To track brand mentions in AI, you need three things: a fixed list of real prompts, separate tests on ChatGPT, Gemini, and Perplexity, and a log of what each engine says, cites, and recommends. Ranking on Google guarantees none of this. A first-page website can vanish when a user asks "what is the best tool for X?".
Tracking brand mentions in AI is the process of measuring if, how, and with what sources AI answer engines talk about your company. Below are nine checks to trade vanity metrics for verifiable data.
Key takeaways
- AI answers to commercial prompts usually cite only 1 to 3 brands, compared to Google's 10 blue links: whoever is left off the list simply does not exist in that response.
- No pair of AI platforms shares more than 24.1% of cited pages, so testing only on ChatGPT tells you nothing about Perplexity or Gemini.
- A "mention" is the brand name in the response text; a "citation" is the URL or domain in the sources; "referral traffic" is actual clicks coming from the assistant.
- A Bain & Company study indicates that over 80% of users use AI answers in at least 40% of their searches, reducing direct organic traffic by between 15% and 25%.
- In GA4, sources like
chatgpt.com,perplexity.ai, andandroid-app://com.openai.chatgptshow how many of these mentions turn into real visits.
How to track brand mentions in AI across the big three engines?
Start with 20 to 30 prompts written the way your customer asks, run each one on ChatGPT, Gemini, and Perplexity in clean sessions, and log the results in a spreadsheet. This is your baseline. Without it, any "visibility score" is just theater.
1. Build a fixed bank of prompts, both informational and commercial. Why: AI answers questions, not keywords, and only a stable list allows you to compare October with November.
Mix three types. Category prompts ("what is the best software for invoicing for SMBs in Portugal?"), alternative prompts ("alternatives to [competitor]"), and problem prompts ("how to reduce customer payment delays?"). If you operate in multiple markets, write the prompts in the native language of each. A word-for-word translation from English is not the question a customer in São Paulo or Madrid asks. The same intent logic we use to choose keywords that sell applies here: prioritize prompts closest to purchase.
2. Test in an incognito window, without logging in, and repeat each prompt. Why: history, memory, and personalization alter the response, and a single run can be a fluke.
In practice, we run each prompt at least two or three times per engine. If the brand appears in one out of three runs, that is unstable presence, not presence. Note both down.
3. Log each engine in a separate tab. Why: since platforms share at most 24.1% of cited pages, lumping everything into a single number hides where you are strong and where you are invisible.
The minimum spreadsheet columns: prompt, engine, date, brand mentioned (yes/no), position in the list, competitors cited, and source URLs. After covering the big three, it is worth expanding to Google AI Overviews, Microsoft Copilot, and Claude, which also figure in many B2B customer searches.
Does the engine cite your official domain or third-party content?
In most answers, the source the AI uses to talk about you is not your website: it is an industry portal, a Reddit thread, a G2 or Trustpilot page, or Wikipedia. Knowing which source feeds the response shows where to act first.
4. Separate mention from citation in each response. Why: the AI can state your name without linking your domain, and then the credit (and the click) goes to whoever was cited as the source.
A common example: Perplexity recommends your brand, but the three sources listed are a competitor comparison article, a forum, and a review site. You won the mention. You lost the citation. Mark both columns independently.
5. Reverse engineer the domains that appear most often in the sources. Why: if Perplexity and AI Overviews always pull the same five domains in your niche, those are the places where your brand needs to be accurately described.
List all source URLs for the month and count the frequency of each domain. The top of that list becomes your action plan: fix your G2 profile, reply on the right forum, update your Wikidata page, or create the missing page on your site. For your own domain, page structure matters. We explain the format commonly extracted in how to get your first page into an AI-generated overview.
6. Create a referral traffic report in GA4 for AI sources. Why: a mention without a click is reputation; a mention with a click is revenue, and GA4 shows the difference.
- Open Reports > Acquisition > Traffic acquisition.
- Change the dimension to "Session source/medium".
- Filter by regular expression with sources like
chatgpt.com,perplexity.ai,gemini.google.com,copilot.microsoft.com, andandroid-app://com.openai.chatgpt. - Save as an exploration and compare the landing pages with the URLs you noted in the spreadsheet.
If a page appears as a source on Perplexity but gets no visits, the issue might be the title or the cited snippet, not visibility.
Does the AI recommend your solution or list you as a secondary alternative?
Appearing in fourth place, after "there is also", is worth far much less than being the top recommendation. Because AI cites only 1 to 3 brands per commercial prompt, the environment is winner-take-most. Measure position and framing, not just presence.
7. Classify each mention by the role the brand occupies in the response. Why: "recommended" and "cited in passing" count as the same mention on a vanity dashboard, but yield opposite results.
Use a simple scale and apply it consistently every time:
| Classification | How it appears in the response | What to do |
|---|---|---|
| Primary recommendation | First option, with explicit reason | Protect: keep sources updated |
| Listed option | Between 2 and 3 brands, without emphasis | Reinforce differentiators in cited sources |
| Secondary alternative | "Other options include..." at the end | Create clear comparison content |
| Absent, competitor cited | Only rivals appear | Top priority in editorial plan |
| Absent, nobody cited | Generic response, no brands | Opportunity: the space is wide open |
8. Check sentiment and factual accuracy. Why: an AI describing old pricing, a feature that does not exist, or a market where you do not operate is selling poorly on your behalf.
For each mention, note: positive, neutral, or negative, and whether there are any factual errors. Errors almost always stem from an outdated source. Go back to the list in item 5 and find out which one it is. Fixing the origin resolves more than complaining about the model.
Does the AI describe your value proposition with your brand's true voice?
If the AI summarizes your company with words you would never use, the texts it read do not reflect your proposition. Tone alignment is the final filter: it shows whether your positioning reached AI answer engines intact or was diluted by third-party sources.
9. Compare the AI's description with your official messaging, in each language. Why: the way the AI summarizes you is the first impression for many buyers, and it varies by market.
Ask each engine directly: "what is [brand]?" and "who is [brand] for?". Then place it side by side with your website's positioning statement. Look for three mismatches: wrong audience (says "for enterprise" when you focus on SMBs), wrong benefit (talks about price when you sell quality), and wrong category.
Run the same test in Portuguese, Spanish, English, and the languages of your markets. It is common for a brand to be well-described in English and poorly described in Portuguese because local content is an awkward translation or nonexistent. Native content in each language gives the AI a reliable local source to cite. Literal translation rarely does this. This is one of the AI SEO mistakes that cost visibility outside your home market the most.
How often should you repeat each check?
A complete monthly audit of the nine items is the minimum useful rhythm for most B2B brands. The most important commercial prompts deserve weekly checks because answers change when models update or when new sources enter the index.
| Check | Suggested frequency | Approximate time |
|---|---|---|
| Top 10 commercial prompts across 3 engines | Weekly | Short, if the spreadsheet already exists |
| Complete prompt bank | Monthly | Half a day of manual work |
| Source domains (item 5) | Monthly | Along with the complete bank |
| GA4 AI referral report | Monthly | A few minutes, once configured |
| Tone and accuracy by language (items 8 and 9) | Quarterly or after launch | Depends on the number of markets |
Manual work scales poorly. With 30 prompts, 3 engines, 3 runs, and 4 languages, you reach over a thousand responses per cycle. This is the point where most teams give up or move to automated monitoring. What matters is that the tool shows real responses, not just an opaque index. If you have not done the technical part yet, a 7-step SEO audit covers the foundation that AI engines also use to crawl your site.
The 3 non-negotiables for tracking brand mentions in AI starting today
If you only have time for three checks, do the fixed prompt bank, the separation between mention and citation, and the classification of the brand's role in the response. Together, these three items tell you whether the AI knows you, where it got that information from, and whether it recommends your solution.
- Fixed prompt bank by engine and language (items 1 and 3). Without a stable list, there is no trend, only loose screenshots.
- Mention vs. citation (item 4). This is where you find out whether the credit goes to your domain or to a competitor who wrote about you.
- Role in the response (item 7). Being the primary recommendation in 5 prompts is worth more than being a secondary alternative in 50.
Everything else (GA4, sentiment, tone) refines the diagnosis. These three create the diagnosis.
Frequently asked questions
Can you track brand mentions in AI for free?
Yes. A spreadsheet, incognito windows, and the GA4 referral report cover the nine items of this checklist at no tool cost. The limit is time: above a few dozen prompts in multiple languages, the manual process gets slow and prone to logging errors.
Why does my brand appear on Google but not on ChatGPT?
Because ChatGPT synthesizes an answer from multiple sources and cites few brands, while Google lists pages. If reviews, forums, and industry portals do not talk about you, the AI tends to recommend who they cite. In responses without web browsing, the model may also rely on training data older than your current website.
Does ChatGPT show the same response to everyone?
No. Memory turned on, conversation history, location, and whether web search is toggled on change the answer. That is why testing must be done without logging in or in a clean session, and repeated more than once to separate a pattern from a fluke.
Which AI platforms should I monitor besides the big three?
After ChatGPT, Gemini, and Perplexity, the most relevant are Google AI Overviews and AI Mode, Microsoft Copilot, and Claude. Grok and DeepSeek enter the list when your audience uses those assistants, which varies greatly by market.
Build your first prompt spreadsheet this week
Your next step fits in an afternoon: write 20 real prompts, run each one on ChatGPT, Gemini, and Perplexity in an incognito window, and fill in the mention, citation, and response role columns. In a week you will know where you are invisible. No inflated scores. Just real answers.
When the spreadsheet gets too big, we take over the work. UPGeoSEO monitors your AI citations on ChatGPT, Gemini, and Perplexity with transparent metrics, learns your brand tone, and publishes native content in multiple languages directly to your site. You can start with just the SEO + GEO audit, monthly or one-off, or turn on the full autopilot. See how an automated SEO platform works and decide at your own pace.
