How the Index Is Made
The measurement pipeline
- Prompts: 38 buyer-intent questions of the kind real customers ask (purchase, comparison, research, local, review, emergency framings).
- Engines: ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Grok (xAI), Perplexity — queried in parallel with per-engine logging.
- Passes: 2 per prompt per engine (AI responses vary between runs; two passes reduce single-run noise) = 380 planned calls.
- Counting: responses parsed for mentions against the pre-registered 46-brand list; 1,103 mentions logged in Edition 01; per-engine call/success counts recorded, including failures.
- Scoring: the AIR Score (a trademark of Deep AI Solutions Inc) composites mention, recommendation, and sentiment 0-100 for individual businesses.
What the index does NOT measure
- Not service quality. AI visibility is mindshare inside AI answers — a marketing measurement, full stop.
- Not a permanent state. Engines re-ground constantly; every edition is a dated snapshot (Edition 01: 2026-07-09) and stale claims from old editions shouldn't outlive their date.
- Not exhaustive. 38 prompts and 46 brands is a designed sample, not the universe. Local businesses are measured through individual scans, not the national brand list.
- Not pay-to-play. No brand can buy inclusion, exclusion, or position — 8 tracked brands scored zero mentions and remain published.
Publisher and conflict disclosure
The index is published by Deep AI Solutions Inc (Houston, TX), which sells AI-visibility services — the measurement demonstrates the discipline the company sells, and that conflict is disclosed everywhere the data appears. The control that matters: numbers only ever come from logged, dated runs, and the live dataset at deepaivisibility.com/ai-visibility-index is publicly inspectable. If a number can't be traced to a run, it doesn't get published.
Frequently asked questions
Why only two passes per prompt?
A cost-noise tradeoff, disclosed rather than hidden: AI responses vary between identical runs, and two passes cut single-run noise substantially while keeping the run reproducible. More passes would tighten confidence further; edition-over-edition consistency is the stronger signal to watch.
What happens when an engine call fails?
Failures are logged and reported per engine rather than silently retried into the data — Edition 01's run log records each engine's call and success counts. Reporting failure rates keeps the denominator honest and shows exactly how much each engine contributed to the measurement.
How were the 46 brands chosen?
Pre-registered before the run: the recognizable national HVAC and plumbing brands a U.S. consumer might be recommended. Pre-registration matters because choosing brands after seeing results is how measurement quietly becomes marketing. Local businesses enter through individual scans instead.
How is sentiment scored?
As part of the AIR Score composite: logged responses are assessed for whether a mention is positive, neutral, or negative in context, alongside whether the brand was merely mentioned versus actively recommended. The three dimensions — mention, recommendation, sentiment — composite to the 0-100 score.
Can I audit the data?
The live dataset — brand table, engine breakdowns, run metadata — is public at deepaivisibility.com/ai-visibility-index. Cite it dated. If you find an error, email hello@deepaisolutions.com; corrections ship with a dated changelog note, the same standard we apply everywhere.