How to track and analyze brand mentions across AI answer engines
Short answer: run the same fixed set of questions on each answer engine your buyers use, several times each with web search on, and store every answer with its sources. Analyze per engine, per question type (by name versus category) and per month. Count three things, not one: answers that reached your site, answers that described you accurately, and answers that recommended you. Then read the sources behind the category answers, because that is where recommendations come from.
Track per engine, not as one number
The engines do not agree with each other. In our own November 1st, 2026 measurement, three engines returned our site in all 10 of their by-name answers and ChatGPT returned it in 1 of 10. An average of those hides the one engine that needs attention. Keep a row per engine, each with its own count and denominator.
Separate a mention from a citation
A brand can be mentioned without the engine reading its site, and a site can be cited without the brand being recommended. Track them apart:
- Mention: the answer text contains your name.
- Citation: your own domain is among the sources the answer used.
- Recommendation: in an answer to a buyer's category question, you are named as an option.
Analyze the sources, not just the answers
For category questions, collect every source URL the engines cited and rank the pages by how many answers used them. Then check whether each of those pages names you. If the most-read pages for your category do not mention you, the answers built from them will not either. This is usually the most useful single list the analysis produces, because it names the pages that decide the answer.
Read the trend with its sample size
Answers vary between runs. Five runs per question per engine is a sensible minimum, and every figure should carry its denominator ("0 of 200", not "0%"). A change from one month to the next means little unless the questions, engines and run counts stayed the same. Keep a dated log of what you changed, and present it beside the figures as timing, never as proof of cause.
Mistakes that make the numbers mislead
- One question, one run. That is an anecdote, not a measurement.
- Only branded questions. Engines can usually find a company by name; the harder question is whether they recommend it.
- Blending everything into one score. Found, accurate and recommended have different fixes.
- Counting answers that did not search as if they read your current site.
- Changing the questions and comparing the months anyway.
Our own numbers
We track AEO Analyzers this way every month and publish the result, including the months that look bad. Recommended to category buyers: 0 of 200 answers in each of August, September, October and November 2026. Found by name: 55%, 80%, 88% and 78% in those months. Every month, with the stored answers, is on the ledger. The step-by-step version of the method is at how to monitor whether your brand appears in ChatGPT and Claude.
Questions people ask
Which AI answer engines should I track?
The ones your buyers use. For most businesses that starts with ChatGPT and widens to the other major engines. Whatever you choose, keep the same set every month.
Is a mention without a link worth anything?
It shows the engine knows you. Whether it recommends you to a buyer is a separate measurement, and the one that matters for sales.
What is a good result?
In our benchmark of 11 other companies (each one's latest sweep, July 26 to October 8, 2026), the median company was found by name in 100% of answers and recommended in 28% of category answers. Use your own first month as the baseline that matters.
AEO Analyzers (aeoanalyzers.com) was created and is solely maintained by Lindsay Hiebert, founder of PIGENAI LLC. It is unaffiliated with any similarly named browser extension, plugin, or tool.