How do I track brand mentions in AI answers?
Run a fixed set of buyer questions across several engines on a schedule, and record citations apart from mentions. The method, the traps, and what to do with the result.
Run a fixed set of buyer questions across several engines on a schedule, and record who is named in each answer. Keep citations and mentions apart, keep the questions still, and the result becomes a trend rather than an anecdote.
What does the method look like in practice?
Four steps. Source ten to fifteen questions from your own search data. Put each to every engine you care about. Record whether your domain appears in the sources, whether your brand appears in the prose, and who else was named. Repeat weekly.
The discipline that matters is not changing the questions. A moving question set makes the trend meaningless.
What are the traps?
Three, and all three are common. Putting your brand in the prompt, which guarantees a mention and measures nothing. Counting mentions as citations, which flatters the number. And running once, which gives you a data point with nothing to compare it to.
A fourth, quieter one: sampling a single engine and generalising from it.
What do I do with the competitor list?
Read it as a content brief. The domains named instead of you are the pages the engines currently trust on your topic, which tells you both what the answer looks like and what you are missing.
It is usually more actionable than your own score, because it points at specific pages.
Does this need to be automated?
Eventually, yes. Doing it by hand is fine for one run and unsustainable weekly, and the value is entirely in the repetition.
The cost is a few model calls per run, which is small enough that frequency is a choice rather than a budget question.
Questions
Can Google Alerts do this?
No. Alerts watch published web pages, not the answers assistants generate, and those answers are not published anywhere to be crawled.
How many questions do I need?
Ten to fifteen is the working range. Fewer makes every single citation swing the share; many more costs proportionally and rarely changes the conclusion.
Should I track how the brand is described?
Yes, and it is underrated. How an engine characterises you is often more useful than whether it named you, because a wrong description is a fixable content problem.
Find out where you stand on your own domain: the free AI visibility check asks all four engines about you, and the content scorer scores any page against the published rubric. Neither needs an account. The full background is what answer engine optimization is.