How to Track Brand Mentions in ChatGPT and Perplexity (2026)

Short answer: To track your brand, fix a panel of 30 buyer prompts, run each one three times a week on ChatGPT and Perplexity, and log whether your brand is named, where it appears, which competitors show up and which URLs are cited. Through the APIs the panel costs about $5.85 a month in search fees before model tokens, on vendor prices we read on 10 October 2026. By hand the method costs staff time only. Your mention rate is the share of answers that name you.

Toolradar data: Toolradar, the software directory we run, evaluated 28 brand monitoring tools for its October 2026 ranking, and only 2 of the 27 it ranks have a free plan. A tool built to watch news and social posts will not show you what an assistant said, so check whether a product names ChatGPT or Perplexity among its sources before you pay for it. Toolradar's full brand monitoring ranking and our own best brand monitoring tools roundup cover that category.

What a brand mention is, and what to measure

Four measures matter. Three of them are columns in the log, and the fourth is worked out from those columns.

• Mention: your brand or product name appears in the answer text. A link to your site without your name does not count.
• Mention rate: answers that name you, divided by all answers in the panel for that week, or for one engine. This is the number to watch.
• Position: where your brand first appears, if it is listed at all. A brand named last in a list of nine is still a mention, but it is not a recommendation.
• Cited sources: the URLs the answer links to. They show where the assistant found its evidence, which is often more useful than the mention itself.

One answer proves very little. The same question can come back with a different list, a different order or different wording. A rate across many answers is what you can act on, and a screenshot of one good answer is not a rate.

Step 1: Build a fixed panel of 30 prompts

Use four groups that add up to 30 prompts. The category group carries the most weight, because buyers ask those questions before they know your name.

Take the wording from real buyers. Sales call notes, support tickets, Search Console queries and the questions people post in community forums all work. Write each prompt the way a buyer types it, with no keyword stuffing, because a prompt written to suit you tells you little about what a buyer will see. For the everyday ways marketers already use assistants, our ChatGPT for marketing guide is a good source of problem prompts.

Freeze the panel for at least four weeks. Change one prompt and the trend line breaks. New questions go into a second panel, with their own rate, so the first one stays comparable.

Step 2: Run it the same way every time

Run the same prompts, in the same order, on the same days each week. Open a new chat for each prompt so earlier questions do not steer the answer. Log whether web search was on, because an answer that searches the web and one that does not can rest on different sources. Note the account type (logged out, free or paid) and the location, and keep that setup fixed from week to week.

Run each prompt three times. Answers are generated text, so one run is an anecdote. Three runs give you a rate you can compare from one week to the next, and they show you when the list of brands itself is shifting.

Step 3: Log every answer, including the misses

Use one row per answer. A spreadsheet with these columns covers the job.

Log the misses. A log with only the answers where you appear overstates your rate, and it hides the prompts where a competitor owns the question. The rows where you are absent are the ones that tell you what to fix.

Step 4: Score the panel by group and engine

Mention rate is answers that name you, divided by all answers. Here is an example. If your brand is named in 45 of 180 answers in one week (30 prompts, three runs, two engines), the rate is 25 percent, because 45 divided by 180 is 0.25. The figures in this example are illustrative arithmetic, not data.

A single rate hides the split, and the split is the useful part. Break the same log down by group and engine. A 25 percent overall rate could come from strong category visibility and weak comparison visibility, or the reverse. Those are different problems with different fixes, so score each group on its own line.

Read the rate over several weeks before you act on it. One week shows a snapshot. A four-week average shows a direction, and it is the figure to put in front of anyone who funds the work.

Step 5: Read the sources, then act

The cited-URLs column is the most useful part of the log. For every prompt where a competitor is named and you are not, open the pages the answer cites. Those pages are where the assistant found its evidence.

You have two routes from there. The first is to update your own page if it should be the cited source and does not answer the buyer's question directly. The second is to get your brand into the third-party page the engines already cite, such as a comparison, a review or a directory listing. Start with the sources that appear in three or more answers, since those are the pages the assistant keeps returning to. If a cited source is an SEO platform's comparison, our best AI SEO tools roundup shows which platforms also report on AI answers.

Step 6: Move to the API when hand logging breaks

By hand, a 30-prompt panel on two engines, run three times a week, produces 180 answers a week. Logging that many answers every week is hard to keep up, and the quality of the log slips first. The APIs run the same panel on a schedule and keep the sources for you.

The arithmetic runs like this. Thirty prompts times three runs is 90 calls per engine each week. Over 52 weeks divided by 12 months, that is 390 calls per engine per month. Add the two request fees and the total is $5.85 a month before model tokens. Token charges depend on answer length and on the model you pick, so budget them on top. The sources are the OpenAI API pricing page and Perplexity's pricing page. Perplexity's page publishes no date of its own, so its figures are dated to our read on 10 October 2026.

One caveat matters more than the price. The API is not the consumer app. Sources and wording can differ from what a buyer sees in the ChatGPT or Perplexity app. Spot-check a handful of prompts in each app every month, and say which numbers came from the API whenever you share them.

When a paid tracker earns its price

Paid tools make sense when you need alerts, dashboards, more prompts than your team can run, or several seats. Check engine coverage before you compare prices, because a plan that skips an engine your buyers use is cheap for the wrong job. Our AI visibility tracker pricing page prices several trackers on one 50-prompt scenario, and our best GEO and AI search optimization tools ranking shows which engines each one covers. If you are weighing a single product, our Peec AI review walks through one tool that bills in euros, and Toolradar keeps software profiles for Peec AI and Otterly. If the tool you already pay for is an SEO suite, check its AI prompt allowance first: our Ahrefs pricing breakdown lists the limits on Ahrefs' plans, and our Surfer SEO review covers that platform.

Gotchas that break the trend line

• Logging only the answers you like. The panel has to include the runs where you are absent.
• Counting any name as a recommendation. A name in a list is a mention with a position, not a pick. Read the sentence around the name before you count it as a recommendation.
• Rewording prompts to chase a result. Freeze the panel. New questions go into a second panel with their own rate.
• Mixing accounts and locations. A logged-in account with chat history and a logged-out session can return different answers. Pick one setup and keep it.
• Ignoring the sources column. A mention with no cited source gives you nothing to act on.

What to do each month

Pick two actions. First, fix the page behind the most-cited source where you are missing, or write the page that answers the prompt directly. Second, earn a named mention on a third-party page the engines already cite. Then rerun the panel four weeks later and compare the averages, not a single week.

FAQ

How do I track my brand in ChatGPT for free?

Build a 30-prompt panel, run each prompt three times a week in ChatGPT, and log the brand named, the position, the competitors and the cited URLs in a spreadsheet. The fee is zero, and the cost is staff time. Keep the account and location fixed, because both change the answer.

How many prompts do I need?

Thirty prompts cover the four buyer groups well enough to show patterns. With 20 prompts and one run each, a single answer moves an engine's rate by 5 points, so a larger panel holds steadier from week to week.

How often should I check?

Weekly is enough to see a trend. Run each prompt three times that week, and compare four-week averages before you decide anything.

Is the ChatGPT API the same as the ChatGPT app?

No. The API returns answers through your own request, so the wording and the sources can differ from what a buyer sees in the app. Spot-check a few prompts in the app every month.

Does Perplexity show sources?

Yes. Perplexity answers carry numbered citations, so log each URL in the cited-URLs column in the order the answer gives them.

Do I need a paid tool?

Not to start. Add one when you need alerts, dashboards, more prompts than your team can run or several seats, and check which engines it covers before you look at the price.

Cite this: MarketingShot, "How to Track Brand Mentions in ChatGPT and Perplexity (2026)", October 2026.

How we verified this: The OpenAI web search price of $10 per 1,000 calls comes from OpenAI's API pricing page, and the Perplexity Sonar prices come from Perplexity's pricing page. Both were read on 10 October 2026. The brand monitoring figures come from Toolradar's brand monitoring ranking, updated October 2026. The panel, the logging method and the scoring are our recommendation for a team that wants a repeatable measure. We have not published a multi-week study on this page.

Some offers on this page may be paid placements or contain affiliate links.