AI for UK insurance brokers is a noisy category in 2026. Every vendor claims to "use AI". Few are honest about the limits. This is the playbook we'd give a friend running a broker practice today.
What general LLMs are good at
ChatGPT, Claude and Gemini are excellent at summarising long PDFs, drafting client emails, and explaining concepts. If you paste a Bupa broker doc into Claude and ask "what's covered for mental health?", you'll get a plausible answer.
What general LLMs are bad at
They hallucinate the specifics. The same prompt run twice gives slightly different answers. They confuse 2023 and 2026 schedules. They mix one insurer's terminology with another's. And critically — they don't cite the page in the broker doc, so you can't verify the answer in the 30 seconds you have on a client call.
What works in a real broker workflow
Three patterns are reliably useful in 2026:
- Retrieval-augmented answers grounded in the actual broker doc, with page citations. This is what HealthCareCompare does.
- Email and quote-letter drafting from a fact sheet you provide.
- Renewal-difference summarisation: feed the AI last year's schedule and this year's, ask "what changed?".
What to avoid
- AI that promises to "recommend the best insurer". Recommendation requires regulated advice; AI should give cited facts, not opinions.
- AI tools that don't show their sources. If you can't click through to the page in the broker doc, you can't defend the advice.
- Generic LLM wrappers with no insurer-specific corpus. The whole point is grounding.
How HealthCareCompare fits
HealthCareCompare is a cited answer engine over the latest broker documents from Aviva, Bupa, AXA, Vitality and WPA. Brokers ask one question, get five side-by-side cited answers, and place business faster. It doesn't replace the broker — it removes the 20 minutes of PDF scrolling per client question.