Every week we ask ChatGPT, Claude and Gemini the questions real UK consumers ask about health insurance — and record exactly which brands they put forward. Right now, Bupa owns the conversation.
0 AI answers this run · 0 brands named · measured, never estimated
Named in 38% of health insurance answers, typically in position 1.4. Aviva follows at 35%, Vitality at 32%.
Visibility is the share of answers naming the brand; citation share is its slice of all brand-owned pages cited; read→cited is how often its pages convert a read into a citation.
| # | Brand | Visibility | Avg. position | Citation share | Read→cited |
|---|---|---|---|---|---|
| 1 | Bupa | 38% | 1.4 | 26% | 14% |
| 2 | Aviva | 35% | 2.4 | 15% | 25% |
| 3 | Vitality | 32% | 3.3 | 24% | 53% |
| 4 | AXA | 30% | 2.4 | 11% | 8% |
| 5 | WPA | 8% | 3.0 | 6% | 13% |
| 6 | The Exeter | 7% | 3.5 | 2% | — |
| 7 | Saga | 3% | 3.0 | 2% | — |
| 8 | Simplyhealth | 2% | 1.0 | 2% | 17% |
| 9 | Allianz Care | 2% | 2.0 | 2% | 0% |
| 10 | Benenden Health | 2% | 2.0 | 2% | 100% |
| 11 | BHSF | 2% | 2.0 | 0% | 0% |
| 12 | April International | 0% | — | 0% | 0% |
| 13 | Freedom Healthinsurance | 0% | — | 8% | 13% |
| 14 | Post Office | 0% | — | 2% | 0% |
Being named is visibility. Underneath it, asked to rate each brand directly, the AI assistants judge it three ways — how much it stands out, how much it belongs, and how much it is returned to. They put 15% of the 27 brands rated above indifference. Bupa leads the brands the answers name, at 77% of maximum Desire. On attachment, the most divided driver, Bupa indexes 136 against Pi Financial at 69, on a category average of 100.
Each whorl lights one ring per index threshold cleared (84 / 92 / 100 / 108 / 116). Above indifference means Desire above two-thirds of its maximum. See the AI Desire Index, market by market →
An assistant does not answer from memory alone. It runs its own web searches first, and whoever it searches for is the shortlist it then chooses from. Across 99 searches in health insurance, Aviva is named in 31% of the ones that name a brand at all. ChatGPT Web does not report its searches, so these shares are measured against the ones that do.
These are the assistants' own searches, read from their answers — not what anyone typed into a search engine.
The independent sources AI models cite when answering health insurance questions — the places a brand needs to be seen, reviewed and ranked.
When Claude or ChatGPT opens a brand's own pages while answering, does the page make the cited-sources list? Vitality converts 53% of reads; Post Office just 0%. That gap is a content problem, not a visibility problem.
| # | Brand | Visibility | Avg. position | Citation share | Read→cited |
|---|---|---|---|---|---|
| 1 | Bupa | 38% | 1.4 | 26% | 14% |
| 2 | Aviva | 35% | 2.4 | 15% | 25% |
| 3 | Vitality | 32% | 3.3 | 24% | 53% |
| 4 | AXA | 30% | 2.4 | 11% | 8% |
| 5 | WPA | 8% | 3.0 | 6% | 13% |
| 6 | The Exeter | 7% | 3.5 | 2% | — |
| 7 | Saga | 3% | 3.0 | 2% | — |
| 8 | Simplyhealth | 2% | 1.0 | 2% | 17% |
| 9 | Allianz Care | 2% | 2.0 | 2% | 0% |
| 10 | Benenden Health | 2% | 2.0 | 2% | 100% |
| 11 | BHSF | 2% | 2.0 | 0% | 0% |
| 12 | April International | 0% | — | 0% | 0% |
| 13 | Freedom Healthinsurance | 0% | — | 8% | 13% |
| 14 | Post Office | 0% | — | 2% | 0% |
Ask about health insurance and the brand you are shown depends on the assistant: Claude → Bupa, Gemini → Aviva, ChatGPT → Bupa. Their top-five lists share just 38% of names.
Brands whose visibility swings most between models — the bar per model, widest gap on the right.
| Brand | Claude | Gemini | ChatGPT | Spread |
|---|---|---|---|---|
| AXA | 25% | 60% | 5% | 55pp |
| Aviva | 20% | 70% | 15% | 55pp |
| Vitality | 25% | 60% | 10% | 50pp |
| Bupa | 30% | 65% | 20% | 45pp |
| WPA | 15% | 5% | 5% | 10pp |
| The Exeter | 10% | 10% | 0% | 10pp |
| BHSF | 5% | 0% | 0% | 5pp |
| Saga | 0% | 5% | 5% | 5pp |
| Simplyhealth | 0% | 0% | 5% | 5pp |
| Benenden Health | 0% | 0% | 5% | 5pp |
Every page AI opened while answering, classified by what kind of page it is. Editorial supply the most citations; faq convert best — 100% of the ones models opened were actually cited.
| Page type | Cited | Read but not cited | Conversion |
|---|---|---|---|
| Editorial | 189 | 20 | 90% |
| How-to guides | 133 | 4 | 97% |
| Comparisons | 48 | 2 | 96% |
| Product pages | 36 | 3 | 92% |
| faq | 16 | 0 | 100% |
| Other | 15 | 12 | 56% |
| pricing | 12 | 0 | 100% |
| Documentation | 11 | 4 | 73% |
Of the health insurance pages models actually opened, how much more often each feature's pages ended up in the cited sources. Structured data is worth 28 percentage points; cites its sources pages fare 17pp worse. Measured within the read population, so selection can't flatter it.
| Feature | Pages with | Cited % | Pages without | Cited % | Difference |
|---|---|---|---|---|---|
| Structured data | 460 | 93% | 110 | 65% | +28pp |
| Named author | 312 | 92% | 258 | 81% | +11pp |
| Comparison table | 216 | 94% | 354 | 83% | +11pp |
| Answer up front | 326 | 92% | 244 | 82% | +10pp |
| Updated this year | 243 | 90% | 327 | 85% | +5pp |
| Hard statistics | 145 | 78% | 425 | 91% | -13pp |
| Cites its sources | 31 | 71% | 539 | 88% | -17pp |
Readiness scores each brand's own pages on the structure models reward — a direct answer up front, tables, hard numbers, named authors, structured data, recency, specifications. General & Medical is the widest gap: well-built pages that still aren't converting into citations.
| Brand | Readiness | Pages cited | Pages profiled |
|---|---|---|---|
| Post Office | 35 | 40% | 5 |
| Vitality | 35 | 14% | 116 |
| WPA | 28 | 21% | 33 |
| Bupa | 28 | 17% | 143 |
| General & Medical | 27 | 0% | 7 |
| Saga | 26 | 20% | 10 |
| AXA | 25 | 6% | 130 |
| Allianz Care | 25 | 60% | 5 |
| Freedom Healthinsurance | 24 | 25% | 16 |
| Aviva | 23 | 7% | 144 |
Readiness is the share of eight measurable structural signals present on a brand's profiled pages. It is a build-quality measure, not a ranking of the brand.
Read straight off the pages AI cited: the consumer questions cited content answers most often. Content that fails to settle these is content models have no reason to quote.
Method: 20 unbranded consumer prompts per category, spread across the buying journey, run weekly against ChatGPT, Claude and Gemini. Read→cited uses Claude and ChatGPT only (Gemini does not separate the two). Correlational observations of model behaviour at the time of the run. Data from the run completed Wed, 21 Oct 2026 07:40:25 GMT.