Measuring your AI visibility: 6 useful metrics

Why one number is not enough

Because AI visibility has several sides: being cited, being cited early in the answer, carrying weight against competitors, being backed by good sources, being described in the right tone, and having readable pages. A single score flattens all of them.

Each of the six metrics below answers a different question. Their definitions are those of iPand’s methodology; other tools use similar names, but the logic is the same.

One rule applies to all six: read them across many answers, never just one. SparkToro had the same questions put to ChatGPT, Claude and Google nearly three thousand times, and found less than a one-in-a-hundred chance of getting the same list of brands twice^[1]. A metric only means something as a trend.

Definitions from iPand’s methodology^[2].

Visibility: are you cited?

Visibility is the share of collected answers that mention your brand. It is the baseline: if you are not cited, the other five have little to say.

It is calculated over a set of questions, engines and days: for instance, every ChatGPT and Google AI Overviews answer on your tracked questions over the last month. For Meeple Junction, our fictional store in Denver, being cited often on advice questions and never on buying questions would say far more than one overall score.

It is also the metric SparkToro considers the most sensible: a presence rate calculated over dozens or hundreds of repeated questions holds up, where the detail of any single answer does not^[1].

Position: where do you appear in the answer?

Position is the average rank at which your brand is mentioned, calculated only on the answers where it appears. Being named first does not carry the same weight as being named fifth.

Read it with care. SparkToro’s study shows that the order of brands changes from one answer to the next, to the point that its author calls any “ranking position in AI” meaningless^[1]. An average position over many answers is still useful to spot a trend; a position taken from a single answer proves nothing.

Share of voice: how much weight do you carry?

Share of voice compares your mentions with those of every entity of the same type cited in the answers: brands on one side, products on the other. It tells you whether you are gaining ground on competitors, even when your visibility stays flat.

It only makes sense if you know who the others are. A share of voice that rises because a competitor dropped out of the answers is not the same as one that rises because you appear more often. Always look at the list of cited players next to the number.

Citation share: which sources do AI engines rely on?

Citation share measures the weight of each domain among the sources AI engines cite on your questions. It tells you where your visibility is decided: on your site, a directory, a media outlet or a review platform.

It is the most directly useful metric for taking action. If a specialist outlet’s comparison or a forum thread shows up in half the answers, being on that page matters more than rewriting your homepage. iPand’s sources module groups those domains by type, so you can see at a glance whether to work on press, reviews or directories.

Sentiment: what tone is used about you?

Sentiment sums up the tone of the passages about your brand, not of the whole answer. Being described as “pricey but reliable” is not the same as “best avoided”.

Scored on the whole answer, it would mislead you: a positive answer about your sector can still contain one negative line about you. Also look at the recurring themes, such as price, welcome, selection or delays: they are often the themes of your customer reviews, and our article on reviews explains how to shift them.

Technical score: can AI engines read your pages?

The technical score, which iPand’s methodology calls the optimization score, is the average of the criteria that apply to a page, weighted by importance, across the 24 criteria of the framework. It tells you whether an engine can read, understand and reuse your page.

It is the only one of the six you fully control, and the only one that does not measure answers. A good score guarantees no citation; a poor score, on the other hand, can prevent one: a page blocked by robots.txt or invisible without JavaScript will not be cited. Our article on robots.txt and AI crawlers covers the most common case.

How often should you measure?

Often enough to smooth out the variability of answers, and for long enough to see a trend: daily collection read weekly or monthly is a good balance. A one-off monthly check confuses noise and signal.

Keep the same questions from one period to the next, or you are comparing two different things. Also write down the date of every action, such as a new page, a review campaign or a technical fix, so you can link a change in the curve to a cause.

For a first snapshot, the free test asks ChatGPT, Gemini, and Perplexity a few of your customers’ questions. To follow these six metrics over time, iPand’s Visibility module calculates them on your own questions; our comparison of tools covers the other options.

Sources

1. SparkToro — AIs are highly inconsistent when recommending brands — sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/ , accessed October 23, 2026, page dated January 28, 2026
2. iPand — Measurement methodology — ipand.io/methodology