How to Track Your Brand in ChatGPT: A Repeatable 7-Step Method

Buyers no longer build supplier shortlists only on Google. They also build them in AI assistants such as ChatGPT. So how to track your brand in ChatGPT is now a measurement problem for B2B marketing and sales teams, not just a curiosity. The difficulty is that ChatGPT can give a different answer to the same question every time. A one-off "let me ask about my brand" test can make the picture look better or worse than it really is. This guide gives you a repeatable measurement protocol, formulas, a worked example with realistic numbers and copy-ready templates.

Short answer: To track your brand in ChatGPT, write a list of 20-50 questions your customers would actually ask. Ask each question at least three times in a private window with web search turned on. For each answer, log whether your brand is mentioned, its position in the list, whether your site is cited as a source and whether the facts are correct. Repeat weekly alongside competitors, and move to a tracking tool once the work grows.

Key Takeaways (Short Answer)

• Question list: Prepare 20-50 real customer questions by funnel stage. Keep questions that contain your brand name separate from questions that don't.
• Fixed conditions: Use the same setup for every run and record it: private window or logged out, memory off, web search on or off, language and date.
• Repetition: Ask every question at least 3 times, because ChatGPT answers are probabilistic.
• Metrics: Track mention rate, average position, citation rate, share of voice, sentiment and accuracy.
• Cadence: Measure weekly or every two weeks together with competitors and watch the trend. Move to an automated tool once the work outgrows a spreadsheet.

One answer to one question tells you nothing about your ChatGPT visibility. It only shows the roll of the dice at that moment. At the end of this article you'll find a tracking template you can paste into Google Sheets and a weekly checklist.

Which 'Brand Tracking' Do You Mean?

In Turkish, 'marka takibi' (brand tracking) describes two very different jobs, and search results mix the two intents.

Quick distinction

• ChatGPT brand tracking (AI visibility): Measures whether ChatGPT and similar assistants mention your brand in their answers, and at what position, in what tone and how accurately. This is a marketing and sales task.
• Trademark registration monitoring: Follows the status of your trademark application, official trademark bulletin publications, oppositions and new filings that resemble your mark. This is a task for your legal team or trademark attorney.

If you want to check a registration status or whether a name is already registered in Turkey, go to the TÜRKPATENT trademark search screen. This article covers the first meaning: brand visibility in AI answers.

What Is ChatGPT Brand Tracking, and Why Is It Different from Google Rankings?

On Google, your page has one position for a query. In ChatGPT, a single answer has several dimensions at once:

• Mention: Does your brand appear in the answer at all?
• Position: Where are you in the list of recommended brands?
• Citation: Is your website listed among the answer's sources?
• Sentiment: Are you recommended, listed neutrally or mentioned with a warning?
• Accuracy: Are your price, address, features and service area correct?

Is a brand mention always a recommendation?

No. All three of these sentences count as a 'mention', but they have very different effects:

• 'X stands out for this need; it has a local support team and a Turkish interface.' (positive recommendation)
• 'X and Y offer similar features, but Y has broader integration options.' (comparative, in the competitor's favour)
• 'Some users complain about X's support response times.' (negative)

That is why every mention should be logged together with its sentiment. A user can also build a shortlist in ChatGPT without clicking a single link. Your visibility can shape a buying decision before it ever turns into traffic. For that reason, measure answers directly and don't rely on traffic metrics alone.

How Does ChatGPT Decide Which Brands to Recommend?

ChatGPT answers can come from two different sources of knowledge, and each one is tracked differently.

1. Model knowledge (training data): When web search is not used, the model answers from what it learned during training. That knowledge is current only up to a certain date and changes slowly. These answers show no source list. What you can measure is how much your brand appears in the model's 'memory', and how it is described.

2. Web search (RAG): With retrieval-augmented generation, ChatGPT turns the question into one or more search queries, reads pages and builds the answer from them. These answers show source links. The pages ChatGPT read are the most valuable raw data in your tracking. If a 'best X' list or a comparison site pushes your competitor to the top, that page is exactly where you need to work.

What do OpenAI's crawlers do?

Crawler	Purpose	If you block it
OAI-SearchBot	Crawls and indexes sites for ChatGPT search results	Your site may not appear as a source in web-search answers
GPTBot	Collects content for model training	Future models may know less about your brand; it does not directly decide search visibility
ChatGPT-User	Visits a page when a user asks ChatGPT to open it; this is not automatic crawling	ChatGPT may not be able to read your content when a user points it to your page

OpenAI states that these settings are independent of each other. For example, you can block GPTBot and still allow OAI-SearchBot.

Simple flow

User question
   ↓
Is search needed? → No → Answer from model knowledge (no sources)
   ↓ Yes
Search queries → Pages read and sources selected
   ↓
Answer + brand list + source links

How to Track Your Brand in ChatGPT: The 7-Step Manual Method

1. Build the question list by funnel stage. Write at least 5-10 questions each for awareness, evaluation and decision. Base them on real questions your sales team hears, demo notes and Search Console queries.
2. Fix the measurement conditions and record them. Use a private window or a logged-out browser. If you must be logged in, turn off memory and custom instructions, or use a temporary chat. Note whether web search is on or off, the language, the date and the visible model version. If the conditions change, your results can't be compared.
3. Ask every question at least 3 times. Open a new chat for each repetition. Never decide based on a single answer.
4. Log every answer in the sheet. Record mention (0/1), position (blank if not mentioned), citation (0/1), sentiment (positive/neutral/negative) and accuracy (correct/incorrect + error type).
5. Mark 3-5 competitors in the same answers. Don't run separate queries for competitors. Record whether they appear in the same answers, so every brand is measured under the same conditions.
6. Calculate the metrics. The formulas are below.
7. Repeat weekly or every two weeks. A single measurement is a snapshot. The value shows up in the trend.

Which Questions Should You Ask ChatGPT? Prompt Templates

The most common mistake is to test only with branded questions such as 'What is [Brand]?'. When you supply the brand name, ChatGPT will naturally talk about your brand, so visibility looks higher than it really is. Keep two separate sets:

• Non-branded questions: These test real visibility. Calculate mention rate, position and share of voice from this set only.
• Branded questions: These test accuracy and sentiment. What does ChatGPT say about you, and is it correct?

Persona variations

Ask the same question with different context: 'I own an SMB with a 10-person team', 'I'm a procurement lead at an enterprise and data-protection compliance matters', 'I'm a student on a tight budget'. Personas change the answer. Seeing which persona you lose tells you which content you need to create.

Copy-ready template (18 questions)

AWARENESS (no brand name)
1. What are the best [service] companies in [city]?
2. Can you recommend reliable companies offering [service] in [city]?
3. What are the most widely used [product category] solutions for [industry]?
4. What are affordable [product category] options for SMBs?
5. Which local companies offer [product category] in Turkey?
6. I have [problem]; which [product category] would help?

EVALUATION
7. What should I look for when choosing a [product category]?
8. Which criteria matter when an enterprise chooses a [product category]?
9. How much does [service] cost in [city], and which companies are affordable?
10. What are the differences between [Brand] and [Competitor]?
11. [Brand] vs [Competitor]: which is better for SMBs?
12. What are the alternatives to [Competitor]?

DECISION (branded: accuracy and sentiment test)
13. Is [Brand] reliable?
14. How much does [Brand] cost?
15. What do reviews say about [Brand]?
16. Does [Brand] have a branch or office in [city]?
17. Which integrations does [Brand] support?
18. Is [Brand] compliant with data-protection law (KVKK)?

Questions 10 and 11 contain your brand name. Leave them out of the mention-rate calculation and add them to the accuracy and sentiment set. Question 12 names only the competitor, so it is a strong visibility test for you.

Which Metrics Should You Track? Formulas and a Worked Example

• Mention rate = answers that mention your brand / total answers
• Average list position = sum of your positions in the answers where you appear / number of mentions. Answers without a mention are left out here, because the mention rate already captures them.
• Citation rate = answers that cite your website / total answers
• Share of voice = your mentions / total mentions of you and your competitors
• Sentiment mix = positive, neutral and negative mentions as a share of all your mentions
• Accuracy score = error-free answers / total branded answers. Count error types separately: wrong price, closed branch or old address, missing or non-existent feature, wrong service area.

Worked example

A B2B software company asks 30 non-branded questions, 3 times each, and collects 90 answers. Two competitors are marked in the same answers.

Metric	You	Competitor A	Competitor B
Mentions (in 90 answers)	27	52	21
Mention rate	27/90 = 30%	52/90 = 57.8%	21/90 = 23.3%
Average position	92/27 = 3.4	1.9	4.1
Citation rate	12/90 = 13.3%	–	–
Share of voice	27/100 = 27%	52/100 = 52%	21/100 = 21%
Sentiment (27 mentions)	15 positive, 9 neutral, 3 negative	–	–

The company also asks 10 branded questions × 3 repetitions = 30 answers. Four of them contain errors (3 outdated prices, 1 closed office address), so the accuracy score is 26/30 = 86.7%.

How to read it: Competitor A is mentioned more often and higher up the list. Your first question should be which pages cite Competitor A. The outdated price is a quick win you can fix right away.

Why Do You Get a Different Answer Every Time? Sampling and Repetitions

ChatGPT answers are probabilistic. The model chooses each word based on probabilities. In web-search answers, the generated search queries and the pages it reads can also change. The same question might list your brand third in one run and leave it out entirely in the next.

Rule of thumb: Before you make decisions, use at least 20-30 non-branded questions with 3 repetitions each. The reason is simple statistics. When you measure a rate of about 30% with 90 answers, random sampling error alone creates an uncertainty of roughly ±10 points (an approximate 95% confidence interval). Repetitions of the same question are not fully independent, so the real uncertainty can be somewhat wider. With a 10-answer test, the interval widens to about ±28 points, and the result tells you almost nothing.

Factors that increase variability: being logged in, memory and custom instructions, location (IP), language, web search on or off, and the model version. Fixing and recording these factors lets you tell whether a change came from your work or from the conditions.

Turkish vs English queries: A question asked in Turkish may lean on Turkish sources. The same question asked in English may lean more on global sources, which means a different list of brands. If you sell in Turkey, use the Turkish set as your main set. If you also sell abroad, report the English set separately and don't merge the two languages into one rate. Local businesses need a separate set of city-based questions ('in Ankara...', 'in Izmir...'), because recommendations change from city to city.

Noise or real loss? Because of the uncertainty above, a single-week drop of 5-8 points is usually noise. A practical threshold: raise an alert when you see a drop of more than 10 points in two consecutive measurements, then move on to source analysis.

How Do You See ChatGPT Traffic in GA4?

1. In GA4, open Reports > Acquisition > Traffic acquisition.
2. Set the primary dimension to Session source / medium.
3. Type 'chatgpt' in the search box. You will usually see a chatgpt.com / referral row.

ChatGPT search adds the utm_source=chatgpt.com parameter to many of the links it shows. As a result, this traffic often appears with chatgpt.com as the source, which makes it easy to filter.

To group all AI sources into one channel: Go to Admin > Data display > Channel groups and create a new channel group. Add a channel called 'AI Assistants' with the condition 'Session source matches regex' and this value:

chatgpt|openai|perplexity|gemini.google|copilot.microsoft|claude.ai

Move this channel above the Referral channel. Otherwise, these sessions will be assigned to Referral first.

Limitation: GA4 only sees clicks. If a user reads the answer and shortlists you without clicking, no report will show it. Measuring the answers is therefore essential, and GA4 is a complement to it.

Connecting it to business results: Build the chain like this: ChatGPT referral session → form submission or demo request (GA4 key event) → lead source 'AI' in your CRM → qualified opportunity → deal. Writing the UTM source into a hidden form field keeps that link intact. For a broader view of what happens after the form, see this guide to following campaign demand from lead to qualified pipeline.

Manual Tracking or a Tool? A Comparison

The cost of the manual method is time. 30 questions × 3 repetitions = 90 answers. You have to ask each one, read it and log five brands (you + 4 competitors). Even at 2 minutes per answer, that is about 3 hours a week. Add Gemini, Perplexity or Google AI Overviews, and that time multiplies by the number of engines. A practical threshold: move to a tool when you need to measure more than one engine or language, or when you can't keep a weekly rhythm for two months.

Criterion	Manual method	ChatGPT tracking tools (general)	Jumpix AI Visibility
Engine coverage	Whatever you ask by hand	Varies by tool	ChatGPT, Gemini, Perplexity, Google AI Overviews; Claude optional
Repetition automation	None	Most have it	Weekly automated measurement
Source analysis	Noted by hand	Varies by tool	Source site analysis
Competitor comparison	Marked by hand	Usually available	Share of voice for up to 5 competitors
Turkish dashboard	–	Rare	Yes
Price	Free, high time cost	Varies by tool	9.99 USD per month (1 brand)

Questions to ask when choosing a tool:

• Which engines does it measure? Does it measure web-search answers or model knowledge?
• Does it support Turkish and city-based queries?
• How many times does it repeat each question?
• Does it store the raw answer, so you can read the answer itself when you see a drop?
• Does it list the sites that were cited as sources?

One option you can assess against these criteria is Jumpix AI Visibility, which provides brand tracking in ChatGPT and other AI engines. It also offers a free instant check with no sign-up (industry + city, 5 questions, 3 engines).

What Should You Do If Your Brand Doesn't Appear in ChatGPT?

1. Diagnose

• Is OAI-SearchBot blocked in your robots.txt? If it is, you can't be cited in web-search answers. To allow it:

User-agent: OAI-SearchBot
Allow: /

• Are your key pages indexed? ChatGPT search also uses third-party search providers, so checking index status in Bing Webmaster Tools is a practical place to start.
• Can your pages be read without JavaScript?

2. Analyse the source sites

List the sources that push your competitors to the top in web-search answers. In Turkey these are usually Turkish 'best X' lists, comparison sites, complaint and review platforms, forums and communities, news sites and YouTube videos. Getting onto these pages often has a faster effect than editing your own site. That means getting listed with correct details, collecting reviews and giving editors the information they need. Consistent brand mentions across the web also feed model knowledge over the long term.

3. Strengthen your site

• Clear, separate pages for each service and each city
• Explicit prices or price ranges
• FAQs based on real customer questions
• Organization schema, plus LocalBusiness schema for local businesses
• Consistent NAP (name, address, phone) across your site, Google Business Profile and directories

4. Fix incorrect information

If ChatGPT quotes an old price, open the sources behind the answer. The error usually comes from an old blog post, an outdated directory listing or a forgotten page on your own site. Fix the source or request an update, then measure the same questions again. Errors that come from model knowledge take longer to correct, which is why correct information needs to be widespread and consistent across the web.

Measure → act → measure again: a sample task list

• Week 1: Baseline measurement (90 answers); list the accuracy errors
• Week 1: Check robots.txt and Bing indexing
• Week 2: Find and fix the source of the wrong price
• Weeks 2-3: Apply to, or supply content for, 3 lists that cite Competitor A
• Week 3: Publish the missing city pages and FAQ
• Week 4: Measure again under the same conditions and compare the trend

Weekly ChatGPT Brand Tracking Checklist and Template

• Are the measurement conditions the same as last week (private window, memory off, web search, language)?
• Is the question set up to date (new products, cities or competitors)?
• Are the branded and non-branded sets kept separate?
• Was every question asked at least 3 times?
• Were 3-5 competitors marked in the same answers?
• Were mention rate, average position, citation rate, share of voice, sentiment and accuracy calculated?
• Was the alert threshold crossed (a drop of more than 10 points in two consecutive measurements)?
• Was the source of every new accuracy error identified?

Google Sheets template

Copy the header row below and paste it into the first row of a new Google Sheets file. Then use Data > Split text to columns with '|' as the separator:

Date | Engine | Question | Stage | Branded (Y/N) | Repetition no | Mention (0/1) | Position | Citation (0/1) | Sentiment | Accuracy | Competitor A | Competitor B | Competitor C | Notes

• Stage: awareness, evaluation or decision
• Mention / Citation: 1 or 0, so averages give rates directly
• Accuracy: correct, or the error type (price, address, feature, area)
• Competitor columns: the competitor's position in the list, or blank

With 90 answers in rows 2-91, the mention rate is =AVERAGE(G2:G91) and the average position is =AVERAGEIF(H2:H91,">0"). Calculate these on the non-branded rows only.

Conclusion

Brand tracking in ChatGPT only works when it is a protocol, not a single test. Use a fixed question set, fixed conditions, at least three repetitions and the same answers for every competitor. Then you can separate real change from noise. Keep branded and non-branded questions apart, track accuracy as its own metric and follow every error back to its source. Use GA4 for the clicks, but don't forget that much of the influence happens without one. If you'd like to see where you stand before building your own sheet, a free AI visibility check with no sign-up gives you a first picture for your industry and city in a few minutes.

Frequently Asked Questions

How can I check for free whether ChatGPT mentions my brand?

Ask 10-20 non-branded questions, three times each, in a private window with memory off, and log every mention in a sheet. This gives you a rough first mention rate at no cost. For a quick first picture without manual work, you can also use a free AI visibility checker that requires no sign-up.

What is the difference between ChatGPT brand tracking and a TÜRKPATENT trademark search?

ChatGPT brand tracking measures how AI assistants mention and describe your brand in their answers. A TÜRKPATENT trademark search (registration monitoring) checks the legal status of a trademark and similar filings in Turkey. The first is a marketing metric and the second is a legal process. They use different tools and involve different teams.

Why does ChatGPT recommend different brands for the same question each time?

Answers are generated probabilistically, and in web-search mode the search queries and the pages read can also change between runs. Login status, memory, location and language add more variation. That is why you should measure rates across repeated answers instead of trusting a single response.

What is ChatGPT shopping research, and how do I track whether my products appear in its recommendations?

ChatGPT offers shopping-oriented experiences that can ask about your needs and build product comparisons or buying guides from web sources. To track them, apply the same protocol to product-category questions with a budget and use case, and log whether your products appear and whether their price and availability are correct. Clear product pages with Product schema, prices and stock information make your products easier to use as sources.

How do I see ChatGPT visitors to my site in Google Analytics?

In GA4, open Reports > Acquisition > Traffic acquisition and filter Session source / medium for 'chatgpt'. You will usually see chatgpt.com / referral, often tagged with utm_source=chatgpt.com. For a cleaner view, create a custom 'AI Assistants' channel with a regex and place it above Referral.

Related reading

<!-- rank-agent-related -->

• Why Doesn't ChatGPT Recommend My Brand? 8 Causes and Fixes <!-- rank-agent-link:8d7217c5-cd4c-41e7-9d3f-4f837ec342ae -->