Much is claimed about visibility in AI answers, and little is measured. That's no coincidence: answers differ by session, by wording and by platform, and in that noise everyone can find support for their own view. The solution isn't to measure more, but to first agree what you're measuring. The method in four steps:

1. Fix a set list of search queries that your target customer actually asks, before you measure.
2. Agree on what counts as visible: mentioned as a source, linked, or recommended as a party.
3. Run the baseline measurement: every query through every platform, with the date and method noted.
4. Remeasure monthly with the same queries; the difference from the baseline measurement is your proof.

The sections below work through each step.

01 / the yardstickFix the queries before you measure

The yardstick is a fixed list of search queries that your target customer actually asks, set before the first measurement runs. Twenty to thirty queries is workable: enough to see a pattern, few enough to keep up monthly.

Also fix what counts as visible. Are you mentioned as a source with a link, mentioned without a link, or recommended as a party. Those are three different outcomes, and mixing them up means comparing apples with pears next month. The queries themselves come from conversations with customers, from your inbox and from the search terms your site is already shown for.

02 / the baseline measurementRecording the initial state

The baseline measurement is the first complete round: every query through every platform, noting what happens for each combination. Do you appear, who does appear, and with which page. Note the date and method too, because only then can a later measurement be fairly compared.

For each answer, note three things: whether your business appears, which three parties do get mentioned, and which page of those parties is cited. That third point is the most instructive, because it shows what kind of page carries the answer: a knowledge article, a comparison, a review page. That's where your improvement list lies.

Expect a confronting outcome. Anyone doing this for the first time often turns out to be visible on virtually none of the queries, while a handful of competitors and comparison sites keep coming back. That's not bad news but a starting point: now you know for certain, instead of merely suspecting it. How to do a first check like this yourself in ten minutes is in how do you check whether your business appears in ChatGPT.

03 / the remeasurementThe delta is the proof

Every month, the same measurement runs again: the same queries, the same platforms, the same definition of visible. The only number that matters afterwards is the delta: on how many queries are you now visible where that wasn't the case before.

Without a yardstick fixed in advance, any result can be defended after the fact, and is therefore worth nothing.

The delta forces honesty in two directions. If something moves, it's there in black and white and it's not a sales pitch. If nothing moves, that too is visible, and then the right question is why, instead of it staying hidden behind isolated success stories.

If you work with an agency, this is also your control instrument. Before starting, ask for the yardstick and the baseline measurement, and afterwards for the remeasurement in every monthly report. A party willing to do that has nothing to hide; a party that avoids it is asking you to pay for something you will never be able to check.

04 / the meansTools or manual work

There are tools that automate this work, such as Otterly and Peec AI: they run your queries periodically through multiple platforms and keep track of the score. For anyone tracking many queries or multiple brands, that's the logical route. Outsourcing is also an option; what such an outsourced baseline measurement looks like is in what an AI visibility scan is.

Manual work also works, and for a first stretch it's often enough: a spreadsheet with the queries in the rows, the platforms in the columns, and one fixed moment per month when you run the round. For 25 queries across four platforms, count on half a day per month.

The recommendation for anyone starting out: do the manual work for the first few months. The method determines the value of the measurement, not the tool, and anyone tired of the manual work after three months will by then know exactly what a tool should do for them.

05 / the limitWhat this measurement can't do

A snapshot remains a snapshot. AI answers vary by session, so a single isolated measurement proves little; the pattern over months is what counts. So measure fixed queries at a fixed moment, and don't draw conclusions from one deviating round.

And the measurement says nothing about revenue. Being visible in an answer is not the same as being called. Anyone who wants to know what visibility delivers should also record a traffic and enquiry figure alongside this measurement. For businesses without searching customers, this whole instrument is unnecessary, and that's a valid conclusion too.

This text was produced with AI assistance and checked and approved by a human before publication.