AI answers: does your brand show up when customers ask an AI

Why a single measurement is not enough

An answer from a language model is a sample, not a verdict. Ask the same question twice and you may get different answers, and with a single run per question the uncertainty is so large that a green or red marker carries essentially no information at all.

That is why each question is asked several times per round and the result is reported as a frequency with an uncertainty interval. The table shows how much the noise shrinks with the number of runs.

Below three runs no percentage is shown at all, only the count. A number that cannot bear interpretation should not look like a measurement.

Mentioned and cited are two different things

An answer can mention your brand without linking to you, and it can pull text from your website without naming you. The difference decides which action is the right one, so they are kept apart instead of being merged into one number.

• Mentioned and cited: you are there and named as the source.
• Mentioned only: the model knows the brand but did not fetch it from your site.
• Cited only: your text was used without you being mentioned. A content problem, not an awareness problem.
• Searched but passed you over: the model looked and landed somewhere else.
• The model did not search: the question was answered from memory. Then you are measuring awareness, not visibility.

The models behave differently

The result is always reported per model and never as an average, because the differences between them are the whole point. In a measurement on our own site the brand was mentioned in two thirds of the answers from one model and in roughly a tenth from the other, on the same questions over the same period. An average would have hidden exactly what you need to know.

And when the visitor actually clicks?

Showing up in the answer is half the question. The other half is what visitors do when they arrive. Many assume that traffic from AI assistants has higher purchase intent than ordinary search traffic, but that is usually an assumption, not a measurement.

That is why your funnel can be broken down by source. You see how many visits from ChatGPT, Perplexity and the other assistants started the funnel, how many went all the way, and whether the rate differs from other traffic. A difference is only reported when the uncertainty intervals do not overlap. If the data is too thin, it says exactly that instead of showing a percentage that looks certain.

On our own site, the answer today is that visits from AI assistants are too few to settle the question. That is also an answer, and a more honest one than a guess. Links opened inside the AI apps also often lose their referrer, so the count is a floor. That applies to every measurement tool.

What is required

AI answers is an add-on that works with any plan, including the free one, and needs no tracking code on your website. You can have up to ten active monitoring questions. Activation is manual today, so get in touch and we will set it up.

The funnel breakdown by source is part of funnel analysis from Pro and requires the tracking code and the events you want to follow. How many visits come from AI assistants is visible on every plan, including the free one.