See what AI assistants actually recommend — live on a real store. Open the live demo →
Vol. 2026 · Edition I
The verification layer for the AI shelf File · AI-Recommendation Records

AI decides what gets recommended. We make it provable.

AI assistants already shape what shoppers buy. We measure what they recommend across ChatGPT, Perplexity and Gemini, verify it against auditable evidence, and issue it as a signal brands earn — and show.

See if AI already recommends your brand — and what to do if you don't.
https://
No account. If we have already measured your category, you get the record. If we have not, you go on the list and we tell you either way. Earned through evidence, never purchased.
Free category check · badge served free and kept current, always · see how it moves from $39/mo

AI is becoming a discovery channel — a third-party, verifiable signal is trust you can show on the product page.

See the badge live on a demo store → a fictional Shopify shop

How it works A product of AIProof — the evidence layer for AI recommendations.
Sample of the record you'd receive — this could be your brand.
Measured on the instruments shoppers actually ask ChatGPT Perplexity Gemini
§ 01 · The shift
The shift

AI is the new shelf.

When a shopper asks an assistant "what should I buy?", a handful of products make the list — and the rest disappear.

That list is a real distribution channel now. But it is invisible, it changes with every query, and no one has made it verifiable.

The signal that lasts isn't the ranking. It's how often a product is recommended — and that is what we measure, and prove.

The procedure

Measure. Prove. Show.

i
Measure

Across the real engines

We ask neutral buying questions — many phrasings, no brand names — across ChatGPT, Perplexity and Gemini, and record which products each names, and how often.

ii
Prove

Against evidence, not opinion

Every observation is captured with its sources and a timestamp. Frequency is tested against a per-intent null baseline with a conservative bootstrap interval, and source diversity is checked. Reproducible, not asserted.

iii
Show

A signal brands earn

When a product is consistently recommended for a real shopping need, it earns a category record — shown on the brand's own profile, and, through the Shopify app, on its product pages. Every surface links back to the evidence.

The badge

Earned, not bought.

The brand seal is ours. What lands in a store is quieter: a small, neutral badge that adopts your theme and always carries its category, the engines behind it, the date, and a link to the full evidence.

"Frequently recommended by AI for office."
Honest by construction — never "ChatGPT recommends this product."
For the operator

One click in Shopify. Nothing you have to babysit.

01Installs in one click, wears your theme
The badge renders in your store's own styling — light or dark — on the products you map to an earned category. No redesign, no custom code.
02Never in front of your page
The script is deferred, pulls no web fonts, and blocks nothing: your product page paints before ours does. The badge arrives once the page is already up.
03Honest if the signal moves
If a category's 30-day window lapses, the badge shows its date instead of a current claim — it ages, it never vanishes or turns negative on your page.
04Instrumented as little as possible
The badge pings us once per shopper session, and all it carries is your shop domain — no page, no product, no identifier. It tells us the badge is live on your storefront, and nothing else. We do not count impressions, we do not count clicks, and we hold nothing that could later be turned into a person.
Evidence, not hype

We publish the method — not just the conclusion.

Claims are measured across three independent, web-grounded engines, tested against a null baseline, and re-verified monthly. Every record links to the raw, timestamped observations behind it. Anyone can audit them.

3
Independent engines
ChatGPT · Perplexity · Gemini, web-grounded.
Frequency
not ranking
Position is noise; recurrence is the signal.
Bootstrap
confidence interval
It beats a null baseline, or it isn't a claim.
Monthly
re-check
Signal that drifts is revoked, not kept.
For brands

Some brands are the answer AI gives — and don't know it.

01
Know your standing
the categories AI recommends you for today.
02
Prove it
a third-party, auditable claim — cleaner than self-declaring.
03
Show it
a neutral category badge on your product pages and ads.
04
Compare it
see your standing against every measured brand in your category.
The register

A live record of what AI recommends.

Continuously measured
US · English · barefoot-footwear program
Methodology v1.0 · updated 2026·10·04
Category / intentEnginesNull baselineTop frequencyTrendStatusUpdated
hiking Vivobarefoot 3 / 3 28% 95% — ◉ Verified 10·04
everyday minimalist Vivobarefoot 3 / 3 29% 95% — ◉ Verified 10·04
wide feet Lems Shoes 3 / 3 31% 93% — ◉ Verified 10·04
running 3 / 3 41% 92% — Measured · no category yet 10·04
first pair 2 / 3 36% 77% — Measuring 10·04
office / dressy 2 / 3 34% 69% — Measuring 10·04

Each category has its own honest null baseline (chance level for that intent). A record reads Verified only when a brand clears its baseline across engines (with a bootstrap interval) and meets the full C1–C7 bar — enough observations and stability across phrasings. The worked example (Xero Shoes · hiking) reads Verified at 98% across 3/3 engines: it clears the null with room to spare (98% observed vs 28% chance) and holds across phrasings (C4). Like every first-round record it stays on watch for the next monthly re-check (C5) — signal that drifts is revoked, not kept. No brand appears as verified before its evidence does.

§ Colophon

The recommendation you earn, not buy.

Proof of what AI recommends

See what AI recommends in your category.

We're building the verification layer between AI recommendations and commerce — one auditable category at a time. Start with a free check of your own.

https://
Earned, not bought. Every claim links to timestamped, reproducible observations.