How many AI platforms should a benchmark cover?

As many as your customers use. In two measurements across six platforms, more than four in five sources appeared at exactly one platform, so a single-platform benchmark shows a small and unrepresentative slice.

Should I average scores across platforms?

No. Averaging removes the variation that matters. Report each platform separately; the differences between them are the most actionable part of the result.

How often should a benchmark be repeated?

Often enough to distinguish change from noise, with identical questions each time. A single measurement gives a starting point, not a trend.

What should happen to failed measurements?

They should be reported separately, never counted as zero. A platform that was temporarily unreachable is not the same as a platform that did not name you.

Explained · benchmarking method

How to benchmark visibility across multiple AI platforms

Averaging one score across platforms hides the thing you need to see. Platforms disagree, and the disagreement is the finding.

Ask unbranded buying questions to every platform separately and record four levels per answer: retrieval, citation, mention and recommendation. Never average across platforms — in two measurements we found that more than four in five sources appeared at exactly one of six platforms. Repeat with identical questions to separate change from noise.

Four rules for a benchmark that holds

Skip any of these and the numbers stop being comparable across runs.

Keep the brand name out of the question

A question containing your name measures whether the system knows you when pointed at you. Your customer does not type your name; they ask for a solution. Only unbranded questions measure that situation.

Record every platform separately

Do not average. In our measurements a single platform showed between 3% and 46% of all sources found, depending on which one. An average across six platforms conceals exactly the variation you are trying to see.

Separate the four levels

Retrieved, cited, mentioned and recommended are different outcomes. A brand can serve as a source without appearing in the answer — we observed this happening to our own domain. A benchmark that records only mentions cannot show you where the gap is.

Repeat with identical questions

AI answers vary by moment. Without identical repeats you cannot tell improvement from noise. Also record which measurements were unavailable and never count them as zero; that materially changes the result.

What to record for each measurement

These fields make a benchmark reproducible across runs.

Field	Why it matters
Platform	Platforms disagree strongly; averaging hides the pattern
Language	English and Dutch answers pull different sources
Question type	Informational and commercial questions behave differently
Level reached	Retrieval, citation, mention or recommendation
Position	Being named fifth is not being named first
Sources cited	Shows which sites the platform treats as authoritative
Unavailable	Report separately, never as zero

Frequently asked questions

How many AI platforms should a benchmark cover?: As many as your customers use. In two measurements across six platforms, more than four in five sources appeared at exactly one platform, so a single-platform benchmark shows a small and unrepresentative slice.
Should I average scores across platforms?: No. Averaging removes the variation that matters. Report each platform separately; the differences between them are the most actionable part of the result.
How often should a benchmark be repeated?: Often enough to distinguish change from noise, with identical questions each time. A single measurement gives a starting point, not a trend.
What should happen to failed measurements?: They should be reported separately, never counted as zero. A platform that was temporarily unreachable is not the same as a platform that did not name you.

Benchmark your own brand

The free check measures technical AI readiness in about a minute. A full measurement across six platforms starts at €29, one-time.

Run the free check See what a report contains

Figures come from two blind measurements on 29 August 2026, each with eight unbranded buying questions across six AI platforms. See the measurement →

AI Visibility Report Group · Chamber of Commerce 42136638 · VAT NL004345918B24 · company details