The recommendation in the family WhatsApp group came from ChatGPT

It starts as an ordinary Tuesday-night question in a family WhatsApp group: does anyone know a good confinement centre near Kota Damansara? Someone has a cousin to ask. Someone else pastes a shortlist from ChatGPT. The auntie who has actually visited one of the places gets the final word.

This is a plausible scene, not a measured national habit. The distinction is important because the available Malaysian data proves something broader, and less dramatic, than the story.

The Malaysian data shows that asking AI is already normal

A Vodus study published in January 2026 reported that 67% of Malaysians had used an AI tool in the previous three months. Among those users, 81% said they used AI to search for information (Vodus, AI Consumption in Malaysia).

That is strong evidence of general use and information-seeking. It is not a count of people asking for clinics, restaurants, contractors, or confinement centres.

The study also does not tell us which AI product handled the question, which language was used, whether the answer reached WhatsApp, or whether anybody eventually called a business. Saying otherwise would turn one broad statistic into a much narrower finding the research never made.

AI does not have to replace word of mouth to influence it

The interesting possibility is not that Malaysian families suddenly stop asking one another. It is that an AI shortlist arrives before the familiar conversation has finished.

A name introduced by ChatGPT may then be checked against a sister's experience, a neighbour's warning, Google Maps, or the family member who always knows somebody. The human network still decides what deserves trust. The system has simply placed a few names on the table.

For a local business, that earlier moment matters. A referral may remain the strongest reason someone calls, while an AI answer influences which unfamiliar names are allowed into the comparison. Both can be true at once.

This is why “AI has replaced referrals” is the wrong claim. A more useful question is whether AI has started appearing inside the referral journey for your own customers.

The shortlist has no single universal source

ChatGPT or another assistant generates the answer according to the product, feature, tools available, and wording of the request. It may search the current web, use learned information, consult connected sources, or combine several inputs.

When the answer shows citations, open them. Check whether they support the business description, location, and service being recommended. When it shows no sources, leave the source as unknown; confidence is not a substitute for a link.

A website, directory, news article, listing, review page, or other public reference may matter in a particular system and query. None should be sold as the universal key to every shortlist.

A Malaysian customer may change language halfway through the question

The same household may search in English, ask a follow-up in Bahasa Melayu, and use a Chinese business name copied from WhatsApp. The resulting answers can differ even when the underlying need has not changed.

A fluent Malay response does not prove that the sources were Malaysian or written in Malay. It proves only that the system produced Malay text. A proper Bahasa Melayu comparison keeps the prompt, answer, citations, engine, and date together so the difference can be inspected.

EasyFound tracks Bahasa Melayu prompts as their own language rather than using an English result as a proxy. That matters because a business can preserve the exact Malay question and compare its answer with the English version without mixing the two result sets.

Your first useful evidence may already be in your enquiries

Ask new customers how they first heard the name, then listen for the route rather than forcing one channel into the answer. “My sister sent me a ChatGPT list” is different from “I found you on ChatGPT,” and both are different from a direct Google search.

Save the customer's words, the date, and any screenshot they volunteer. After several months you may find a pattern. You may also find that AI-assisted discovery barely appears in your business. Either result is first-party evidence; neither needs a national statistic stretched to fit it.

In parallel, keep the public basics right. Your service, location, opening hours, contact details, and important limits should be easy to find in the languages your customers use. That helps the person checking the shortlist even when the AI answer played only a small part.

For a deliberate test, turn three recent enquiries into questions and use the five-minute checking method. If your customers ask in Malay as well as English, request a free visibility check and include both versions in the baseline.

Sources

• Vodus, AI Consumption in Malaysia Study (vodus.com), January 2026

The Vodus fieldwork ran from 29 December 2025 to 2 January 2026. Its figures describe AI use and information search generally; they do not isolate local-business recommendations.