To monitor what AI is saying about your brand, pick 10 to 20 questions buyers ask, put them to each engine on a schedule, and keep every answer word for word. Each time, log whether you were named, whether the claims are true, which pages were cited, and who was recommended. A spreadsheet works for a handful of questions. Past that, you'll want a tracker.
A one-off check is a different job. Our brand mention audit covers that one, and this post picks up after it, with what to do every week so the audit doesn't go stale.
Why your existing brand monitoring misses this
I asked ChatGPT this exact question on 2 November through its API. The sources it returned were google.com/alerts, brandwatch.com, mention.com, talkwalker.com and awario.com, all bare homepages. So the top answer to "how do I monitor what AI says about me" pointed at a set of brands best known for monitoring the web and social media.
Those tools watch for your name on pages and posts as they are published. An AI answer doesn't get published anywhere. It's generated when someone asks, shown to that person, and gone, so a crawler has nothing to find. Some of these vendors have added AI products: Brand24, for one, has an LLM monitoring page. If you already pay for one of them, ask whether it puts questions to the engines or only watches the web.
What to log each time
| Field | What you're checking | Why it matters |
|---|---|---|
| Named | Does the answer use your brand name? | The basic presence count |
| Accuracy | Are the price, features and audience right? | A wrong answer can hurt more than no answer |
| Cited pages | Which URLs did the engine show? | This is where a wrong claim can be traced and fixed |
| Recommended | Who did it tell the reader to choose? | Being listed and being picked are different things |
Keep the full answer too. Six weeks from now you'll want to read how the wording changed, and a yes/no column can't tell you that.
Picking the questions
Most of the list should be unbranded, like "what's the best tool for X", because that's where a buyer who hasn't heard of you shows up. Add a few branded ones ("is [brand] any good for X?", "[brand] pricing") since those are where wrong facts tend to do the most damage: the person is already considering you. We have a longer method for building the list.
Then check the languages you sell in. My Search Console shows the same question arriving in French, Portuguese, Italian and Czech ("existe-t-il une application qui montre comment ChatGPT décrit nos produits..."). I put the French version to Perplexity. Judging by their URLs, the 15 sources it returned were overwhelmingly English-language pages, and none of the three tools I was looking for got named, even though all three were named when I asked Perplexity a similar question in English. That's one answer, so it proves little. It's enough to make me test each language separately instead of assuming the English result carries over.
How to read changes without overreacting
Answers move with nothing changing on your side. If you were named last Tuesday and not this Tuesday, that's a thing to watch. If you've dropped out of the same question for several runs across more than one engine, that's a change worth investigating. I'd only start looking for a cause at that point.
The order I'd follow when something is wrong:
- Read the cited pages and find the sentence the claim came from.
- If it's your page, fix it. If it's someone else's, send them the correction with a source.
- State the true version plainly on your own relevant page, in text a crawler can read.
- Re-ask the same questions over the next few runs, and write down the date you made the fix.
You won't be able to prove the fix caused the change, since the engines and the other pages are moving at the same time. The dated note still tells you what to look at.
Doing it by hand, or not
By hand, 10 questions on 3 engines is 30 answers a week to run and paste, which is the kind of chore that doesn't survive a busy week. A tracker's real value is that the run keeps happening when you're busy.
CueScout does this: it keeps each answer word for word, labels its sentiment, and stores the pages the engine read. Answers that change are re-asked every day and steady ones at least once a week. Other trackers do similar work, and our comparison has the engine lists and prices. To see one answer first, the free AI Visibility Checker needs no signup.
Frequently asked questions
How can I monitor what AI is saying about my brand?
Choose 10 to 20 questions buyers ask, including a few that name your brand. Put them to each AI engine on a schedule and keep the full answers. For each answer, record whether you were named, whether the claims are accurate, which pages were cited, and who was recommended. You can do this by hand in a spreadsheet or use an AI visibility tracker.
Can Google Alerts or a social listening tool monitor AI answers?
Google Alerts can't, because it watches for newly published web pages and an AI answer isn't a published page. Social listening tools built around web and social mentions have the same limit, although some vendors now sell a separate module that queries AI engines. Check that the product actually asks the engines rather than only watching the web.
How often should I check what AI says about my brand?
Weekly is enough by hand for most small teams. A tracker can run more often, which helps separate a real change from day-to-day movement. Don't act on a single answer.
What should I do if AI says something wrong about my brand?
Read the pages cited in that answer, because the wrong claim often comes from one of them. Correct your own page if the error started there, and ask the publisher to update it if it didn't. Then state the correct fact plainly on your site and re-ask the same questions over later runs.
Do I need to monitor AI answers in other languages?
If you sell in those languages, yes. A question in French may draw on different pages and name different brands than the English version, so it needs its own place on your list.