How to Track Brand Mentions in AI Answers (Free Tracking Template)

To track brand mentions in AI answers, write 20 to 50 questions your buyers really ask, run each one several times in ChatGPT, Perplexity and Google’s AI Mode, and log four things per answer: were you mentioned, were you cited with a link, how were you described, and which competitors appeared. Then add the free first-party data: Search Console’s Generative AI report, Bing’s AI Performance report and AI referral visits in your analytics.

That’s the whole method. The rest of this post is how I run it without fooling myself, because a single screenshot of one AI answer tells you almost nothing.

This is a process guide, not a software roundup. If you’d rather pay for automation, I compare the options in my AI search optimization tools post. For why a mention matters even without a link, see backlinks vs brand mentions.

What Decides How You Track Brand Mentions in AI Answers?

Before you open a single chatbot, answer four questions. Each one changes how much work this is.

1. Where your buyers ask. A local dentist’s patients mostly meet Google’s AI Overviews. A SaaS buyer might compare vendors in ChatGPT or Perplexity. Track those engines first and ignore the rest.
2. How many prompts you need. Under about 50 prompts, a spreadsheet and an hour a month works fine. Past that, manual runs get tedious and error-prone.
3. Whether you have first-party access. If Search Console and Bing Webmaster Tools are verified for your site, you get real impression and citation counts for free. If not, fix that first.
4. Who reads the report. A founder wants one trend line. A marketing team wants per-competitor detail. Build the sheet for the reader.

If you’re a small business with one main engine, run 20 prompts monthly and lean on Search Console. If you sell to researchers who use several assistants, run 40 to 50 prompts across three engines. If neither fits, start with 20 prompts in ChatGPT and grow from there.

Step 1: How Do You Build a Prompt Set That Mirrors Real Buyers?

The prompt set is the whole experiment. Get it wrong and you’ll measure questions nobody asks.

I build mine from three sources. First, the queries already in Search Console, filtered for question words and longer phrases. Second, the questions customers ask on sales calls or in emails. Third, the comparison questions people type when they’re close to buying, such as “best accounting software for a three-person firm” or “is X better than Y”.

Then I sort every prompt into one of three intent groups:

• Category prompts: “best project management tool for agencies”. These show whether you’re in the consideration set.
• Problem prompts: “how do I stop my Shopify store from getting duplicate content”. These show whether your content gets cited as an expert source.
• Brand prompts: “is [your brand] any good” or “[your brand] vs [competitor]”. These show how assistants describe you.

Honestly, most teams overload the brand group because it feels good. Keep it to about a fifth of the set. The category prompts are where new customers find you, or don’t.

Write prompts the way people talk to a chatbot. They’re longer and more specific than Google searches, often with a situation attached. “Which CRM should a two-person real estate team use if we already use Gmail” is a realistic prompt. “Best CRM” isn’t.

Step 2: Run the Prompts by Hand, the Same Way Every Time

Consistency matters more than volume here. My rules for a manual run:

1. Use a fresh chat for every prompt, so earlier answers don’t steer the next one.
2. Turn off memory or personalization where the assistant allows it, or use a logged-out session. Your own history can make your brand look more visible than it is.
3. Note the engine and mode. ChatGPT with search on and Perplexity’s default search can pull different sources, and Google’s AI Mode differs from an AI Overview.
4. Record the date. Answers drift as models and indexes update.
5. Copy the cited URLs, not just the brand names. The URLs tell you which of your pages, or which third-party pages, are doing the work.

Location matters too. If you serve one city or country, run the prompts from there or say the location in the prompt, because assistants often localize answers.

Why Do AI Answers Change Every Time You Ask?

Because these models generate each answer fresh. They sample words with some randomness, and search-connected assistants may retrieve a different set of pages on each run. Ask the same question twice and you can get a different list of brands, in a different order, with a different number of items.

This isn’t a hunch. SparkToro and Gumshoe had 600 volunteers run 12 prompts through ChatGPT, Claude and Google’s AI a combined 2,961 times in November and December 2025. They reported less than a 1 in 100 chance that ChatGPT or Google’s AI would return the same brand list in any two of 100 responses. Identical order was rarer still.

The same study found something more useful. Brands still showed up from a fairly steady consideration set, so how often a brand appears across many runs is a fair metric. My takeaway: never report a “position” in an AI answer. Report a percentage.

In practice, I run every prompt at least three times per engine per month, and five times for the category prompts that matter most. Your visibility rate for a prompt is simply the runs where you appeared, divided by total runs. Three runs is a floor, not a gold standard. More runs give a steadier number.

The AI Mention Tracking Template (Copy and Use)

Here’s the sheet I use. One row per run, not per prompt, so the variance stays visible.

These rows are illustrative, not real data. Paste the CSV below into any spreadsheet to start:

date,engine,prompt_id,intent,prompt,run,mentioned,cited_url,position_in_answer,sentiment,competitors_named,sources_cited,notes
2026-09-15,ChatGPT (search on),P01,Category,best bookkeeping service for Etsy sellers,1,Y,/bookkeeping-for-etsy/,3 of 5,Positive,"BrandA; BrandB","reddit.com; brandA.com",example row

I keep “position in answer” only as context. Given the variance above, I never average it into a score.

Which Metrics Should You Report: Mentions, Citations, Sentiment or Share of Voice?

Four numbers cover almost everything a client asks me. Each answers a different question.

Mentions and citations split more often than people expect. An assistant might name your brand because review sites talk about you, while citing a competitor’s guide for the facts. That gap is a content signal: your brand is known, but your pages aren’t the ones being read.

For sentiment, read the sentence your brand sits in. “Brand X is cheaper but support is slow” is a mixed mention, and I’d log it as neutral with a note. Watch for factual errors too, like an old price or a discontinued service. Those need fixing at the source, which is usually a page on your site or a third-party listing.

AI share of voice is the metric I trust most for trends. It’s relative, so it partly cancels out model-wide swings that affect every brand at once.

How Do You Add First-Party Data From Google, Bing and Analytics?

Prompt runs are a sample you designed. First-party data is what real users actually saw. You want both.

Search Console’s Generative AI report. Google’s help page for the report says it rolled out to all websites as of August 31, 2026. It counts impressions of your links in AI Overviews and AI Mode, by page, country, device and date. It shows impressions only, no clicks or queries, and sites with too few AI impressions may see no data. If the filters are new to you, my Search Console performance report guide walks through them.

Bing Webmaster Tools AI Performance, if your site is verified there. Microsoft launched it in public preview on February 10, 2026. It shows total citations, cited pages and sample grounding queries across Microsoft Copilot, Bing’s AI summaries and some partner integrations. Those grounding queries are gold for your prompt set. Add them.

AI referral traffic in analytics. Visits that come from chatgpt.com or perplexity.ai show up as referrals. OpenAI’s publisher FAQ says ChatGPT adds utm_source=chatgpt.com to links in its search answers, so filter by that source. GA4 also has an AI Assistant channel in its default channel group. Google’s definition names sources like ChatGPT, Gemini, Deepseek, Copilot and Grok, and it doesn’t name Perplexity, so I still check perplexity.ai as a session source by hand.

None of these reports count unlinked mentions. That’s why the manual prompt runs stay in the process.

Your Tracking Decision Matrix

Skip straight here if you just want the setup.

The last row is the one I see most. People guess which assistant their buyers use. Analytics usually settles it in five minutes.

When Should You Stop Tracking by Hand?

Manual tracking breaks down at scale. Once you’re past roughly 50 prompts, several engines and weekly runs, you’re logging hundreds of rows a month, and copy-paste mistakes creep in. That’s the point where a paid tracker earns its fee, and the tools comparison shows which ones fit which budget.

Tracking also only tells you where you stand. If the numbers show competitors named and you absent, the fix is content, entity clarity and reputation work, which is what our AI search optimization service covers. For Perplexity specifically, my guide to Perplexity SEO separates what you can control from the folklore.

Frequently Asked Questions

Can I Track Brand Mentions in ChatGPT for Free?

Yes. Run your prompt set by hand in a fresh chat, log each run in the template above, and filter analytics for utm_source=chatgpt.com. OpenAI provides no report of how often ChatGPT names your brand, so the manual sample is the free option.

How Many Times Should I Run Each Prompt?

At least three times per engine each month, and five for your most important category prompts. AI answers change from run to run, so a single run can’t tell you whether your brand usually appears or just got lucky.

Does Google Search Console Show Brand Mentions in AI Overviews?

Not mentions. The Generative AI report shows impressions when links to your site appear in AI Overviews and AI Mode. If an overview names your brand without linking you, that won’t show up in Search Console. Whether those Overviews cost you clicks is a separate question, and I go through the independent studies in what the data shows about AI Overviews and traffic.

What Is AI Share of Voice?

It’s your brand’s mentions divided by all brand mentions across your prompt set. If assistants named brands 200 times and you 30 times, your share of voice is 15%. It’s a relative measure, so it’s steadier than raw mention counts.

Last updated: September 2026 by Mizanur Rahman