A single mention tells you almost nothing. A mention tracked over time tells you whether the brand is becoming easier to find, harder to find, or staying the same, and whether the facts attached to the name are getting better or worse. Tracking is the discipline of asking the same questions, in the same way, on the same schedule, and writing down the same fields every time. Without that discipline, every check is a fresh anecdote, and anecdotes cannot be compared.
No vendor publishes a mention tracking tool or a mention benchmark. None is claimed here. What follows is a manual routine, built from asking the assistants directly and logging what they say, with the limits stated plainly.
1. Build the prompt set first
The prompt set is the measurement instrument. Build it before you track anything, and do not change it between runs unless the business changed.
- Write the questions your buyers actually ask. A buyer asks for the best trail shoe for wet weather, or a coffee grinder for espresso under a budget, or a gift for a new parent. Write twenty to thirty of those questions, covering your main products, your main categories, and your main buyer problems.
- Add a small set of branded prompts: questions that name your brand directly, and questions that ask for alternatives to your brand. The branded set tells you whether the assistant knows you at all. The unbranded set tells you whether it recommends you when it has a choice.
- Add a small set of comparison prompts: questions that ask for the best option in a category, or for a comparison between two product types. These are the prompts where a brand is most often named alongside competitors, and they are the prompts where a missing mention costs the most.
- Write each prompt down exactly as you will run it, word for word, in a file the whole team can see. Do not improvise the wording at run time. Improvised wording changes the measurement, and you will not know whether the brand moved or the questions moved.
- Keep the prompt set small enough to run in one sitting. Twenty to thirty prompts across four or five assistants is a real afternoon of work. A set of two hundred prompts is a set that will not get run twice.
2. Pick the assistants and the schedule
- Track the assistants your buyers use: ChatGPT, Gemini, Perplexity, Claude, and Copilot. Those five cover the assistants a buyer is most likely to ask. If your buyers use one of them far more than the others, weight your attention accordingly, but do not drop the others entirely, because a brand that is invisible on one assistant is invisible to that assistant's users.
- Run the set on a fixed schedule. Monthly is enough for most stores. Weekly is better when you are launching a product, fixing a problem, or watching a competitor. Pick one and keep it. A monthly run done every month is worth more than a weekly run done twice and then abandoned.
- Run the prompts in the same order, on the same assistants, each time. Run them in a fresh session where the assistant allows it, so the previous question does not shape the next answer. Do not run the branded prompts first and the unbranded prompts second in a way that lets the brand name leak into the unbranded answers. Run the unbranded set first.
- Do not run the set once and judge from one run. Answers change from run to run, even with the same prompt. One run is a snapshot. Three runs is a trend.
3. Log the same fields every time
A log with the same fields every time is a measurement. A memory of what you saw last time is not.
- The assistant, the prompt, and the date of the run.
- Whether your brand was named at all.
- Whether your brand was recommended first, named among others, or named only as an also-ran.
- Which pages the assistant cited, if it cited any.
- Whether the facts it stated about your products were right or wrong, and which fact was wrong.
- Whether a competitor was named, and which competitor, so the log shows the comparison, not just your own brand in isolation.
Keep the log in a spreadsheet, with one row per prompt per assistant per run. That structure is what lets you compare runs: filter by assistant, filter by prompt, and watch the share move. A log kept in notes or in memory cannot be filtered, and it cannot be compared.
4. Read the trend, not the run
- Watch the share of prompts where your brand was named, by assistant, over time. A share rising from one in ten to three in ten is progress, even if the absolute number is still small.
- Watch the share where your brand was recommended first, separately from the share where it was merely named. A brand named last among five is not the same as a brand recommended first, and the two shares can move in different directions.
- Watch the share of mentions with wrong facts. A brand with many mentions and many wrong facts has a different problem from a brand with no mentions at all. The fix for wrong facts is at the source: the page, the markup, or the feed. The fix for missing mentions is at the gate: the crawl, the index, or the content.
- Watch which pages get cited. If the assistant cites your guide but never your product pages, the guide is doing the work and the product pages are not. If it cites a competitor's product pages and your guide, the competitor's product pages are doing work yours are not.
- Watch the competitors in the same log. If a competitor appears in prompts where you do not, compare the two sites: is the competitor indexed, is its product information in the first wave of HTML, is its profile complete, and is its product data in agreement across the page, the markup, and the feed. Fix the gap the comparison shows, then retest with the same prompts after the fix has been recrawled.
5. What to do when the trend moves
- When mentions rise after a fix, do not assume the fix caused the rise until you have two runs that agree. One run can be noise. Two runs in the same direction is a signal.
- When mentions fall and the store's pages have not changed, check what changed on the other side: a competitor's pages, a crawl problem, a stale fact, or a page that fell out of the index. Check the index first, because a page that fell out of the index cannot be cited, and the fix is a recrawl, not new content.
- When wrong facts appear, trace each one to the source before fixing anything. Check the page, the markup, and the feed for that product, and check whether the assistant cited a page that is out of date. Fix the source, ask for a recrawl in Search Console, and retest with the same prompts after the recrawl. Check the vendor docs in Sources for the current path to the recrawl request.
- When a new product launches, add its prompts to the set and run the set more often for a while. A launch is the one time a weekly run earns its keep, because the first weeks are when the index and the assistants are forming their first impression of the new product.
The fundamentals
These mechanics are shared across every engine and covered in full in their own guides:
- •The llms.txt format and a complete sample, see our guide to writing a good llms.txt file
- •The markup this engine reads, see our product schema guide
- •The complete crawler rule reference, see our robots.txt guide for AI search
Say you run a running shoe brand
Say you sell running shoes, and you start tracking mentions monthly. In the first run, ChatGPT names your brand in four of twenty prompts, and recommends it first in one. Perplexity names it in six, and cites your trail shoe guide twice. Claude does not name it at all. The facts check finds one wrong fact: Perplexity states a price from before a sale ended. You fix the feed, ask for a recrawl, and run the same prompts a month later. ChatGPT names you in six, Perplexity in seven, and the wrong fact is gone. In the third run, Claude names you in two prompts for the first time, citing the guide. You did not buy a tracking tool. You asked the same questions, logged the same fields, and watched the direction over three runs. The trend is the measurement, and the trend is up.
What this guide does not cover
This guide covers how to track mentions over time. It does not cover how to measure visibility in full, which is our guide to measuring your brand's AI visibility, or how to run a full audit, which is our guide to running an AI visibility audit on your store, or how to benchmark against competitors, which is our guide to benchmarking AI visibility against competitors. Tracking tells you the direction. The other guides tell you what to do about it.
Check your own brand now
Run the free scan. It takes seconds, no signup, and shows whether AI crawlers can read your pages today: aizii.ai/aeo-scan
Run the free scanSources
- Google Search Central, Optimizing your website for generative AI features on Google Search (specific): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Google Search Central, AI features and your website (specific): https://developers.google.com/search/docs/appearance/ai-features
- Google Search Central, Creating helpful, reliable, people-first content (specific): https://developers.google.com/search/docs/fundamentals/creating-helpful-content
- OpenAI, Bots documentation (specific): https://platform.openai.com/docs/bots
- Perplexity, Bots documentation (specific): https://docs.perplexity.ai/guides/bots
Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
