Health Insurance,
according to AI
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
Bupa is the default answer.
Named in 40% of health insurance answers, typically in position 1.7. Aviva follows at 32%, AXA at 25%.
Every brand AI names, ranked.
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 | 40% | 1.7 | 34% | 18% |
| 2 | Aviva | 32% | 2.5 | 14% | 17% |
| 3 | AXA | 25% | 1.9 | 6% | 40% |
| 4 | Vitality | 23% | 2.6 | 23% | 43% |
| 5 | The Exeter | 7% | 5.0 | 0% | 0% |
| 6 | WPA | 5% | 4.7 | 6% | 50% |
| 7 | Allianz Care | 2% | 3.0 | 6% | 67% |
| 8 | Freedom Healthinsurance | 2% | 4.0 | 6% | 50% |
| 9 | MediCash | 2% | 6.0 | 0% | — |
| 10 | Saga | 2% | 6.0 | 3% | — |
| 11 | BHSF | 2% | 7.0 | 0% | — |
| 12 | General & Medical | 2% | 8.0 | 0% | — |
| 13 | Now Health International | 0% | — | 0% | 0% |
| 14 | Post Office | 0% | — | 3% | 0% |
| 15 | Zurich | 0% | — | 0% | 0% |
Before it answers, it goes looking for AXA.
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 89 searches in health insurance, AXA is named in 33% 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 referees of health insurance.
The independent sources AI models cite when answering health insurance questions — the places a brand needs to be seen, reviewed and ranked.
Allianz Care converts reads into citations. Aviva doesn't.
When Claude or ChatGPT opens a brand's own pages while answering, does the page make the cited-sources list? Allianz Care converts 67% of reads; Aviva just 17%. That gap is a content problem, not a visibility problem.
Full data table
| # | Brand | Visibility | Avg. position | Citation share | Read→cited |
|---|---|---|---|---|---|
| 1 | Bupa | 40% | 1.7 | 34% | 18% |
| 2 | Aviva | 32% | 2.5 | 14% | 17% |
| 3 | AXA | 25% | 1.9 | 6% | 40% |
| 4 | Vitality | 23% | 2.6 | 23% | 43% |
| 5 | The Exeter | 7% | 5.0 | 0% | 0% |
| 6 | WPA | 5% | 4.7 | 6% | 50% |
| 7 | Allianz Care | 2% | 3.0 | 6% | 67% |
| 8 | Freedom Healthinsurance | 2% | 4.0 | 6% | 50% |
| 9 | MediCash | 2% | 6.0 | 0% | — |
| 10 | Saga | 2% | 6.0 | 3% | — |
| 11 | BHSF | 2% | 7.0 | 0% | — |
| 12 | General & Medical | 2% | 8.0 | 0% | — |
| 13 | Now Health International | 0% | — | 0% | 0% |
| 14 | Post Office | 0% | — | 3% | 0% |
| 15 | Zurich | 0% | — | 0% | 0% |
The models do not agree on who leads.
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 67% of names.
Brands whose visibility swings most between models — the bar per model, widest gap on the right.
The data behind this chart
| Brand | Claude | Gemini | ChatGPT | Spread |
|---|---|---|---|---|
| Aviva | 20% | 60% | 15% | 45pp |
| Bupa | 25% | 60% | 35% | 35pp |
| AXA | 20% | 45% | 10% | 35pp |
| Vitality | 10% | 40% | 20% | 30pp |
| The Exeter | 0% | 15% | 5% | 15pp |
| WPA | 0% | 5% | 10% | 10pp |
| BHSF | 0% | 5% | 0% | 5pp |
| Freedom Healthinsurance | 0% | 5% | 0% | 5pp |
| MediCash | 0% | 5% | 0% | 5pp |
| Saga | 0% | 0% | 5% | 5pp |
Editorial carry health insurance.
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.
The data behind this chart
| Page type | Cited | Read but not cited | Conversion |
|---|---|---|---|
| Editorial | 106 | 9 | 92% |
| How-to guides | 59 | 2 | 97% |
| Comparisons | 23 | 2 | 92% |
| Product pages | 14 | 2 | 88% |
| Other | 9 | 9 | 50% |
| faq | 8 | 0 | 100% |
| Documentation | 6 | 2 | 75% |
| Community threads | 6 | 8 | 43% |
Structured data is what separates cited from ignored.
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 27 percentage points; hard statistics pages fare 21pp worse. Measured within the read population, so selection can't flatter it.
The data behind this chart
| Feature | Pages with | Cited % | Pages without | Cited % | Difference |
|---|---|---|---|---|---|
| Structured data | 227 | 93% | 62 | 66% | +27pp |
| Comparison table | 109 | 97% | 180 | 81% | +16pp |
| Named author | 155 | 94% | 134 | 80% | +14pp |
| Answer up front | 150 | 93% | 139 | 81% | +11pp |
| Updated this year | 115 | 90% | 174 | 86% | +4pp |
| Hard statistics | 78 | 72% | 211 | 93% | -21pp |
WPA builds pages AI can quote.
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. Vitality is the widest gap: well-built pages that still aren't converting into citations.
| Brand | Readiness | Pages cited | Pages profiled |
|---|---|---|---|
| WPA | 40 | 83% | 6 |
| Vitality | 35 | 6% | 106 |
| Post Office | 29 | 0% | 3 |
| Bupa | 28 | 8% | 129 |
| General & Medical | 27 | 0% | 7 |
| AXA | 25 | 3% | 126 |
| Saga | 25 | 0% | 8 |
| Allianz Care | 25 | 33% | 3 |
| Freedom Healthinsurance | 24 | 14% | 14 |
| Aviva | 24 | 2% | 110 |
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.
The questions health insurance pages have to settle.
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.
Nineteen more categories.
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, 02 Sep 2026 06:50:23 GMT.