How we score
The AI Brand Index measures fact accuracy with provenance: what AI assistants say about a brand, whether it is true, and the source that settles it. The whole method is disclosed here on purpose.
The questions we ask
We ask every engine the same 3 questions, verbatim (question set v1). {name}is the brand’s name.
Tell me about the brand {name}. Who founded it and when? Where is it headquartered? Who owns it now and who is the CEO? What are its main products and roughly what do they cost?
I'm considering buying from {name}. What should I know about the company before I do — its reputation, any recent major changes or news, and anything controversial?
What has changed at {name} recently — leadership, ownership, pricing, or policies?
The engines, and how many times we ask
We query 6 AI engines, 3 times per question, and label the capture date and engine set on every score. Sampling multiple times reduces run-to-run variance.
| ChatGPT | gpt-4o-search-preview |
| Perplexity | sonar |
| Gemini | gemini-2.5-pro |
| Claude | claude-haiku-4-5 |
| Claude (web search) | claude-haiku-4-5 + web_search |
| Grok | x-ai/grok-4.3 |
What each verdict means
- Correct. The claim matches verified fact.
- Wrong. The claim contradicts a verifiable source.
- Outdated. The claim was true once but a more recent verified fact supersedes it.
- Fabricated. The claim asserts something with no basis in any verifiable source.
- Unverifiable. The claim can be neither confirmed nor refuted from available sources.
- Unverified disagreement. Engines disagree, but we found no source that settles it — so we never call it wrong.
A claim we call Wrong or Fabricatedalways carries a verification source that meets a quality bar, the brand’s own site or an allowlisted authority. If engines disagree but we find no source that settles it, we publish it as an unverified disagreement, never as wrong.
Human review
Nothing appears on this Index until an operator reviews the audit by hand. On-demand checks are never auto-published.
Run the same questions yourself. Ask any of the engines below the exact prompts on this page, on the capture date shown, and compare. The method is fully disclosed on purpose.
What this Index is, and isn’t
A growing cohort with a disclosed method — not a definitive league table.
Disclosure: Jinn sells brand-fact infrastructure — verified, machine-readable Brand DNA that AI engines can read. That is the business this Index measures the need for. We publish verbatim model output, capture dates, and our verification sources so you can check every verdict yourself.