Local businesses
50.7
Highest evidence gap. Local sites often explain themselves well, but do not always expose review, directory, and community proof clearly.
60
same businesses rescored
54
completed evidence checks
85
old-style website readiness average stayed high
65
new external proof average was much lower
The change in one picture
Old benchmark
Can AI understand the website?
Clear copy, schema, crawlability, trust pages, FAQs, pricing, and answer-ready content.
+
New benchmark
Can AI find proof elsewhere?
Reviews, Reddit, YouTube, Wikipedia, Wikidata, directories, publisher mentions, and public profiles.
=
Stricter score
Lower scores, clearer fixes
The drop shows evidence gaps, not a sudden business decline.
What changed?
Round 2 was mainly a website-readiness benchmark. It asked whether each business explained itself clearly enough for an AI system to crawl, parse, and reuse.
Round 3 tightened the test. It asks whether public proof outside the website confirms the same story. That matters because AI answers often lean on corroborating evidence before making a recommendation.
Why lower can be better
A lower rescored result is more useful if it tells the business what to fix. A score of 92 with no explanation is vanity. A score of 58 with a clear proof gap gives the team a practical next step.
This is why GetVisus now separates website readiness from external citation-surface strength.
Category comparison
National businesses
73.2
Best average in this rescore. More brands had public profiles, knowledge-base signals, and corroborating evidence.
International businesses
72.3
Strong entity proof, but still visible gaps around reviews, community evidence, and sampled recommendation citations.
The most common proof gaps
| Gap | How often it appeared | Why it matters |
|---|---|---|
| No AI citation in sampled recommendation prompt | 54 of 54 completed checks | AI may know the brand, but not cite it when answering buyer-intent questions. |
| No Reddit/community evidence detected | 50 of 54 | Community discussion helps AI compare real experiences and common objections. |
| No review/category platform evidence detected | 48 of 54 | Reviews and category profiles are easy verification surfaces for recommendation systems. |
| No news/publisher evidence detected | 40 of 54 | Publisher mentions can support authority and reduce uncertainty. |
| No Wikipedia or Wikidata evidence detected | 9 of 54 | Entity databases help AI confirm what the business is and how it should be classified. |
Plain-English version: a website can be clear, but AI still wants backup. If the backup is missing, scattered, or inconsistent, the business has a recommendation opportunity.
What businesses should fix first
Create a proof hubLink review profilesAdd sameAs schemaClarify entity factsPublish useful videosEarn category mentions
The practical fix is not "write more SEO content". It is to make the business easier to verify: who it is, what it does, who trusts it, where it is discussed, and why it belongs in a recommendation.
Use the data
The full report includes the rescored CSV, JSON, category averages, evidence gaps, outreach angles, and score movement explanations.
Disclaimer: this is a directional benchmark based on a controlled sample, not a complete market-wide study, ranking guarantee, or proof of exact commercial causation. The OpenAI sampled recommendation step hit provider quota limits during this run, so the strongest use of this rescore is website readiness, external evidence surfaces, and score movement explanations.