Where the practice stands today.

Cumulative measurement output. Updated whenever a new category clears pattern-readiness or a new teardown publishes. Last refreshed 2026-05-25.

The scoreboard

Categories measured

Each category was measured the same way: 18 buyer questions locked by hash, 3 repetitions per question per AI tool, 7 AI tools per run. Five are citation-grade engines that read the live web (Perplexity, ChatGPT search, Gemini grounded, Microsoft Copilot/Bing, Google AI Overviews) and two are model-knowledge engines answering from training data (Claude, Gemma). A category needs 3+ usable runs to clear the internal pattern-readiness rule.

The cross-category gradient (banking and med spas highest at about 51% own-site share, hotels and real estate lowest at about 11 to 14%) is the subject of the cross-category teardown. The honest finding: own-site AI citation share varies widely by category. Banking and med spas pull roughly half of citations to the firm’s own site, while hotels and real estate are dominated by third-party sources. Categories where buyers lean on aggregators and review platforms distribute citations away from the firm’s own site.

What we publish vs what we keep private

Two-tier transparency:

• Public (anonymized at the firm level): the teardowns at /teardowns/. Cohort sizes, source-type distributions, per-AI-tool breakdowns, named third-party platforms (Chambers, BestLawyers, SmartAsset, getwarmer, HMSA, Delta Dental, etc.), and the cross-category gradient. The banking cohort is named in full because all 23 firms are publicly competing brands.
• Private (named in full): the 1:1 deliverables inside paying engagements. Each customer’s engagement names every firm in their cohort, names the questions they’re missing on, and ranks the closable conditions. That deliverable is private to the customer.

Non-customer firms appearing in our cohorts can request removal at any time by emailing takedown@neverranked.com. 24-hour SLA.

The discipline behind the numbers

• Pre-registration. Every question set is locked by hash before measurement starts, and the hash is committed to a timestamped record. The lock never breaks. The method is documented at /methodology/, and every dated run is listed on the claims ledger.
• Inspectable model. Gemma is open-weight, so the model itself is independently inspectable rather than a closed box we describe from the outside.
• Pattern-readiness. No category claims a pattern until 3 usable runs land, per the internal pattern-readiness rule. The catalog tool enforces this and labels any premature pattern claim "data point, not pattern" in the deliverable itself.
• Cohort-coverage discipline. Each run surfaces additional cohort members through automated coverage scans. The wealth cohort grew from 15 to 42 firms over three rounds. The headline number rose from 27% to 47% as coverage improved. The teardowns document this kind of cohort-driven shift transparently.
• Fail-closed factual grader. Every prospect-facing artifact is graded against a canonical fact list before ship. The grader rejects unsubstantiated claims. Today across every published teardown and the Hamada brief, the grader rejected zero claims (everything substantiated cleanly).

What is NOT in this scoreboard

• Customer engagement count. We are at zero current paying customers under the rebuilt positioning. The measurement work and the discipline are real. The demand for honest research-only measurement is being validated, not assumed.
• A full national or international map. The methodology works for any geography. Most of the categories measured so far are in Hawaii because that is where the first measurement effort landed, with one published cross-geo measurement, Austin CPA, outside it.
• Categories we have not yet measured. New categories publish here as they clear pattern-readiness.

NeverRanked home · All teardowns · How we measure · FAQ