Brand Monitoring

Brand monitoring is the practice of tracking where and how a business is named, historically across news, social platforms, forums and review sites. AI assistants added a channel the established tooling cannot see, because the thing being monitored is not published anywhere: it is generated, once, in reply to a question, and then it is gone.

Why the usual method does not work here

Conventional brand monitoring works by crawling published material and matching a name against it. The material sits at a URL, it is the same for everyone who loads it, and it stays there, so the job is coverage and deduplication.

None of that holds inside an assistant. The answer exists only in a conversation, it is composed for the person who asked, and asking the same question again can produce a different reply. There is no page to crawl. A monitor that relies on finding published text will report that nothing is being said about the business, which is not the same as nothing being said.

What monitoring the channel requires instead

If the text cannot be found, it has to be produced deliberately: put a fixed set of questions to an assistant at set points and record what comes back. That turns an unobservable channel into a sampled one, with the properties of a sample: it is only as representative as the question set, and a single answer tells you much less than the pattern across a set of questions.

This is what a visibility score is built from, and why the question panel is frozen between readings. Changing the questions between two measurements makes the difference between them uninterpretable.

One question people ask

Can this be done by just asking ChatGPT about the business occasionally?

It can be done that way, and it is worth doing once for the shock value, but it is not a measurement. Assistants vary from run to run, so a single answer is a single draw from a distribution, flattering or alarming largely by chance. What makes a reading comparable to the next one is a fixed question set, the same scoring both times, and a result read across the whole set rather than from any one answer. Without those three, a change from one reading to the next cannot be told apart from noise.

Sources

The panel and the scoring referred to here are this site’s own method, described on the methodology page. No third-party figures are used on this page.