The 75-point engine spread that no single-engine tool can see.
First empirical teardown from the 7-engine NeverRanked methodology. 23-bank Hawaii consumer banking cohort, 18 hash-locked queries, 3 usable runs, 5,153 citations captured. Pattern-readiness cleared. Subject brand and cohort anonymized.
The cohort (anonymized)
The category cohort is the Hawaii financial institutions AI engines cite for consumer banking queries: a set of major banks and credit unions. Individual institutions are anonymized in this public teardown (banks labeled Bank A onward, credit unions CU A onward, ranked by mentions). Categories, counts, source-type distributions, and named third-party platforms are public. Individual institution names are not. The methodology demonstration value comes from the pattern across the group, not from singling any one out.
Methodology summary
7 AI surfaces measured, all on the same day, all on the same query set:
- 5 citation-grade engines (search the web at query time): Perplexity, ChatGPT search, Gemini (grounded with live web search), Microsoft Copilot via Bing organic SERP, Google AI Overviews.
- 2 model-knowledge engines (answer from training data, no web search): Claude (via the Anthropic API), Gemma (open-weight, run on a model-hosting provider so the model itself is independently inspectable).
18 queries hash-locked at 48b6c76f.... 3 repetitions per query per engine for noise control. 3 usable runs all on 2026-05-23, totaling 1,134 successful API calls and 862 cited-text rows scanned. Pattern-readiness rule of 3+ usable runs is cleared.
Full methodology, with the hash-locked question sets and the dated runs on the /claims/ ledger, at /methodology/.
Source-type distribution (cohort-wide)
Across 5,153 citations captured over 3 hash-locked runs, where AI engines pulled from for Hawaii consumer banking queries:
| Source type | % of citations | Count |
|---|---|---|
| Competitor (cohort bank-owned sites) | 53% | 2,745 |
| Independent web | 42% | 2,184 |
| Wikipedia | 2% | 83 |
| 1% | 60 | |
| Social (Facebook, Instagram, TikTok) | 1% | 32 |
| YouTube | 1% | 30 |
| Review directories (Yelp, TripAdvisor) | <1% | 19 |
The cohort-wide read. For Hawaii consumer banking, AI cites bank-owned sites slightly more than third-party content (53% vs 42%). That's not the result a buyer expects. Conventional SEO instinct says "AI cites third-party content about you, so optimize the content, not the site." For Hawaii banking, the data says the opposite is closer to true: own-site optimization is the majority of where AI is looking, third-party content is the minority. The actionable shape of the punch list changes accordingly.
But the cohort-wide number hides the bigger finding. The per-engine breakdown is where the moat lives.
Per-engine breakdown · the 75-percentage-point spread
| Engine | Competitor | Independent web | Other |
|---|---|---|---|
| Perplexity | 75% | 23% | YouTube 1% |
| Claude (model-knowledge) | 72% | 28% | — |
| ChatGPT search | 71% | 27% | Wikipedia 2% |
| Gemma (model-knowledge) | 70% | 30% | — |
| Google AI Overviews | 59% | 27% | Reddit 7%, social 4%, YouTube 2%, Wikipedia 1%, review_dir 1% |
| Gemini grounded | 30% | 67% | Reddit 1%, Wikipedia 1% |
| Microsoft Copilot (Bing) | 0% | 87% | Wikipedia 9%, review_dir 2%, social 1%, YouTube <1% |
The non-obvious read
Five AI engines (including both model-knowledge engines, Claude and Gemma, which answer from training data without searching the web) cite Hawaii bank-owned sites 60-75% of the time. Microsoft Copilot, powered by Bing's organic SERP, cites bank-owned sites 0% of the time, surfacing third-party editorial content (Wikipedia 8%, review directories 3%, news/independent web 88%) instead.
This is not a measurement error. Bing's SERP for these queries genuinely surfaces editorial and reference content above bank pages. A Hawaii bank with strong AI presence on five surfaces could still be invisible to a user who asks Microsoft Copilot, and the only way to know that is to measure all seven surfaces, in parallel, on the same queries.
That's the entire structural argument for measuring more than one engine. The data is the argument.
Other per-engine findings worth naming
- Google AI Overviews is the only engine that surfaces Reddit (7%) and social (4%) for banking queries. Editorial guidance on those surfaces won't help with AIO presence. User-generated content can.
- Reddit, Wikipedia, YouTube, social, and review-directory citations are essentially absent from the other six engines for this category. Banking is not a forum-driven category in AI answers. (Compare to consumer electronics or software categories, where Reddit shows up at 15-30% in AI Overviews and Perplexity.)
- Claude and Gemma (model-knowledge engines) cite the major Hawaii banks by name with no web search at all. 72% and 70% competitor share respectively, just from training data. That means even before any AI search happens, these banks have measurable model-knowledge presence. No comparison tool measures this layer.
- ChatGPT search surfaces location-specific URLs heavily. For one major bank, a single location-specific subdomain alone accounts for 54 of OpenAI's 60 citations of that bank. OpenAI strongly favors specific branch/location pages over brand home pages for "best bank in [neighborhood]" queries. Actionable optimization finding.
Top cited brands · mentions across engines
Brand-host mentions across the 862 cited-text rows, with position-in-answer breakdown (Q1 = lead position, Q4 = tail). Top 10 by mentions:
| Institution (anonymized) | Mentions | Q1 lead | Q2 early | Q3 mid | Q4 tail | Position bias |
|---|---|---|---|---|---|---|
| Bank A | 612 | 265 | 219 | 95 | 33 | lead-heavy |
| Bank B | 584 | 144 | 259 | 135 | 46 | lead-heavy |
| Bank C | 463 | 83 | 149 | 146 | 85 | early-heavy |
| Bank D | 257 | 71 | 72 | 62 | 52 | early-heavy |
| Bank E | 148 | 17 | 27 | 79 | 25 | mid-heavy |
| CU A | 115 | 20 | 21 | 60 | 14 | mid-heavy |
| CU B | 96 | 4 | 18 | 49 | 25 | mid-heavy |
| Bank B (branch pages) | 76 | 35 | 22 | 11 | 8 | lead-heavy |
| CU C | 66 | 18 | 13 | 26 | 9 | mid-heavy |
| CU D | 58 | 19 | 17 | 12 | 10 | lead-leaning |
What the position-bias column reveals. Bank A is "lead-heavy," with 79% of its 612 mentions in the early half of AI answers (Q1+Q2). That's the dominant AI-recommendation pattern. Bank E, with notably less total visibility (148 mentions), is "mid-heavy," at 53% Q3 alone. Same engine, same query universe, fundamentally different competitive posture. Total mention count alone misses this. Position bias is the depth signal no comparison tool surfaces.
True third-party citation sources (the actionable list)
After cohort classification, the third-party content AI cites for Hawaii consumer banking, ranked by recurring-host frequency across the 3-run dataset:
| Host | Cites | What it is |
|---|---|---|
| reddit.com | 60 | Hawaii-banking discussion threads (Google AIO + Gemini) |
| apps.apple.com | 58 | App Store listings for bank mobile apps |
| hmsa.com | 48 | Hawaii Medical Service Association (HMSA-direct-deposit query artifact) |
| bankbonus.com | 47 | National banking-product review site (mostly AIO) |
| play.google.com | 42 | Google Play Store listings for bank apps |
| hawaiistar.com | 41 | Local Hawaii business publication |
| en.wikipedia.org | 64 | Hawaii banking history pages (mostly Bing) |
| facebook.com | 21 | Social posts about Hawaii banks (Google AIO only) |
| yelp.com | 7 | Review listings (Bing only) |
These are the surfaces where editorial mentions actually move citation visibility for Hawaii consumer banking. Notice the engine-specific patterns: Reddit threads matter for Google AIO and Gemini specifically (invisible to the other 5 engines). Wikipedia and Yelp matter for Bing/Copilot specifically. Social platforms (Facebook, Instagram, TikTok) only show up in Google AI Overviews. The path to AI presence in each engine is structurally different, and the strategy for moving citation share has to be specific to which engines matter to the bank's actual buyers.
What this teardown does not yet include
Side-by-side comparison with a dashboard-style AI-citation tool's free AEO Report on the same subject. The original plan for this teardown was a head-to-head against another tool's published output for the same Hawaii bank. That comparison artifact has not been pulled yet. When it is, this teardown will be updated with the side-by-side and the differentiator analysis it produces.
In the meantime, the 7-engine measurement above stands on its own as a methodology demonstration. The headline finding (75-percentage-point Bing/Copilot gap) is independent of any vendor comparison. It's a structural finding about the AI-engine landscape itself, and a single-engine measurement methodology cannot detect it.
Why this is anonymized
Individual institutions are anonymized in this public teardown. These are publicly-competing brands appearing in any local "best Hawaii bank" comparison, but publicly naming a specific brand alongside competitive findings creates trademark, trade-libel, and endorsement-implication exposure that the cohort-level, anonymized analysis above does not. The unredacted version (subject-named with the full per-brand competitive analysis) is available 1:1 to qualified prospects in conversation. The same anonymization pattern is used across NeverRanked's public artifacts.
Methodology + reproducibility
Full methodology at /methodology/. The method is documented end to end there: the measurement runner, the source-type classifier, the within-citation analyzer, and the aggregator, with the question sets hash-locked at 48b6c76f... and every dated run posted to the /claims/ ledger. Gemma is open-weight, so the model itself is independently inspectable.
Raw data manifests for the 3 runs that produced this teardown are at dryrun/out/2026-05-23T10-13-34Z/ (2-engine seed), dryrun/out/2026-05-23T11-14-31Z/ (7-engine pass), and dryrun/out/2026-05-23T21-42-59Z/ (third 7-engine pass clearing pattern-readiness). All three runs aggregate cleanly into the same query-set cohort.
If you're a Hawaii financial institution evaluating this methodology
Email Lance@hi.neverranked.com for the unredacted version with the specific subject-bank's competitive position, or to scope a full kickoff engagement on your own brand. Engagements are $4,500 kickoff per category, $1,500/month ongoing.
Return to NeverRanked · How we differ from the dashboards · How we measure · Example deliverable · Takedowns & opt-out
Measurement window: 2026-05-23. AI engine behavior changes weekly, so we will re-run the comparison on request.
Methodology & substantiation: every quantitative claim above is derived from a hash-locked, pre-registered measurement (query set hash 48b6c76f...), captured by the documented measurement method, against named engines on a named date with a named query set, and graded by a fail-closed factual checker. Full chain at neverranked.com/methodology.
Takedown & opt-out: if your business is named on this page and you want it removed, email takedown@neverranked.com. 24-hour SLA. Full process: neverranked.com/takedowns.