The Hartford Has the Best Top-Three Rate—but Ranks Eighth Overall

1. The position paradox

Top-Three Rate and share of model measure different things, and The Hartford sits at opposite ends of the two. Top-Three Rate asks a conditional question: when this brand is named, how often does it land in the top three? Share of model — the same measure that search reporting calls share of voice, as the methodology explains — asks an unconditional one: how much of the total organic mention volume in this category belongs to it? A brand can be excellent at the first and negligible at the second, and The Hartford is the clearest example of that in the study cohort.

The composite explains the ranking. AIVI™ weights share of model at 35%, recommendation strength at 25%, position weighting at 20%, citation influence at 15%, and top-three rate at 5%. The Hartford leads the top ten on Top-Three Rate, which accounts for 5% of the composite. Position is also reflected separately through the 20% Average Position Score component. Its strength in early placement is therefore rewarded, but not enough to overcome its smaller Share of Model and weaker recommendation performance. Reordering the same underlying performance against a different weighting would move it several places; that is not a criticism of the weighting, it is the reason a single composite number should never be read without its components.

The operational read is more useful than the ranking. An 88.1% Top-Three Rate shows that once The Hartford is mentioned, it is usually given prominent placement. The remaining question is why it enters fewer of the measured answers in the first place.

2. What we measured, and what we did not

This section is deliberately placed before the analysis rather than after it.

• Public model outputs only. Every figure comes from organic brand mentions in responses generated by ChatGPT, Gemini, Perplexity, Copilot, and Claude. No advertising data, no analytics access, no internal reporting from any carrier.
• No client relationship. Brainpan.AI has no commercial relationship with The Hartford. The company was not contacted, did not participate, and did not review this page. It appears here because it is the most analytically interesting profile in the Q2 2026 top ten, not because of any dispute or engagement.
• Organic-only, first-mention attribution. Sponsored placements and paid surfaces are excluded. Where a response names several brands, attribution follows the first-mention rule described in the methodology.
• A sample, not a census. One quarter, 300 stratified prompts, 1,500 model responses. Prompt sets, model versions, and retrieval behavior all move between quarters. Treat quarter-over-quarter movement as directional until a second cycle confirms it.
• Open data, CC BY 4.0. The underlying benchmark is published under a Creative Commons Attribution 4.0 license. Reuse it, check it, disagree with it in public.

Category-level figures in section 3 use Visibility Points, which are the sum of position, recommendation, and citation scores inside a single line of business. They are not normalized 0–100 AIVI™ scores and should not be compared across lines of different sizes.

3. It wins its category outright and still loses the index

The Hartford is the number one brand in Commercial insurance in the Q2 2026 study, and it is not close. It records 598 Visibility Points against Hiscox at 322, Travelers at 281, Progressive at 222, and Chubb at 214 — nearly double the runner-up in the line where it actually competes.

That Commercial dominance does not translate proportionally into the overall index because AIVI™ measures performance across the complete prompt universe, not leadership within one line of business, and share of model is computed across the whole prompt set rather than inside a line. A brand that owns a narrow, high-value segment of the category can be simultaneously dominant in the segment and marginal in the index.

That is a real finding rather than an artifact. Buyers do not prompt an answer engine by line of business; they prompt it by situation. A question phrased as what insurance does a 20-person contracting firm need may never surface the word commercial, and whether The Hartford is retrieved depends on whether its content is written against that situation rather than against the internal product taxonomy. Winning the vertical inside AI answers and being visible in AI answers are not the same achievement.

4. Listed, but not recommended

Recommendation rate is the share of a brand’s mentions in which the model does more than name it — where the answer actively suggests it, ranks it, or steers the reader toward it. The Hartford has the lowest rate in the top ten.

No other brand in the set sits at both extremes. The Hartford holds rank one of ten on position and rank ten of ten on recommendation. Chubb, the next lowest on recommendation at 32.1%, is mid-pack on position. USAA, the highest at 46.3%, is mid-pack on position too. The combination is specific to The Hartford, and it points at a specific failure: the brand is entering answers as a reference point rather than as an option.

Rates conceal the size of the problem, so it is worth converting them into counts.

Thirty-seven. Across 1,500 model responses, The Hartford was actively recommended 37 times. State Farm was recommended 187 times. The recommendation-rate gap between them is 13.3 percentage points; the gap in actual recommendations is 5.05× against a mention-volume gap of 3.4×. A weak conversion rate applied to a small base compounds the volume deficit rather than offsetting it.

The counterfactual is the number worth carrying into a planning conversation. Hold mention volume and Share of Model constant and raise recommendation conversion to the top-ten median: the modeled result increases from approximately 37 to 54 recommended mentions — a 46% gain without requiring additional reach.

5. Citation availability does not appear to be the primary constraint

The reflexive enterprise prescription for weak AI visibility is more content, more authority, more third-party corroboration. For The Hartford, the data does not support it.

The Hartford records a 0.425 citation-to-mention ratio, above the pooled top-ten ratio of 0.371. That makes visible citation capture less likely to be the primary constraint in this dataset. It does not eliminate source quality, provenance, or third-party authority as contributing factors, but it shifts the first diagnostic priority toward consideration-set entry and recommendation conversion.

There is a second reason a pure source-authority program would under-deliver here, and it applies to every brand in the category.

Only Perplexity and Copilot emitted citations in this dataset (citation rate above is the share of each engine's organic mentions carrying at least one source citation). ChatGPT, Claude, and Gemini operated as non-citing recommendation environments — and between them they produced 2,549 of the study’s organic brand mentions, including the single largest mention surface in the study. Citation-only measurement is blind to those three engines: their influence has to be evaluated through direct measures such as mentions, placement, recommendation behavior, and narrative alignment, not inferred from visible citations alone. What those three engines respond to is more likely to be whether the brand reads as a recommendable option inside the passage the model retrieves, which is a content-structure and entity-clarity problem rather than a link-acquisition one.

6. The dead heat with Nationwide

The clearest illustration of what share of model does to a composite score is a side-by-side with a brand three places higher.

Multiply mentions by top-three rate and The Hartford produces roughly 118 top-three placements. Nationwide produces roughly 117. In terms of premium real estate inside AI answers, these two brands delivered the same quarter. The AIVI™ scores are 42.0 and 63.4 — 21.4 points and three ranks apart.

Nationwide reaches the same placement volume from 252 mentions at a 46.4% top-three rate. The Hartford reaches it from 134 mentions at 88.1%. One of those is a breadth strategy and the other is a precision strategy. The index rewards Nationwide's approach here because it pairs that breadth with stronger recommendation and citation performance, not because breadth alone decides the composite. For a brand deciding where to spend, the comparison still sets the trade cleanly: The Hartford does not need to become better at being placed well. It needs to be present in more of the conversations where placement is available.

7. Strategic diagnostic priorities

Four moves follow from the data above. They are stated as diagnostic direction rather than as an execution plan — sequencing, prompt targets, and content specification are what an audit produces, and they depend on inputs no public dataset contains. The operating frameworks that carry that work — content standard, technical foundation, publishing protocol, and monthly cadence — are published in full at Brainpan.AI frameworks.

Treat consideration-set entry as the primary constraint, not authority

Position rate and citation efficiency are both above benchmark. Mention volume is not. Any program that starts from build more authority is optimizing a variable that is already performing and leaving the binding constraint untouched. The first question is which prompt territories generate insurance answers that never name the brand at all.

Convert reference mentions into recommended mentions

A 27.6% recommendation rate against a 40.2% median is the highest-leverage single number on this page, because it moves without requiring any growth in share of model. The work is in how the brand’s content states fit, eligibility, and comparison — the material an answer engine needs in order to recommend rather than merely cite.

Write to situations, not to the product taxonomy

Category dominance in Commercial that does not translate into index visibility is a signal that content is organized around what the company sells rather than around what the buyer is trying to resolve. Retrieval happens at the passage level against a situation, and the passage has to be self-contained enough to survive being lifted out of its page.

Instrument the three non-citing engines separately

ChatGPT, Claude, and Gemini produced most of the mention volume in this study and zero citations. They cannot be measured with a citation-tracking approach, and a program reported purely on citations will show no movement on the surfaces that matter most. Recommendation presence has to be measured directly.

8. Data, method, and corrections

All figures are drawn from the Q2 2026 AIVI™ Insurance Benchmark, a study of 134 evaluated insurance-sector brands and entities across 300 stratified prompts and 1,500 model responses on five engines, producing 4,328 organic brand mentions and 1,742 organic citations, of which 1,348 resolved to a brand within the scored cohort (the remaining 394 resolved to sources outside it — see the entity-type table in the Allstate teardown). The Hartford’s 134 figure in this teardown refers to its organic mentions across the prompt set, and is unrelated to the 134-brand study cohort.

AIVI™ is an organic-only, first-brand-mention composite of share of model (35%), recommendation strength (25%), position weighting (20%), citation influence (15%), and top-three rate (5%), normalized so the category leader scores 100. The top-ten machine-readable dataset is published at /api/aivi/insurance/q2-2026/top10.json and the per-brand record at /api/aivi/insurance/q2-2026/brands/the-hartford.json, both under CC BY 4.0.

The controlling rules behind every number above — prompt-set construction, attribution, the component weights, and the known limitations — are published in the versioned AIVI™ methodology and correction policy. Prompt design follows the generative-engine framing set out in Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024). The published dataset is declared as a schema.org Dataset under CC BY 4.0, and the carrier universe is framed against the National Association of Insurance Commissioners record.

If any figure on this page is wrong, we would like to know. Corrections go to kwalsh@brainpan.ai and are handled under the published correction policy — acknowledged in two business days, determined in ten, and applied to the page with a dated note rather than silently.