To track brand mentions in ChatGPT, ask ChatGPT the same list of real buyer questions every week in a clean session, and log three things for each answer: whether it names your brand, where you sit in any list, and which URLs it cites. ChatGPT gives brands no dashboard for this, so a fixed prompt list and a simple log are the whole method.
Doing it by hand costs nothing and shows you exactly what a buyer sees. It also gets slow once you track more than a couple of dozen questions, which is where a tracker earns its price.
In this guide, you'll learn:
- What counts as a mention, a citation and a position
- How ChatGPT search changes the answer you see
- The manual method, step by step
- How to read your log
- When a tool beats doing it by hand
The rest of this guide covers it in order.
What you are tracking
Three different things can happen to your brand in one ChatGPT answer. Log them separately, because they move for different reasons.
| Signal | What it looks like | What it tells you |
|---|---|---|
| Mention | Your brand name appears in the answer text | ChatGPT knows you and connects you to the question |
| Citation | A link to one of your pages appears as a source | Your page was retrieved and used to build the answer |
| Position | Your place in a list of recommended brands | How strongly you are recommended against competitors |
A mention and a citation often do not go together. An answer can name you without linking you, and it can link one of your pages without writing your name. For the definitions behind these terms, see brand mentions in AI, AI brand mentions and AI citation rate.
How ChatGPT search changes the answer
Before you start logging, it helps to know which settings change what ChatGPT says. Every claim below comes from OpenAI's own help pages, checked 12 November 2026.
Search can be automatic or forced. ChatGPT "may search the web automatically when your question would benefit from current information." You can also force a search by choosing Search from the tools menu, or by typing / in the composer and choosing Search (OpenAI: Searching the web with ChatGPT).
Search answers show sources. Responses that use web search may include citations, and a Sources button lists the cited sources and other relevant links (same page). If an answer shows no citations and no Sources button, treat it as an answer from the model's own knowledge.
Logged-out users get search too. Web search is available on all ChatGPT plans, from Free to Enterprise, and "people who are not signed in can also use web search" (same page).
Memory personalizes answers. When Memory is on, ChatGPT can draw on past chats, saved memories and custom instructions (OpenAI: Memory in ChatGPT). It may also use saved memories when it rewrites your question into a search query (OpenAI: Searching the web with ChatGPT).
Location shapes local results. ChatGPT uses an approximate location based on your IP address and may share that general location with its search providers (same page).
Ads are separate from the answer. Free and Go users may see a labeled ad below a response. OpenAI says ads "do not influence ChatGPT's answers" (OpenAI: Ads in ChatGPT). An ad for your brand is not a mention, so leave it out of your log.
Blocking the search crawler hides you. OpenAI's crawler docs say sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers" (OpenAI: Overview of OpenAI crawlers). Check your robots.txt with the free AI crawler checker before you blame your content.
How to track brand mentions in ChatGPT by hand
Step 1: Build a prompt list from real buyer questions
Write 15 to 25 questions in the words your buyers use. Good sources are sales call notes, support tickets, the "People also ask" box in Google and your own Search Console queries. Mix three kinds:
- Category questions, such as "best invoicing software for freelancers"
- Problem questions, such as "how do I stop chasing late client payments"
- Comparison questions, such as "Brand A vs Brand B for a small agency"
Leave your brand name out of most of them. A question that names you will almost always mention you, and it tells you little. Freeze the list once it is written, since changing the wording breaks the week-to-week comparison.
Step 2: Set up a clean session
Your own account carries memory and chat history that a buyer's account does not. Use one of these two setups, and use the same one every week:
- Logged out. Open a private browser window and use ChatGPT without signing in.
- Logged in, unpersonalized. Start a Temporary chat and choose Unpersonalized before the first message. OpenAI says this option "does not use memory, custom instructions, or plugins" (OpenAI: Temporary chat in ChatGPT).
Run from the same country and network each time, because ChatGPT uses your IP address for location.
Step 3: Run each question twice, once on Auto and once with Search
First, send the question as is and let ChatGPT decide whether to search. Then start a fresh chat and send it again with Search selected.
The two runs answer different questions. The Auto answer can come from what the model already knows about your category. The forced Search answer shows which pages ChatGPT finds and trusts today. OpenAI documents no switch that turns search fully off in the ChatGPT app, so the Auto run is the closest you get to a no-search answer.
Step 4: Record the answer, the brands and the cited URLs
Copy each answer into a sheet with these columns:
| Column | What to write |
|---|---|
| Date | The day you ran it |
| Question | The exact wording from your frozen list |
| Session | Logged out, or Temporary chat unpersonalized |
| Search | Auto or forced |
| Citations shown | Yes or no |
| Brand named | Yes or no |
| Position | Your place in the list, or "prose" if there is no list |
| Competitors named | Every other brand, in order |
| Your URLs cited | Each of your pages that appears as a source |
| Other URLs cited | The other sources, so you know where to get covered |
| Notes | Anything wrong about you: old prices, missing features |
Save a screenshot of each answer too. It settles arguments later.
Step 5: Repeat every week
The same question can return a different answer on the next run. ChatGPT even has a Try again control that regenerates a response (OpenAI: Searching the web with ChatGPT). One run is a sample. Run each question three times per check and record how many of the three name you. Then repeat the full check on the same weekday each week.
How to read the results
Turn the log into three numbers per week:
- Mention rate: runs that name you, divided by all runs.
- Citation rate: runs that cite one of your pages, divided by all runs.
- Average position: your mean place across the list answers that include you.
Then look at where the signals disagree, because that is where the work is:
- Named on Auto, not cited on Search. ChatGPT knows your brand, but it does not find your pages for this question. Write a page that answers the question directly. Our guide on optimizing for ChatGPT covers the page shape.
- Cited, but not named. ChatGPT uses your page as a source for a fact and recommends someone else. Make your brand and what it does explicit on that page.
- Neither named nor cited. Look at the "Other URLs cited" column. Those review sites, lists and forums are where ChatGPT learns about your category. How AI citations work explains why those sources win.
- Named, but wrong. Fix the facts on your own site first, then on the third-party pages that ChatGPT cites.
Compare yourself with the competitors in your log, too. Your share of the brand names across all runs is a fairer score than your mention rate alone.
When a tool beats doing it by hand
The manual method scales badly. Twenty questions, two modes and three runs each come to 120 answers a week, before you copy anything into a sheet. A tool makes sense when one of these is true:
- You track more than about 25 questions, or more than one brand.
- You need history and charts to show a manager or client.
- You want the same questions checked in other AI engines as well as ChatGPT.
Before you buy, ask each tool how it collects ChatGPT answers. An answer collected through an API can differ from what the ChatGPT app shows a signed-in buyer. Whatever you pick, keep a monthly manual spot check of five questions to confirm the tool matches what a buyer sees.
Rankwise is one option. It checks ChatGPT for your tracked questions on every plan, from Starter at $59 a month with 50 questions per website (pricing). It asks ChatGPT through an API with web search turned on, re-checks every question every 14 days, and records whether your brand was named and whether your pages were cited as separate results. It has no prompt-volume data, and it has no daily re-check option. For other options, see our list of ChatGPT rank trackers.
Frequently asked questions
Is there a free tool to see if ChatGPT mentions your brand?
ChatGPT itself is the free tool, used with the method above. Rankwise has no free ChatGPT checker. For the other engines, the free Claude rank tracker and Perplexity citation tracker show whether each one names your brand and which URLs it cites, with no account needed.
Does ChatGPT give everyone the same answer?
No. Memory, location, the search setting and plain run-to-run variation all change the answer. That is why the method fixes the session, runs each question three times and tracks rates over weeks.
How often should I check?
Weekly is enough for most brands. Check more often for a week or two after you publish or change a page you expect ChatGPT to cite.
Next step
Start with the free AI audit. It needs no account and shows your AI visibility score, whether AI crawlers can reach your site and what to fix first. When you are ready to track ChatGPT every week without the spreadsheet, Rankwise does it on every plan.