10 November 2026

How to track brand mentions in ChatGPT: a manual method, its limits and when to automate

A weekly routine for tracking brand mentions in ChatGPT: unbranded questions, a clean test set-up, a one-row-per-answer sheet, and the sample size you need before trusting a percentage.

To track brand mentions in ChatGPT, write a fixed list of customer questions, ask each one in a fresh chat with search switched on, and record four things every time: the date, whether your brand was named, where it came in the order of brands, and which pages were cited. Repeat on a schedule and count. ChatGPT gives site owners no report of its own, so a log you keep yourself, or a tool that keeps it for you, is the only way to know.

This is a method for a UK small business or a small agency. It takes about an hour to set up and twenty minutes a week to run.

Key takeaways

  • ChatGPT has no equivalent of Search Console. Tracking means sampling answers yourself.
  • Use unbranded questions. If the question names you, the answer tells you nothing about being found.
  • Start a new chat for every question, with memory and personalisation off, or earlier chats will colour the answer.
  • Record the cited sources as well as the brands. The sources tell you what to fix.
  • One answer is a snapshot. Judge a trend only on a few dozen answers per period.

Why can't you just look it up?

Because OpenAI publishes no dashboard that tells a business how often ChatGPT names it. Google reports AI feature appearances inside Search Console, and Bing Webmaster Tools has an AI Performance report in public preview for Copilot. ChatGPT has neither.

What OpenAI does document is how search answers are shown. Its help page says ChatGPT searches the web when a question "might benefit from information on the web", adds inline citations to the reply and lists all sources behind a Sources button (OpenAI Help Center, ChatGPT search). Those two things, the text and the sources, are what you can record. For the wider picture across engines, see our guide to AI search visibility.

What should you set up before you start?

Set up three things: a question list, a competitor list and a clean test environment. Skip the third and manual tracking gives misleading results.

  1. Ten to twenty questions. Write what a customer would ask before they know you exist: "best payroll software for a UK business with five staff", "who repairs Miele washing machines in Leeds", "alternatives to a named market leader". Do not include your brand name. Keep each question under about 30 words.
  2. Three to five competitors. Include the obvious rival and the one that wins comparisons you think you should win.
  3. A clean environment. Use a separate ChatGPT account or a temporary chat, turn off memory and custom instructions, and start a new chat for each question. Otherwise ChatGPT may tailor answers to what it already knows about you, including the fact that you keep asking about your own company.

Check access while you are at it. OpenAI's crawler documentation says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers", and that a robots.txt change takes about 24 hours to take effect (OpenAI crawler documentation). If your robots.txt or your CDN blocks that crawler, fix it before you measure anything.

How do you run the check, step by step?

Run every question the same way each time, so the only thing that changes between weeks is the answer.

  1. Open a new chat. Make sure web search is available; you can also switch it on from the search control in the message box.
  2. Paste the question exactly as written in your list. Do not add "in the UK" one week and leave it off the next.
  3. Wait for the full answer, then open the Sources list.
  4. Copy the answer text into your log, or save it as a file named with the date and question number.
  5. Record the brands named, in the order they first appear.
  6. Record the cited domains, and note whether yours is among them.
  7. Close the chat. Open a new one for the next question.

Your location affects the result. OpenAI says a general location based on your IP address may be shared with search providers to improve accuracy, so run the checks from the country your customers are in, and do not mix runs done on holiday with runs done at your desk.

What should the tracking sheet look like?

One row per answer, not one row per question. That is what lets you count properly later.

Column What goes in it Example
Date Day the answer was collected 10 Nov 2026
Question ID A short code from your fixed list Q07
Searched? Whether the answer showed sources Yes
Named? 1 if your brand appears in the text, 0 if not 1
Position Order in which you were first named among brands 3
Brands named All brands in order, separated by commas Brand A, Brand B, You, Brand C
Cited domains Domains in the Sources list a comparison site, a trade magazine, Brand A's site
Your page cited? 1 or 0, and which URL 0
What was said One line on how you were described, and any error "Good for small teams"; price quoted is out of date

That sheet gives three figures. Visibility is the answers that named you divided by all answers. Share of voice is your mentions divided by all brand mentions, counting a brand once per answer. Citation rate is the answers that cited one of your pages divided by all answers that showed sources.

How many times do you need to ask?

More than once. The same question can return a different set of brands on the next run, so a single check can make you look invisible or dominant by luck.

A workable minimum for a manual routine is ten questions, three runs each, every week. That gives 30 answers a week and about 120 a month. With 30 answers, a visibility of 40% carries a 95% interval of roughly 25% to 58%; with 120 answers it tightens to about 32% to 49%. So compare months, not weeks, and do not celebrate a rise from 40% to 45% on a week's data.

Always keep the count beside the percentage. "Named in 12 of 30 answers" is honest. "40% visibility" on its own hides how thin the evidence is.

What do you do with the results?

Read the cited sources before anything else, because they show where ChatGPT got its picture of your market.

  • You are named and described correctly. Note which pages were cited and keep them current.
  • You are named but something is wrong. Find the cited page carrying the error. If it is yours, fix it in plain text. If it is someone else's, ask for a correction with the right fact and a link.
  • Competitors are named and you are not. Look at the domains cited across those answers. If the same comparison article or directory keeps appearing and you are not in it, that is the gap. Being accurately listed there is the first job.
  • No sources at all. The answer came from the model's training, not a search. You cannot change that directly; consistent public information about you is the long game.

Businesses that depend on reputation locally will find reviews among the material engines read. Our guide on how to get more Google reviews covers what you may and may not ask customers for.

When is it time to automate?

Automate when the routine starts to slip or the sample is too small to answer your question. Typical triggers: more than about twenty questions, a need for daily data, several engines, or client reporting that needs stored evidence.

The manual method also has blind spots a tool removes. You are one person in one place on one account, your checks cluster on the day you remember to do them, and copying brand lists by hand introduces errors. Our comparison of AI visibility tools lists what the published plans cost and include. The same routine applies to Google's surfaces, with differences set out in AI Overviews vs AI Mode.

Doing this with CiteSieve

CiteSieve, from Braivex, runs this routine for you. You add your brand, the other names people use for it, your competitors and your questions. It asks each question once a day on every engine you have switched on, and stores each answer in full with the brands named in order and the sources cited. You sign in at citesieve.braivex.com with a Braivex account.

Your brand and competitors are matched by exact text, visibility is shown with the number of answers and a 95% interval, and everything can be exported as CSV. ChatGPT, Google AI Mode and Gemini are on by default. One answer from one engine costs one credit, so a question tracked daily on ChatGPT alone uses about 30 credits a month; plans start at $95 a month for 470 credits, and the trial is 14 days with 70 credits and no card.

What it cannot do: it does not see what real ChatGPT users type, it does not change what ChatGPT says, and its answers are collected through a third-party collection service as a searcher in your chosen country would get them, which may still differ from what a logged-in person with a long chat history sees.

Other Braivex tools for this job

  • Peakwyn is a paid, clearly labelled leaderboard where you can list a website; every rank is advertising, and our guide on where to list your startup or SaaS covers the free and paid places a new brand can be found.
  • ReelSieve turns an Airbnb listing into a short walkthrough video, which gives a holiday let something other than text for people to find.
  • HireSieve screens CVs against a rubric written from your advert, for the day the marketing work needs another pair of hands.
  • Reachwyn writes LinkedIn posts in the owner's own voice and publishes only what you approve.
  • Housieve scores UK sale, rental and commercial listings for agents and sourcers.
  • BriefSieve (VirtualPA) writes a morning brief from Shopify, Klaviyo and ad data; our ecommerce marketing plan shows where search fits in a small shop's year.
  • Loculens gathers Google reviews for every branch and drafts replies a person checks and posts; see our guide to multi-location reputation management.
  • VoiceSieve is Braivex's AI phone agent. It is in development and is not live.

Frequently asked questions

Can you see how often ChatGPT mentions your brand?

Not from OpenAI. There is no owner dashboard. You have to ask a fixed set of questions repeatedly and count the answers, by hand or with a tracking tool.

Does ChatGPT give the same answer every time?

No. The brands and sources can change between runs of the same question, which is why you need several answers before quoting a percentage.

Should the questions include my brand name?

Not for visibility tracking. Use unbranded questions. Keep a separate short list of branded questions if you want to check that ChatGPT describes you accurately.

Why does ChatGPT sometimes answer without sources?

It searches only when it judges the question would benefit from web information. Without a search, the answer comes from training and shows no citations.

Does blocking GPTBot stop ChatGPT mentioning me?

OpenAI says GPTBot is for training and OAI-SearchBot is for search, and that the settings are independent. Blocking OAI-SearchBot is what removes a site from ChatGPT search answers.

How often should I check?

Weekly is a sensible manual rhythm. Daily collection gives a larger sample and a clearer trend, which is where a tool helps.

For a first data point without a spreadsheet, the free CiteSieve report runs a short one-off check on ChatGPT and Google AI Mode and emails you the brands named and the sources cited. Treat it as a snapshot and start your log from there.