How do you track your brand in AI search?

Tracking your brand in AI search means measuring, over time and across engines, whether AI answers name you, cite you, and describe you well. It is not a single check; it is a repeated measurement, because AI answers change between runs and between engines. Done properly, it tells you five things: whether you are cited, whether you are mentioned, how your share of voice compares to competitors, the sentiment of how you are described, and where you sit when an engine lists options. Those five together are the picture, and each can move independently.

What to measure

Five signals make up AI visibility, and they answer different questions.

• Citations are when an engine links to you as a source. They tell you your content is being used.
• Mentions are when an engine names your brand without a link. You can be mentioned without being cited, which still shapes the reader's decision.
• Share of voice is how often you appear relative to competitors on the same questions. It turns raw presence into a competitive read.
• Sentiment is how you are described, favourable, neutral, or critical. Being named negatively is not the same win as being named well.
• Position is where you fall when an engine lists several brands. Being named third reads differently from being named first.

Tracking only "are we mentioned" misses four of the five. The value is in watching them together.

How to do it

There are two practical approaches, and most brands use both.

Manual checks work for a small scope. Choose the questions your customers actually ask, run them on each engine that matters to you, and record whether you are named, which sources are cited, and how you are described. This is the fastest way to get a feel for reality, and everyone should do it at least once. Its limits are real: AI answers vary between runs, so one check is unreliable, and manual testing does not scale past a handful of prompts.

Continuous tracking runs a fixed set of prompts across engines on a schedule and records the results over time. This is what turns anecdote into trend, and it is the only way to cover many prompts and engines without the effort ballooning. Whether you build it or use a monitoring tool, the principle is the same: a stable prompt set, the same engines, a consistent cadence.

The starting point for either is knowing how to check whether you appear in an engine in the first place.

How often, and why numbers disagree

Check more often than you would a traditional ranking. AI answers change between runs, when models update, and as fresh content is retrieved, so a single snapshot is noisy. A stable read comes from the same prompt set checked repeatedly over weeks, watching the direction rather than any one answer.

This is also why two tools can report different AI visibility numbers and both be right: they measure different prompts, engines, and windows, against a non-deterministic system. The fix is not to find the one true number; it is to fix your own method so your numbers are comparable to themselves over time. Consistency with yourself beats agreement with someone else's methodology. For the metric that most often differs, see share of voice in AI search.