How to Track Whether AI Mentions Your Brand (GEO Monitoring)

Why AI visibility is now a metric

A growing share of buyers no longer start with a search results page. They ask an AI assistant "what's the best tool for X?" and act on the two or three names it returns. If your product is one of those names, you get considered. If it is not, you are invisible at the exact moment of intent — and no amount of Google ranking fully compensates, because the buyer never ran a Google search.

That makes "does AI mention my brand?" a real, trackable metric, the same way keyword rankings became a metric fifteen years ago. The difference is that AI answers are non-deterministic, personalized, and change as models update, so you cannot check once and call it done. You have to measure repeatedly and look at the trend. If you want the background on why models behave this way, see how AI answers questions and what GEO is.

What to measure

"Am I mentioned?" is the yes/no version. To actually improve, you want a handful of signals you can track over time. These are the ones that matter most and that a founder can realistically check by hand.

A simple manual method

Before you pay for anything, do this by hand. It takes about an hour to set up and thirty minutes each time you rerun it. The value is in doing it consistently, not in doing it perfectly.

1. Write a fixed prompt set. Pick 8 to 15 questions your ideal customer would actually ask an assistant — the buying questions, not your brand name. Keep them identical every time so results are comparable.
2. Run them across the four major assistants. ChatGPT, Claude, Gemini, and Perplexity each pull from different data and behave differently, so test all of them. Use a fresh chat with no memory or personalization where possible.
3. Log the raw answer. For each prompt on each assistant, record whether you were named, your position in the list, the exact wording used, sentiment, and any sources cited. A spreadsheet with one row per prompt-per-model is enough.
4. Repeat on a schedule. Monthly is a sensible cadence for most founders; move to weekly around a launch. Same prompts, same setup, so the trend is real and not noise.
5. Diff against last time. Look for movement: new mentions, dropped mentions, changed framing, new competitors appearing, corrected or newly introduced inaccuracies.

Here are three example prompts you can paste in as-is and adapt to your category:

What are the best tools for [the specific job your product does]? List a few options and say who each is best for.

I'm a [your ideal customer, e.g. solo founder] trying to [the outcome your product delivers]. What would you recommend and why?

What are the main alternatives to [your closest competitor], and how do they compare?

Follow up with "where did you learn that?" or "what sources support that?" The model will often name the kind of content it is leaning on — community discussions, review sites, or specific pages. That answer is your treasure map.

What tools and signals help

The manual method scales poorly once you are tracking many prompts across four assistants every week. A few things genuinely help, and a few are overrated.

• APIs for batch runs. The OpenAI, Anthropic, Google, and Perplexity APIs let you script your prompt set and log answers automatically. This removes the tedium and makes weekly tracking realistic. Be honest that API answers can differ from the consumer apps.
• Dedicated AI-visibility trackers. A category of tools now monitors brand mentions in AI answers for you. They save time, but treat their numbers as directional — non-determinism means any single score has noise, so watch trends, not decimals.
• Citation signals. For assistants that show sources (Perplexity, Gemini, ChatGPT with search), the cited URLs are the highest-value signal you get. They tell you exactly which pages the model trusts in your category — and therefore where to earn presence.
• Your own logs over time. The most underrated tool is a boring, consistent spreadsheet. Trend data you own beats any one-off snapshot from a dashboard you do not control.

Closing the loop: improve the sources

Monitoring is only half the job. The point of measuring is to change the answer, and you change it by improving the material these models draw on. When you asked "where did you learn that?", the assistants almost certainly pointed at community discussions and third-party pages — and for product recommendations, that overwhelmingly means Reddit and similar peer forums.

So the loop is: find the threads where your buyers ask the questions from your prompt set, participate genuinely where your product is honestly the right answer, and earn mentions in the review pages and roundups the models cite. Then rerun your monitoring and watch whether presence, sentiment, and accuracy move. If a model got a fact wrong, the fix is usually to make the correct information abundant and clear in the sources it reads, not to argue with the model.

This is the same playbook covered in how to get ChatGPT to recommend your product — monitoring just gives you the feedback signal that tells you whether it is working. Measure, improve the sources, rerun, repeat.

Start small, stay consistent

You do not need a full stack to begin. Pick ten buyer prompts, run them across the four assistants this week, and write down what you see. Do it again next month. Within two or three cycles you will have something most founders never build: a real, trending picture of how AI describes your brand — and a concrete list of sources to go improve.

If you would rather not hunt for those source threads by hand, Reddily helps you find the exact Reddit conversations most likely to shape what AI assistants say about your category, so you can turn your monitoring data into action.