People ask AI which insurance to buy, and comparison sites get named. What should insurers fix?

Why do AI answers name comparison sites instead of insurers?

Because the buyer's prompt asks for a comparison, and comparison sites publish exactly that. A person types "best health insurance for a family of four in the UAE" or "term life cover for a 35-year-old non-smoker in India". The engine looks for pages that already rank several policies against a need, and an aggregator's table fits that shape better than an insurer's product page.

The insurer is often present in the answer, but second-hand. Its name arrives through the comparison site's description, which may be months old, and its exclusions arrive through whatever the aggregator chose to summarise. Our view: the problem for most insurers is accuracy before visibility. Being named with the wrong waiting period is worse than not being named.

We found no published study of AI citations for insurance queries specifically, so this post works from broader finance and citation data and says where it does. Sigzen AI sells AI visibility audits and programmes at published prices, so we have an interest in the question.¹

What the 2026 surveys say about insurance buyers and AI

People already use AI to pick financial products, including insurance, and most of them do it in a general chat tool. In MDRT's survey of 2,000 adults in Singapore, run online by Opinium on 6–9 April 2026, 35% of people using digital platforms for their finances used AI tools for financial purposes.² Among those who had acted on AI-generated advice or used it as a starting point, 47% had used it when choosing or switching a product such as a loan, credit card or insurance plan.²

The same survey found 29% used general chat tools such as ChatGPT or Gemini for money matters, against 14% who used a chatbot run by their bank or financial institution.² MDRT is an association of financial professionals, so it has a stake in the advice question, though the fieldwork was done by an independent pollster.

Other countries show the same pattern at a lower level. In Gallup and Edward Jones's US survey of 5,075 adults (March to April 2026), 18% of Americans seeking financial guidance had used AI tools for it.³ In AustralianSuper's survey of 1,009 Australians by Quantum Market Research (April to May 2026), 17% used AI tools to inform financial decisions, and 45% of those users asked it to compare financial products.⁴

Money questions and AI, 2026

Choosing a product is one of the main jobs people give AI

47%of those acting on AI advice used it to choose or switch a product such as an insurance planMDRT and Opinium, Singapore, Apr 2026 ²

29% vs 14%use general AI chat tools vs a financial firm's own chatbotMDRT and Opinium, Singapore, Apr 2026 ²

18%of Americans seeking financial guidance had used AI tools for itGallup and Edward Jones, US, Mar–Apr 2026 ³

45%of Australians using AI for money ask it to compare financial productsAustralianSuper, Apr–May 2026 ⁴

Do buyers trust the answer, or check it?

Most of them check it, which gives an insurer a second chance if its own pages are clear. In the AustralianSuper survey (April to May 2026), 54% of people using AI for financial decisions cross-checked its answers against other sources, and 32% consulted a financial professional.⁴ In Gallup's US and Canadian fieldwork (March to April 2026), only 3% of Americans and 4% of Canadians had a great deal of confidence in AI as a source of financial guidance.³

Low trust does not make the answer harmless. The engine sets the shortlist, and the check that follows usually starts with the names it gave. A buyer who is told your policy has a two-year waiting period on a condition you cover after one may never reach your site to find out.

The checking also happens outside your channels. Gartner surveyed 3,566 B2B and B2C customers in February and March 2026 (geography not stated) and found people were about three times as likely to use a third-party GenAI tool as a company's own chatbot.⁵ Our view: a quote-and-buy assistant on your site helps people who already chose you, and does little for the comparison that happens before they arrive.

Where an insurance answer gets its facts

An engine answering an insurance question pulls from a mix of company pages, comparison pages, reviews and press, and the mix varies by engine. Profound, which sells AI visibility software, analysed 11.84 billion citations across eight engines between 16 April and 16 July 2026. About 57% went to company-owned sites, with ChatGPT at 47% and earned media taking 30% of its citations.⁶

Engines also differ in how much they read. Semrush, which also sells AI visibility software, found ChatGPT cited about 15 sources per response against about 3 for Gemini in 126 million US prompts from January to April 2026.⁷ In the same data, the top three brands held 41.4% of AI visibility in finance, against 82.9% in news and media.⁷

Read those together. Finance answers are less concentrated than some sectors, so a mid-sized insurer or broker can still be named. A broad reader such as ChatGPT will weigh comparison pages and press alongside your site. A narrow reader such as Gemini leans on fewer pages, so a clear product page of your own counts for more there.

How a policy question becomes a named answer

1. Buyer prompt
2. Retrieval
3. Comparison pages
4. Policy facts
5. Reviews and press
6. Named shortlist

The two highlighted stages are where an insurer loses. At the first, the comparison page describes you in its own words. At the second, the engine looks for a fact it can quote, such as an exclusion or a waiting period, and finds it only in a forty-page PDF.

Which insurance facts do engines get wrong?

The facts engines get wrong are the ones insurers publish least plainly: the limits and conditions, not the headline benefits. A marketing page lists what a policy covers. The answer a buyer needs is often about what it does not cover, and that sits in the policy wording.

• Exclusions and pre-existing conditions: which conditions are excluded, for how long, and whether a declaration changes that.
• Waiting periods: the general waiting period, and any longer ones for named treatments or conditions.
• Excess, deductible or co-payment: the amount, and whether it applies per claim or per year.
• Age and residency limits: entry ages, renewal ages, and who counts as resident for the policy.
• Claim steps: what to submit, where, and how the insurer tells the customer about a decision.
• Effective dates: when the current terms took effect, so an engine can tell a new rule from an old one.

Our view: the effective date is the cheapest fix and the most neglected. Insurance terms change every renewal cycle, and an engine that finds two versions of a waiting period with no dates will pick one. Often it picks the older one, because more pages repeat it.

The same problem appears with company facts. Where a wrong AI statement about your company comes from, and how to fix it covers the general method. For insurers, the source of the error is usually an old product page, an aggregator's summary or a PDF the engine read once.

The policy fact page: what to publish for each product

Publish one HTML fact page per product that states its limits in plain sentences and short tables, with the date the terms took effect. Link the full policy wording from it, but do not make the wording the only place a fact lives. The page should answer the questions a buyer types, in the order they type them.

Illustrative. The fields we use for an insurance fact page; your product and compliance teams decide the wording.

Mark up the insurer or broker itself with structured data, using the InsuranceAgency or Organization type, and keep its name, registration and contact details identical to the regulator's register.⁸ Structured data will not get you cited on its own. It helps an engine match the entity it reads about on a comparison site to the one on your site.

Where a policy fact lives, and what an engine can do with it

PDF policy wording only

• Fact sits on page 23 of 40
• No effective date near the fact
• Several versions online at once
• Engine quotes the aggregator instead

HTML fact page plus wording

• Fact in a plain sentence or table row
• Effective date on the page
• Old pages redirect to the current one
• Engine can quote you directly

Retire the old versions. A product page from three renewal cycles ago, still live at its old address, is a source an engine can find. Redirect it to the current fact page, or mark it clearly as a closed product with the date it closed.

What should you do about comparison sites and aggregators?

Treat them as sources to keep accurate, not rivals to outrank. A comparison site answers the category question better than any single insurer can, because it lists several. The engine will keep citing it, so the useful work is making sure what it says about you is current.

Start with the data you send them. Most aggregators build their tables from product data supplied by the insurer or broker, and a stale feed produces a stale listing. Check each listing against your fact page after every product change, and ask for corrections in writing when they are wrong.

Then look at your branded searches, because that is where a buyer goes to check. Ahrefs, which sells SEO software, analysed US desktop Google results from July to September 2026. On 29 September, 82.91% of branded results showed an AI Overview.⁹ Someone searching "is [your company] good at paying claims" now often reads a summary first, and our note on branded AI Overviews covers what those summaries cite.

Our view: brokers have more room here than insurers. A broker can publish an honest comparison of the policies it places, with its own reasoning, and that page has the shape engines look for. An insurer comparing itself with rivals rarely does, and readers know it. Our post on self-ranked best lists covers why a page that ranks its own author first is a weak bet.

Claims reputation is the part you cannot write your way out of

Claims experience reaches AI answers through reviews, complaints data and press, and no fact page changes it. When a buyer asks which insurer pays claims fairly, the engine draws on review sites, forum threads and any published complaint figures. Your own pages are one input among many, and a weak one on this question.

What you can control is the process description. Publish how a claim works, what documents it needs, how long each stage usually takes in your own words, and how a customer can escalate. A clear process page will not outweigh a pattern of poor reviews, but it gives the engine an accurate description to quote beside them.

Review sites matter for the same reason. Which review sites and directories the engines cite covers the general evidence. For insurance, the useful step is to find which review and complaint sources the engines cite for your product lines and answer what is there. Do not try to bury it.

How do you check what AI says about your policies?

Write buyer prompts per product line, run them on five engines on three separate days each month, and record what each answer says about you. Twenty prompts per product line is enough to start. Keep them stable from month to month so you can see movement rather than noise.

A good insurance prompt names the cover, the buyer and the country, because that is how people ask. The table below shows the shape, across four countries; swap in your own products and markets. The financial services and insurance playbook lists the prompt groups we draft for insurers.

Illustrative prompts; the right-hand column is what we record, not a measurement.

A monthly insurance prompt check

1. Write the prompts

   Twenty per product line, naming cover, buyer and country, plus a few branded ones.

2. Run them on five engines

   ChatGPT, Gemini, Microsoft Copilot, Google AI Overviews and Google AI Mode, on three separate days.

3. Record four things

   Named or not, who is named instead, each fact stated about you, and the sources cited.

4. Log every wrong fact

   With the source it came from: your old page, an aggregator, a review or a PDF.

5. Fix at the source

   Correct the page or listing, add the effective date, then re-run the same prompts next month.

Add a handful of branded prompts: your name on its own, your name with "claims", and your name beside the rival most often compared with you. They show what the engine believes about you today, and the wrong facts they surface become your correction list. Our guide to checking AI visibility yourself shows how to run a small version by hand.

Publishing facts under financial-promotion rules

Fact pages are easier to clear through compliance than marketing copy, because they state terms rather than promises. Most insurance regulators expect product information to be clear, fair and not misleading. A page that states exclusions and waiting periods plainly is closer to that standard than one that lists only benefits.

Three habits keep the review short. Draft from the approved policy wording rather than from marketing copy, so compliance checks a summary against a document it already signed off. Put the effective date and a link to the full wording on every fact page. And keep performance claims, such as claim payout rates, off the page unless you publish the full figure with its period and source.

Our view: insurers who publish almost nothing because every page is a financial promotion are handing the description of their products to the aggregators. A short, dated, compliance-approved fact page is lower risk than an AI answer built from someone else's summary.

Where to start

Start with your two largest product lines and the facts buyers ask about most. In the first month, publish a dated HTML fact page for each, redirect old versions, and check your comparison-site listings against them. In the second, run the twenty-prompt check per line and log every wrong fact with its source.

From the third month, fix what the check finds at its source and re-run the same prompts. Add product lines one at a time. What financial firms in Singapore should make answerable covers the wider set of money facts, such as rates, fees and licence details, if you also sell savings or investment products.

This post is desk research. Every figure comes from a published study, none from our own captures, and none of the studies isolates insurance. A captured version, built on our AI answer monitor's runs across five engines, would show insurance answers directly. Until then, the fact page and the monthly check are the parts you can act on today.