How Infera measures AI visibility

What a trial is

The smallest unit of a measurement is a trial: one question, put to one model, one answer. Only many trials add up to a reliable statement — a single answer is a random event, not a measurement.

A run therefore consists of the full cross product: every topic cluster, every prompt, every model, several repetitions.

Why we ask more than once

Generative models do not answer deterministically. The same question can name different providers the second time round. Ask once and you measure noise.

Infera puts every question repeatedly and forms the share across all answers. That is why mention coverage is a percentage and not a yes/no statement: “named in 34 % of answers” is a measurement, “is named” would be a claim.

Mention coverage and citation coverage

Both metrics use the same denominator — all answers of a run — but measure different things:

• Mention coverage: share of answers in which the brand is named. Counts mentions without a source too.
• Citation coverage: share of answers in which the brand’s own domain is cited as a source.

Why the gap between them matters

Citation coverage is almost always well below mention coverage. That is normal — it becomes notable when the gap is large and competitor domains supply the citations. Then the competition shapes the picture of your own brand.

That is why Infera reports both values separately instead of folding them into one score. A combined index would hide exactly the information that matters.

Placement, not just mention

Being named is not the same as being recommended. Infera additionally records where a brand sits in the answer: as the top pick, within the first three, or somewhere in a list.

Region and language are part of the measurement

AI answers differ by language and region. A measurement without a fixed region and language is not reproducible.

Every brand therefore has a default region and a default language; individual topic clusters can deviate where a market requires it. Both values appear in the report.

How the bands are cut

Percentages alone say little if you do not know the market. Infera therefore sorts mention coverage into fixed bands — the same thresholds for every brand and every industry, so values stay comparable:

• very weak: below 8 %
• weak: 8 – 19 %
• moderate: 20 – 39 %
• good: 40 – 59 %
• strong: 60 % and above

Where the prompts come from

The questions are not maintained by hand. Infera derives them from the brand: your site yields a brand analysis, that yields topic clusters and subtopics, and those yield decision questions — classified by the intent behind them, from information seeking to purchase intent.

That is deliberate: choose your own prompts and you end up measuring your own assumptions.

What Infera cannot measure

A scan shows what AI systems answered during the measurement period. It does not say how many people actually asked those questions, and it does not explain model behaviour causally — which signals trigger a mention can be reasoned from externally observed answers, not proven.

Models also change. A value from today is a snapshot; statements about development need several runs over time.