AI Visibility Index.
An open, pre-registered study of which websites an AI answer engine cites for local-service queries, and whether common on-page features predict being cited. Mindstorm Labs uses the same measurement approach in its AEO and GEO consulting.
Schema and FAQ blocks did not predict being cited.
Across five local-service verticals, Gemini (with Google Search grounding) cited business sites that look the same, on these two features, as organic top-10 sites it did not cite. Both pre-registered hypotheses came back null.
| Hypothesis | Feature | Cited sites with it | Control sites with it | Odds ratio (95% CI) |
|---|---|---|---|---|
| H1 | LocalBusiness-type or Organization schema | 75.3% | 76.2% | 0.96 (0.70–1.31) |
| H2 | FAQ block | 46.5% | 46.2% | 1.01 (0.77–1.33) |
| H3 | Name, address and phone consistency | Not tested | ||
An odds ratio of 1 means no difference. Adjusted for vertical, the odds ratios are 1.01 for H1 and 1.08 for H2; both intervals include 1. 857 query-by-site rows across 532 reachable business sites (454 cited, 403 control).
Reddit and review sites dominate.
Across all 300 queries, answers averaged 9.0 cited sources from 823 distinct domains. The most-cited sources were not business websites. This is a description of this query set, not a tested hypothesis.
How it was run.
- Queries: 300 = 5 verticals (plumber, dentist, HVAC company, personal injury lawyer, landscaper) × 10 US metros × 6 fixed templates, committed before collection.
- Engine: Gemini with Google Search grounding, collected 2026-10-12.
- Controls: organic top-10 business sites that Gemini did not cite for the same query, for a seeded random sample of 100 queries (20 per vertical).
- Features: detected from each site's homepage (JSON-LD parse; FAQ = FAQPage schema, an expandable details element, or an FAQ heading).
- Analysis: odds ratios with 95% confidence intervals and a logistic regression with vertical fixed effects.
Read this before quoting it.
- Confidence intervals are optimistic: the same site can appear in several rows and standard errors are not clustered by query.
- The schema feature is broad (an Organization type counts), and about three quarters of sites in both groups have it, so it separates little.
- Features were read from homepages only.
- One engine, US metros, five verticals, one collection date. Correlational, not causal.
- H3 (name, address and phone consistency) and a domain-age control were not run. The control sample was reduced from 300 to 100 queries because of API rate limits. Every deviation is logged in DEVIATIONS.md.
Citation and license.
Jenson, J. (2026). AI Visibility Index [Data set]. Mindstorm Labs. https://mindstormlabs.dev/jeffrey/research
Data: CC BY 4.0. Code: MIT. Repository: github.com/mindstorm-code/ai-visibility-index.