The State of AI Answer Accuracy in Malaysian Insurance, A Preliminary Barometer

A question asks for the maximum financial loss the Financial Markets Ombudsman Service will accept for an insurance dispute. The answer gives RM250,000, says larger claims cannot be handled and points to litigation or arbitration. The published rules leave another possibility open: referral with both parties' written agreement.

That is the first finding in this reading. The other cases concern a claim-notification deadline that was left out and a general office number presented as a dedicated healthcare line. These are individual answers checked against official sources, not evidence of how often errors occur.

How we checked October's Malaysian insurance answers

The October capture included ChatGPT, Copilot, Gemini, Google AI Mode, Google AI Overview and Perplexity. Questions concerned Malaysian insurance rules and practical service information.

Copilot and Perplexity coverage was degraded by capture failures. We make no comparison of providers or their coverage, and successful responses from an incomplete capture do not establish performance. This reading is qualitative: each featured finding depends on its own answer and source evidence.

We read the full answers and the questions they were responding to, including responses that the automated detector had not flagged. We checked material statements against the relevant official rules, policy document or service page, retained source captures and distinguished a wrong statement from an omission, a qualified estimate and a harmless wording difference.

The featured examples were selected for review, not drawn as a random sample. Private insurers and their identifying contact details are withheld. Public bodies are named where their rules provide the reference point.

What a short insurance answer needs to include

An answer need not reproduce an entire policy. It does need to keep the part of the rule that decides the question.

For FMOS, RM250,000 is the ordinary monetary limit, but the rules include a written-agreement exception. A claim-notification answer needs the default deadline as well as any allowance for late notice. When someone asks for a healthcare service, a real telephone number is not enough if it belongs to a different function.

The check is therefore more specific than asking whether an answer sounds plausible. Does it give the relevant rule, with the qualification the reader needs?

Three findings about disputes, claims and healthcare help

The FMOS limit was made absolute

A question asked for the maximum financial loss FMOS would accept for an insurance dispute. The answer gave RM250,000, then said FMOS could not handle larger claims and that consumers must pursue litigation or arbitration.

RM250,000 is the ordinary limit. Rule 5.1(a) of the FMOS Rules, however, permits a dispute above that limit where the parties jointly agree in writing to submit it to FMOS. [2]

This does not mean a policyholder can require an insurer to agree or that every larger dispute will be accepted. Higher monetary awards also have conditions under Rule 11.22(b), including written agreement involving FMOS and the parties. Other eligibility rules still apply.

The error was the absolute refusal, not the headline figure. It could lead a policyholder to dismiss an available route before asking whether the necessary agreement can be obtained.

A notification deadline was replaced with general advice

A product-specific question asked when a critical-illness claim had to be notified under a master policy. The answer said to notify the insurer as soon as reasonably possible. It mentioned 30 days, but attached that period to survival after diagnosis, not to notifying the claim.

The checked master policy sets a separate default: written notice within 30 days from diagnosis or surgery. It also preserves an exception. Failure to meet that timeframe does not invalidate a claim if giving notice then was not reasonably possible and notice was given as soon as reasonably possible. [3]

The answer described late-notice circumstances without supplying the deadline the question asked for. Survival and notification are different requirements, even when both use a 30-day period.

This is a finding about the particular policy checked, not a deadline for all critical-illness insurance. Nor does it show that a claim was rejected. A reader relying on the answer could miss the default timeframe without understanding when the exception applies.

A general office line was labelled as hospitalisation support

A question asked for an insurer's healthcare hotline for hospitalisation assistance. The answer supplied a real general office number and repeatedly described it as the dedicated hospitalisation and healthcare service line.

The checked official service pages distinguish that general contact from a separate healthcare channel for hospitalisation assistance. The healthcare line is listed as available around the clock; the general contact has weekday office hours. [4] Another reviewed answer made the distinction correctly.

We are not saying the general line could never help or redirect a caller. The narrower error is that the answer labelled one service as another and omitted the channel the question requested. Someone seeking admission assistance may have to make an avoidable extra call.

Accurate handling was also present. A reviewed FMOS answer explained that a dispute above RM250,000 may be referred with joint agreement rather than treating the limit as absolute. A claim-notification answer supplied the 30-day default and the late-notice exception. A healthcare answer distinguished the dedicated service from general customer contact. These examples are reported alongside the findings, without comparing providers.

Why these details matter to insurers

The consequences here are possible, not observed. We checked answers, not what a customer did after reading them.

An absolute refusal could discourage a policyholder from exploring a dispute route. Missing a notification deadline is a different problem: general advice to act promptly does not tell a claimant which timeframe applies. The contact error may be less consequential, but it can still add friction when someone is trying to arrange hospitalisation help.

Insurers cannot edit public AI answers. They can check material answers about their services, keep the relevant source evidence and document where the answer departs from it. That work is distinct from maintaining their own website: the published information may be clear while an answer about it leaves out the detail a person needs.

Checking an insurance answer before relying on it

Start with the actual question. If it asks for a deadline, look for the event that starts it and any exception. For a dispute route, check the forum's scope rather than assuming a familiar monetary figure settles eligibility.

Contact information needs its own check. Match the line or department to the service requested, not just to the institution's name. A valid general number and a dedicated assistance number are not interchangeable descriptions.

Read the whole response before deciding it is wrong. A qualification later in the answer may change the meaning of its opening. Then compare the material statement with the current official source and retain the question, full answer, source and check date. Automated flags are leads for that review, not its conclusion.

This is answer-level examination, not continuous monitoring or a claim to have tested every possible answer.

Reading history

This page is refreshed in place at the same address. Open a previous reading to see the findings checked in that month, the published boundaries used to assess them and the limits of that reading. Each month stands on its own. We do not infer a trend or compare how often errors occurred across readings.

Read on

Lawnise maintains this barometer as an evidence-led view of public AI answers about insurance. To examine selected answers about your institution or market, scope a private AI answer baseline.