Your standing in the answers AI gives.

Presence, not rankings.

The AIProof Index is presence: the percentage of a round’s answers in which an assistant names your brand, mapped to 0–100. It answers the question SEO never could — when a shopper asks an assistant what to buy, how often are you the answer?

These are the real figures from the sealed round of 2026·09·05: 900 answers, 6 buying categories, 3 engines, 33 brands with any presence.

There are brands in all ten deciles, which is the only real test of a scale: it has to separate the crowd in the middle, not just the leader from the tail.

Being found vs. being chosen.

SEO measures whether you can be found. The Index measures whether you are named when the assistant gives its answer. One is presence in an index of documents; the other is being the recommendation. Different question — different moat.

And it is measured the way a shopper actually asks: neutral buying questions, no brand names in the prompt, five different phrasings of the same need, repeated across three web-grounded engines.

Not share of voice.

Share of voice — your slice of all brand mentions — is the number most AI-visibility tools report. We don’t, and the reason is arithmetic, not taste.

Share of voice is zero-sum. It cannot rise for everyone, and yours rising requires someone else’s falling. An index whose only route to improvement is a competitor getting worse does not measure work you did.

Presence is not zero-sum. If assistants start naming more brands per answer, it rises for everybody — because more brands are in fact being recommended. That is the property we wanted, and it is why the metric is presence even though “market share” is the phrase everyone reaches for.

What this number cannot tell you.

• It is not the badge. The Index says how much of the AI shelf you occupy. The badge says that presence is demonstrably above chance, agreed across engines and across phrasings. A brand can sit at 58 with no badge at all — present, and not yet provably better than chance in any single category.
• It is not a quality score. It counts how often assistants name you. It says nothing about whether the product is good.
• It is not universal. It reads AIProof Index · barefoot footwear · US/EN — six categories, one market, one lexicon. Like Domain Rating, it is “according to our crawl”.
• It has no trend yet. One sealed round is a photograph. The value carries its date and never an arrow; a rise or fall starts existing at the second round.
• It does not explain itself. We can show what the assistants answered. We will not tell you which words to change — advising a brand on how to improve a position we then certify is a conflict we won’t take, at any price.

Published, versioned, reproducible.

An opaque index inside a product whose whole argument is auditability would be the largest contradiction available to us. So here it is:

Rounded to the integer. Presence 0 gives index 0. The unit is the answer, not the mention: an answer that names you twice is not twice the presence.

Why logarithmic

One arithmetic criterion: how many points you gain by doubling your presence, depending on where you start.

On a linear scale a brand already at 45% gains twenty-two times more points than one at 2% for doubling exactly as much: the index would stop measuring work and start measuring prior size. Logarithmic brings that to 1.3×.

Why the anchor is fixed

The anchor is a fixed 90% of presence, not the highest brand observed. With a relative anchor your number would move because someone else moved. Fixed, the Index is yours and only moves if you do.

Three sealed inputs, and nothing else

The round seal, the resolved mentions, and the round verdict. In particular, not the merchant’s plan: the Index does not change because someone paid. Anyone can recompute it: python3 engine/index_score.py --round 2026-09-05. Full specification: AIPROOF-INDEX-v1.0.

What makes a product the answer.

This part is a research question, not a roadmap. Google’s edge wasn’t more content — it was understanding the relationships between documents. The AI shelf may have an equivalent, and it probably isn’t links: some web of corroborating evidence — mentions, citations, agreement across independent sources — that makes a product the recommendation.

We can see correlations in our own data. We are not publishing them as causes, and we are not selling optimisation against them.