The State of AI Answer Accuracy in Singapore's Insurance Sector, A Preliminary Barometer

Ask a public AI assistant where to take a complaint your insurer could not resolve, how long that insurer has to handle it, or how quickly help reaches you at the scene of an accident, and the answer comes back fast and confident, often with a named body, a tidy number of days, or a hotline. The fact it is describing is usually a public one: the insurer's own complaints page, its published service standards, its accident-assistance page. As of August 2026, this is our third monthly reading, and we again checked answers against those published facts, one by one.

The certified findings in this reading concern the practical machinery a consumer reaches for when something has gone wrong: where to escalate, how long a complaint takes, and how fast help arrives. Accurate handling was also present in the reviewed examples. This is one of several sector readings in our barometer, and it sits alongside our Singapore banking reading as a companion view of the same problem.

How we checked AI answers about Singapore insurance

In August 2026 we again put high-intent questions, the kind a person types before complaining, escalating, or calling for help after an accident, to a set of public AI systems, among them ChatGPT, Gemini, Copilot, Perplexity, Google AI Mode and Google AI Overview. Questions about how to escalate a complaint, how long an insurer takes to handle one, and accident-hotline response were compared against the insurer's own published complaints, service-standard and contact pages, and against the dispute-resolution body's own page. Where an answer conflicted with the published record, we flagged it for review, and we hand-verified each finding below against a dated official-source capture before standing behind it.

This is a preliminary reading, and we would rather say so plainly than dress it up. It covers a single month, so it is a directional reading rather than a trend, and the examples we reviewed were selected for diversity, not drawn as a random or complete sample, so we make no claim about how often any error occurs. We report findings, not a scoreboard. We name the AI systems only as the set we tested, never to rank one against another. The dispute-resolution body and the public schemes and regulators, in both Singapore and, where an answer wrongly reached for them, Malaysia, are public bodies, so we name them and cite their official pages directly. The insurers we do not name: where a finding turns on one insurer's own complaints page, service standard or accident-assistance page, we describe it as "a major Singapore insurer" and withhold the name, to preserve the anonymised research design. We retain the underlying evidence and result identifiers internally for audit and right-to-reply.

Where AI answers about Singapore insurance drift

Two answer-level errors underlying separate findings share the same observable shape: answers to Singapore questions reproduced facts the insurer publishes for its Malaysian operation. The insurer involved in those two errors operates in more than one market and publishes separate pages for each; an answer that reproduces the wrong market's page is confident, specific, and wrong for the consumer asking it.

A complaint was routed to another country's ombudsman. A consumer asked how to escalate a complaint their Singapore insurer had not resolved. A public AI assistant pointed them to Malaysia: the Financial Markets Ombudsman Service and Bank Negara Malaysia's BNMLINK, the escalation channels the insurer publishes on its Malaysian contact page. For an eligible unresolved dispute in Singapore, the route the insurer's own Singapore page identifies is FIDReC, the Financial Industry Disputes Resolution Centre. A consumer who follows the answer may take their complaint to bodies in another country and may delay reaching the appropriate Singapore route. A second assistant, asked the same kind of question about a different insurer, went wrong in the other direction: it omitted FIDReC entirely and offered avenues that do not resolve insurer disputes the way FIDReC does. Either way, the one step a dissatisfied consumer most needs signposted, the external route, was the step that failed.

A complaint timetable came from another market. A consumer asked how long their Singapore insurer takes to acknowledge and resolve a complaint. A public AI assistant answered with that insurer's Malaysian operation's complaint-handling schedule, reproducing even a "1 April 2026" change that applies in Malaysia, with figures like next-working-day acknowledgement and five working days for simple cases. The Singapore insurer's own published service standard is different: acknowledgement within 2 working days, and a final resolution within 14 working days. The numbers are not invented, which is part of what makes the answer hard to catch. They are simply the wrong country's numbers, given to a consumer who may measure their Singapore insurer against a commitment it never made.

A published accident-response standard was dropped. A consumer asked for their insurer's 24-hour car-accident hotline and how quickly help arrives at the scene. A public AI assistant gave the hotline number and the 24-hour availability correctly, but on the response time it omitted the insurer's published target, help at the scene within 20 minutes of your call. The hotline details were right; the specific, checkable figure the consumer actually asked about was left out. Omitting a published response target tells a consumer less than the insurer has set out, at exactly the moment they are counting on it.

Even-handed note, accurate handling was also present in the reviewed examples. A review is only as trustworthy as the answers it confirms and the flags it declines to run, so this belongs alongside the findings above. On the same question type that produced the escalation-routing errors described above, other reviewed answers named FIDReC correctly as the Singapore route, with the correct process. Reviewed answers about the public protection scheme's coverage, whether all life insurers belong to it, the CareShield Life disability trigger and MediSave payment, the direction of the new Integrated Shield Plan rider premiums, and the Motor Claims Framework reporting window were answered correctly. Flags that did not hold on checking against the published source were cleared.

Why AI answer accuracy is a governance risk for insurers

None of these answers were written by the insurers or the dispute body, and none of them can be edited the way an institution's own website copy can. That is the difficulty. A Singapore insurer can keep its Singapore complaints page exact, its Singapore service standards current, and its accident-response target plainly stated, and a consumer can still arrive having received a conflicting, misplaced or incomplete account from a system the institution has no contract with and no visibility into. The published record was right. The representation circulating about it was not, and here it was not even wrong in the ordinary way, it was a different country's version of the same institution. And the potential consequences, which include a complaint directed to another market's bodies instead of the route identified for Singapore, a consumer measuring their insurer against a commitment it never made, or someone acting on less precise information than the insurer published about accident response, together with a governance question about information customers may rely on, are the institution's to understand, evidence and manage, regardless of who typed the answer.

Conventional governance controls often focus on information the institution publishes or systems it operates directly, and here the published record is accurate. This risk sits outside the institution's controlled publishing perimeter, where consumers may consult public AI before approaching the insurer or the dispute body, and where a multi-market insurer faces the added hazard that its own other-market pages can be reproduced as if they were local.

The Singapore context makes the issue sharper. The escalation route, the complaint-handling standard, and the accident-response target are all published, public-facing facts, so a consumer has a concrete record to be measured against, and a concrete basis for checking or challenging the answer when the AI restates it wrongly. The certified findings in this reading concern the steps a consumer takes when something has already gone wrong: where to complain, how long it should take, and how fast help arrives, which is the moment a misled consumer can least afford bad information. A complaint sent to the wrong country may defer redress. A timetable reproduced from another market may set a false expectation. A published target left out may leave the consumer unaware of the standard the insurer has set. None is exotic; all are the ordinary business of dealing with an insurer, which is why getting them wrong, invisibly, is worth a risk function's time.

How to govern AI answers about your institution

The answer is not to chase the AI, which cannot be corrected the way an institution's own website copy can, but to govern the surface with the same seriousness applied to any surface through which customers encounter claims about the institution. In practice that is a repeatable loop: watch what the major systems are telling consumers about your complaint and escalation routes, your service standards, and the assistance you commit to; check each answer against the official published fact, your own Singapore pages and the dispute body's page; separate the genuine misstatements from the false alarms before acting on either, and where a flag turns out to be correct, say so; work out why a real one drifted, since a wrong-country escalation route, a borrowed complaint timetable and a dropped response standard call for different fixes, and a multi-market insurer should check specifically whether answers are reproducing facts from its other-market pages in place of its local ones; rank what could actually mislead a consumer ahead of what is merely imprecise; and keep a dated record of what was said and when. Done steadily, an open-ended worry becomes a managed process. The institution can identify and document these issues proactively, review the clarity and market labelling of sources it controls, and show its work to a board, or to a regulator, when asked what it is doing about AI.

Reading history

This page is refreshed in place each month at the same address. Earlier readings are demoted here rather than deleted, so the record stays visible. Each entry below is a standalone summary of that reading. We do not draw comparisons between readings.

First reading (June 2026). A scheme-only reading, checking public AI against Singapore's public insurance and protection schemes. Its certified findings concerned scheme eligibility, dispute limits and scheme scope: a CareShield Life severe-disability trigger understated, the FIDReC adjudication limit overstated, the FIDReC non-injury motor scheme described partly backwards, and the Policy Owners' Protection Scheme's general-insurance protection called unlimited when it is capped. Accurate handling was also present in the reviewed examples.

Second reading (July 2026). A reading covering the public schemes and insurers. Its certified findings concerned a national CPF scheme attributed to a former administrator rather than its current one, a complaint left with no external escalation route named, the Policy Owners' Protection Scheme general-insurance cover stated as having no caps when it is capped, an insurer's complaint-response time given as longer than its published 20-business-day standard, and an interim-cover condition broadened past its terms. Accurate handling was also present in the reviewed examples, including a MediShield Life flag that on checking was correct and was rejected.

Third reading (August 2026). This reading's certified findings are summarised above: a complaint routed to another market's ombudsman with FIDReC absent or omitted, a complaint-handling timetable taken from another market, and an omitted accident-response standard.

We are not drawing a trend line across these readings. We do not yet have a stable, re-asked tracking panel with comparable provenance across readings, so each is a standalone single-month reading. We will consider trend language only when that panel exists.

Read on

This is an early read from an ongoing barometer Lawnise is building on how public AI answers questions about Singapore's insurance sector, refreshed in place each month with more categories to follow over time. If you would like to see what public AI is currently saying about your own institution and the schemes around it, checked, the way these were, against the official published facts, we can scope a private AI answer baseline: sector-context, no obligation.