How to Track Brand Mentions in ChatGPT and AI Answers

What counts as a brand mention in an AI answer?

A brand mention is any answer that names your brand in its text, whatever the spelling. Agree the rules below before your first run, write them at the top of the sheet, and keep them fixed for the whole period.

• Count variants. Your brand name, common misspellings, your domain (yourbrand.in) and your product line name all count. List the variants in the sheet header so everyone matches the same ones.
• Skip look-alikes. A word that matches your name but is used in its ordinary meaning does not count. A brand called Simple and the word simple are different things.
• Position by first appearance. Number the brands in the order they first appear. If the answer names five brands and you come third, write 3. Count only brands, and leave out retailers and marketplaces unless you compete with them.
• Sentiment by the sentence. Positive if the answer recommends or praises you, negative if it warns buyers off or leads with a drawback, neutral for a plain listing. Paste the sentence so anyone can check the call.
• Citations in their own column. Cited sites are the source links an engine shows. A site can be cited without naming you, and you can be named with no link to your site. Log both. How to get cited by AI shows what to do with the list.
• No Overview is a result. Google does not show an AI Overview for every search. Write No Overview, leave those rows out of your mention share, and count separately how often an Overview appears at all.

How do you set up the tracking sheet this week?

Plan about half a day of work for the first round, spread over three days. The steps below assume one person and a spreadsheet, and they use ChatGPT, Gemini, Perplexity and Google as the engines.

• Day 1, write the questions. List 10 to 20 questions a buyer asks before choosing a product like yours. Take the wording from reviews, support chats, marketplace Q&A and sales calls. Leave your brand name out: an answer to a question that names you almost always names you back, which inflates your share.
• Day 1, add Hinglish versions. If you sell in India, write a Hinglish version of your 5 most important questions. Hinglish is Hindi and English typed in English letters. Give each pair the same row ID.
• Day 2, run the questions. Ask each question 3 times per chat engine, each time in a new chat, logged out or with memory off. Search Google in a private window for the AI Overview. 15 questions on 3 chat engines with 3 runs each come to 135 answers, a little over 2 hours at a minute an answer.
• Day 3, roll up. Use the formulas further down to get your mention share per engine, your average position and your most cited sites. That first result is your baseline.
• Every week or month, repeat. Run the same questions on the same weekday. Rewording a question starts a new trend line, so add a new row ID and keep the old one.

Why do you need several answers per question?

AI engines rarely give the same answer twice. In our August 2026 study of 480 answers from India, two ChatGPT answers to the same question, asked within one hour, shared 54.7% of the brands they named, and two Perplexity answers shared 68.3%. A single run can show you first or absent by chance. That is why this method asks each question 3 times and adds up the results across questions: 15 questions give 45 answers per engine per round. Does ChatGPT give the same answer to everyone? explains why, and how to measure AI visibility covers how many answers you need.

How to track brand performance in ChatGPT responses over time

To track brand performance in ChatGPT responses over time, freeze your questions, ask them on a fixed schedule, and compare the share of answers that name you in one period with the period before. Report every share with the number of answers behind it, because a small sample moves a lot on its own.

A 95% confidence range shows how far a share can move by chance: it is the band the true share most likely sits in. The table shows how the range narrows as you add answers, for a brand named in half of them.

How do you track Hinglish prompts next to English ones?

Give each Hinglish question the same row ID as its English twin, set the Language column, and report the two languages side by side. A single blended share hides differences between engines.

Those differences can be large. In our English vs Hinglish study, we asked 10 buying questions in both languages, 8 times each, on ChatGPT, Gemini and Perplexity from inside India on 14 Aug 2026. We compared the share of brands two answers have in common. On ChatGPT, that share was 2.2 percentage points lower for an English and a Hinglish answer than for two repeat answers in the same language. The gap was 7.8 points on Gemini and 23.4 points on Perplexity. Gemini attached sources to 90.0% of English answers and 41.3% of Hinglish answers. In the October 2026 re-run (479 answers), the gap was 0.8 points on ChatGPT, 2.0 on Gemini, no longer significant, and 28.2 on Perplexity, and Gemini attached sources to 73.4% of English and 56.3% of Hinglish answers.

• Write Hinglish the way buyers type it. Use English letters, everyday words and English product nouns. "India mein oily skin ke liye 500 rupees ke andar konsa face wash kharidna chahiye?" reads like a real buyer. A word-for-word translation usually does not.
• Log the answer language too. Engines usually reply in the language you used. In our August 2026 study ChatGPT answered all 80 Hinglish questions in Hinglish, while Gemini answered 8 of 80 in Hindi script.
• Have a native speaker check each pair. In our study, 3 of 10 question pairs failed an independent review and were rewritten before any answers were counted.

When should you switch from a spreadsheet to an AI brand monitoring tool?

Switch when filling the sheet takes more time than acting on it. Daily tracking of 20 questions on 3 chat engines is 60 answers a day before any repeats, about 1,800 a month, which nobody scores by hand. The usual signs: you skip runs, two people score the same answer differently, or you want daily numbers.

LLM brand monitoring tools, also sold as AI brand monitoring or AI visibility tools, run the questions for you, store the answers and do the scoring. They differ on engines, countries, languages and price. The best AI visibility tools compares them.