list is the measurement.

Built by StyleLine FZE in Sharjah. Email support@mentionclock.com. We reply within one business day.

ChatGPT does not publish a brand dashboard. You measure visibility by asking the questions a buyer would ask, writing down what the answer did, and repeating that list often enough to see a change. The five steps below are the manual method. The product that stores the same rows is the ChatGPT brand tracker. The short form of the five steps is also on that tracker. A source-linked vendor table is on best tools to track ChatGPT brand mentions.

Step 1: Freeze a prompt list

Write the questions buyers ask before they shortlist a vendor. Include a category question, a problem question, an alternatives question, a comparison, and a "best for a small team" variant. Twenty to thirty questions is a full manual list. Five is enough to learn the habit before you automate.

Freeze the wording. Week 4 has to be the same sentence as week 1, or a change in mention rate is just a change in the question. Good prompts sound like a buyer:

  • "What is the best [category] for [audience]?"
  • "Which tools do [job title] use for [job to be done]?"
  • "[Competitor] versus [competitor] for a team of 10?"

Avoid a list that only names your brand. You need both "do you know us?" and "who wins the category?" If you only ask the branded question, every answer can mention you and still hide a category miss.

Before you freeze a long list, run three description questions once so a wrong price or a thin description shows up before you commit a quarter to the wrong prompt. Then lock the shortlist.

Step 2: Use a clean session

Your own ChatGPT account has memory and custom instructions. That is not a typical buyer. Turn memory and custom instructions off for the account you use to test, or use a fresh workspace that has never been told your positioning.

Write down the browser profile, whether web search was on, and the model name in the notes column. When the default model changes, expect one week of movement in mention rate or position. Do not treat that week as a content win or a content failure until the next run uses the same settings.

A clean session still is not a private customer. It is a public prompt with no saved brief. If you paste your own homepage into the chat before the question, you have changed the test.

Step 3: Repeat the full list

The same prompt can name different brands on different days. Count how often you appear, not whether you appeared once. Run every prompt on the frozen list. Skipping the ones that looked bad last week makes the mention rate flattering and impossible to defend.

Paste one prompt, wait for the full answer, and log it before you open the next. Do not stop at the first paragraph if the brands are named later. Do not rerun a single prompt until it names you and then record only that reply.

Weekly is the default cadence for a manual list. During a launch, a tighter loop is worth the extra time. A random subset each week makes the series noisy.

Step 4: Record more than yes or no

For every prompt, write one row:

  • Date
  • Prompt, copied, not paraphrased
  • Engine, set to ChatGPT
  • Search on or off
  • Mentioned, yes or no
  • Position among the brands named, such as first, second, or third
  • Rivals named in the same answer
  • Sources cited when search is on
  • Notes, including a wrong price or a wrong product description

A mention with the wrong price hurts more than silence when sales repeats the error. Flag that in notes. Position is blank when you are absent. Do not write zero, because zero would look like a rank.

Download the Excel brand mention tracker or the CSV brand mention tracker. The Excel file has a Formulas tab for mention rate, average position, and share of mentions, and a Cited domains tab for the hosts that repeat in a week.

Step 5: Turn the rows into a mention rate

Mention rate is the share of rows that say yes. If 10 prompts name you on 4 of them, the mention rate is 40 percent for that run. Average position is the mean rank on the rows where you appear. Lower is better. Share of mentions is your brand names divided by all tracked brand names on that list, so a prompt that names four rivals counts them, not just a win or a loss.

Read four to eight weeks of the same list before you call a trend. One strong week after a press hit is not a strategy. Four flat weeks after a site rebuild might be. When rivals gain while you stay flat, read the cited sources. Those pages are the outreach list. If the cited reviews name competitors and not you, that is the work, not another prompt rewrite.

Worked example from a real run (28 Oct 2026)

On 28 Oct 2026 at 09:39 Dubai time Mention Clock ran 10 buyer prompts on the category project management software for small teams.

Engines: ChatGPT (model-4.1-mini-2025-04-14), The second chat engine (the second chat engine-sonnet-5-5), Perplexity (sonar). 30 of 30 API calls succeeded; total logged cost $0.70. The upcoming engine was not included (quota error during probe).

ChatGPT often returned answers with no URL citations in the logged payload. When citations appeared, clickup.com showed up often because roundup answers cited ClickUp pages and comparison sites.

MetricAsanaMonday.comClickUpTrelloNotion
Mention rate on ChatGPT100%90%90%90%40%
Mention rate on the second chat engine100%90%90%80%40%
Mention rate on Perplexity70%40%90%80%30%
Average position when named22.82.62.23.6
Top cited domain when namedtie: clickup.com, ones.com, rock.so, thedigitalprojectmanager.comthedigitalprojectmanager.comclickup.comclickup.comtie: ones.com, rock.so

Download the full run log (CSV) (30 rows: prompt, engine, model, timestamp, brands with position, cited domains, excerpt).

30 Oct 2026: removed float.com from row 1. The answer linked float.com in the text, and the stored citations for that response did not include it.

SHA-256 of this file: b826be1b1e9e2729c2bd1aeefa1ec8a301e483afcca1efae8f51210484fe8f82.

Use this pattern on your brand: swap the category, keep the prompt count stable, and watch whether your mention rate climbs after you improve pages that match cited domains.

How often to check

Weekly is enough for most brands. The manual list takes the time of a person pasting prompts, so a daily manual run usually collapses into a shorter, biased list. Keep the weekly habit until the series is stable.

Check more often when a launch, a pricing change, or a public incident can move the shortlist inside a week. On Mention Clock, the free plan re-runs 5 prompts every week. Starter is $29 a month for 25 prompts, still weekly. Pro is $79 a month and runs every 3 days. Those are the published cadences. A tighter loop is for a launch window, not for every category page.

Do not change the prompt on the same week you change the cadence. You will not know which change moved the rate.

When a tool is worth it

Stay on the sheet while the list is five prompts and one person can paste them every Monday. Move to a tool when any of these is true:

  • The list is long enough that a missed prompt changes the rate.
  • More than one engine has to stay on the same wording.
  • Someone other than the person who ran the chat needs the history.
  • You want an alert when a prompt that used to name you stops naming you.

The ChatGPT brand tracker stores the prompt, the mention, the position, the competitors, and the cited sources, and re-runs the free allowance every week. The best tools comparison is the place to read other vendors' public engines, refresh rates, and starting prices before you buy. Run one product on your real brand for a month. A vendor page is not a measurement of your category.

Cited domains from a paid report are the same columns as the sheet, collected without the paste step. Prioritize hosts that already appear and that name competitors. A tool does not write the page those hosts would need to cite.

What the manual method cannot see

A single session is not the buyer's session. Memory you forgot to turn off, a prompt you softened after a bad answer, and a model change you did not note will all move the row. The method also cannot see a chat that never cites a page. When search is off, the sources column stays empty and you only have the brand names.

Do not mix ChatGPT rows with the second chat engine or Perplexity rows and average them into one rank. Keep the engine column honest. The matching logs are The second chat engine brand mentions and Perplexity rank tracking. The upcoming engine is being added, so the upcoming engine row is still a hand log.

After the rate moves

If the rate is flat and the cited hosts are rivals, pick one URL you can edit and leave the prompt list alone. If the rate jumps for one week and the notes say the model name changed, wait for the next identical run before you brief a writer. If a prompt never names you and never names a source you could earn, it may be the wrong question. Replace it only at the start of a new series, and mark the break in the date column.

The publishing work that follows a miss is on how to get mentioned by ChatGPT. The log and the publishing checklist are different jobs. Finish the row before you rewrite a page, or you will not know whether the rewrite moved anything.

FAQ

How do I track brand mentions in ChatGPT?

Freeze the buyer prompts, ask them in a clean ChatGPT session, and log the prompt, mention yes or no, position, competitors named, and cited sources. Repeat the same list. One reply is a sample. Four to eight weeks of the same list is a tracker. The free plan covers 5 prompts and re-runs them every week, with no card, on the ChatGPT brand tracker.

Can I track mentions without paying for a tool?

Yes. Use the free tracker sheet, run the prompts in a clean session, and log the rows weekly. A product automates the same workflow when you want history and alerts without rebuilding the sheet. The comparison of paid tools is on best tools to track ChatGPT brand mentions.

Is one ChatGPT reply enough?

No. The next session can name a different shortlist. Count how often you appear across the frozen list.

Start a free project