How to Track Brand Mentions in ChatGPT and Perplexity (Step by Step)

Tracking means asking the same buyer questions again and again and counting what the answers say about you. Three numbers matter. Mention rate is the share of answers that name your brand. Citation rate is the share that link to a page on your own website. Position is where you appear when the answer is a list. A brand can be named without being cited, for example when the answer links a directory or a marketplace instead of your site.

This is not social listening, which counts posts that people write about you. Here the AI engine writes the text, the text changes from run to run, and only the person who asked sees it. If you only want a one-off answer to "does ChatGPT know us?", our guide on how to check if ChatGPT mentions your brand is enough. This page is about tracking over time, by hand or with the Leadee Visibility Tracker. Every other ChatGPT guide is listed on our ChatGPT SEO hub.

Both engines can search the web before they answer, but they do not pick the same pages or the same brands. In Leadee's tracker round of 2026-10-25, Perplexity named one of seven Netbase JSC brands in 46 of 252 runs (18.3%, 95% interval 14.0-23.5%). OpenAI's gpt-5 with web search, measured through OpenAI's API, named one in 3 of 25 runs (12%, 4.2-30.0%) before the account ran out of credit.

The two intervals overlap, so this round cannot say which engine names these brands more often. It does show that a number for one engine tells you little about the other. Keep each engine on its own line and compare like with like. Our side-by-side study of ChatGPT, Gemini and Perplexity citations has the full table, including Google AI Mode.

Pick 15 to 30 questions a buyer who does not know you yet would type. Mix a few shapes: "best [product or service] for [need]", "[option A] vs [option B]", "alternatives to [a well-known brand]" and "who supplies [thing] in [country]". Use the buyer's words, not your marketing words. Add three to five questions that contain your name, such as "is [brand] reliable", because those show whether the engines describe you correctly.

Questions without your name are the real test. In the same round, pooled over three engines, the seven Netbase JSC brands were named in 62 of 84 US answers to questions with their name, but in only 20 of 274 answers to questions without it. The free ChatGPT Prompt Finder suggests questions for your category, and our guide on finding the prompts your customers ask shows how to use your own search and sales data.

Decide which engines and which countries you track. A US buyer and a UK buyer can get different answers, so keep each market on its own line. Then run every question at least three times on each engine in every round. AI answers are not fixed: the same question can return a different list of brands a minute later.

Here is why three runs matter. In Leadee's tracker, 159 question sets were asked exactly three times on 2026-10-25. In the 35 sets where a brand appeared at least once, one run picked at random would have missed it in 22 of 105 runs (21%, 95% interval 14.3-29.7%). So a single check by hand gets it wrong about one time in five. Our comparison of manual checks and automated tracking has the full count and the cost of each approach.

Use one row per answer: date, engine, market, question, run number, named (yes or no), cited URL, position and a short note. Count a mention only when the brand is clearly yours. A similar product name, or another company with the same name, does not count. Our ChatGPT prompt tracking sheet gives you the columns ready to copy.

Through the official APIs, citations come back as data, so you do not have to read links out of the text. OpenAI's web search guide says each citation is a url_citation annotation with the URL, title and location in the answer, and that "the number of sources is often greater than the number of citations". Perplexity's Sonar documentation shows a citations list of URLs and a search_results list with title, URL and date. Record the pages that are cited, not every page the engine looked at.

Divide mentions by runs to get a mention rate, and do the same for citations. Then add a 95% confidence interval: a range that shows how far the rate could move by chance alone. Leadee uses the Wilson score interval, a standard method described in the NIST/SEMATECH statistics handbook, because it stays sensible with small samples and with rates close to zero.

Sample size changes everything. At the same observed rate of 30%, 3 mentions in 10 runs gives an interval of 10.8% to 60.3%, while 30 in 100 gives 21.9% to 39.6% (arithmetic examples, not measurements). With ten runs you can say little more than "sometimes". Our guide to the AI visibility metrics that matter explains each rate and which to read first; our methodology page shows how Leadee computes them.

Run the same list, the same number of times, on the same engines, every week or every month. Change the list only between periods, and write down the date when you do. Compare two periods only when their intervals stop overlapping; a change inside the overlap is normal variation. Log every change to your site, listings or reviews in the same sheet, so you can see what happened before a number moved.

A spreadsheet works for a handful of questions on one engine. It gets heavy once questions times engines times markets times runs adds up to hundreds of rows per round. That is the point where automated tracking saves more time than it costs.

Look at the questions where you are missing. Read who is named instead and which pages are cited. If the cited pages are directories, review sites or articles, those are the places to be present. If the engines cannot reach your own pages, fix access first: our guide to OAI-SearchBot, GPTBot and ChatGPT-User shows the exact robots.txt lines. To be quoted more often, see how to get cited by ChatGPT.

If the answers say something wrong about you, such as an old price or a mix-up with another company, follow how to fix wrong information in ChatGPT. Then re-measure with the same questions about four weeks after the change goes live, and only call it a win when the new interval sits clear of the old one.

We track the seven websites of Netbase JSC, the company behind Leadee, in Leadee's own tracker. In the round of 2026-10-25, across all markets, Perplexity named one of the brands in 46 of 252 runs and cited its domain in 41. OpenAI's gpt-5 with web search named one in 3 of 25 runs, all of them CMSmart in the US, before the account ran out of API credit. Treat that OpenAI number as an indicator only.

Two lessons shaped how we track. First, the gap between questions with and without the brand name is large, so a list full of your own name flatters you. Second, single runs disagree often enough that one check can mislead. The per-brand figures are in the Netbase portfolio case study.

Leadee is an AI-native platform that stands on the shoulders of giants: it asks the engines themselves through their official APIs and never scrapes a consumer chat app. The Visibility Tracker runs the steps of this guide for you, round after round.

Leadee runs the questions, stores every answer and computes the rates. Your team owns the question list, reads the results and decides what to change. Leadee never changes your website on its own. If you want experts to set up the prompt set and turn the results into a fix list, start with the AI Visibility Audit. If you want the fixes done for you as well, see Managed GEO.

Plans and prices are on the pricing page. To compare tracking platforms before you decide, including Leadee and other market players, see our overview of AI visibility tools.