2 in 3 Consumers Trust AI. Only 3 in 100 Say It Helped Them Decide

Picture a 28-year-old woman in Malang. Her skin has been breaking out, so she asks an AI chatbot which serum suits oily, acne-prone skin. In seconds she gets an answer: two ingredients to look for, three products to try.

She does not buy any of them. Not yet. She opens a marketplace and starts reading reviews.

Now imagine that evening she asks the same AI what snack to bring to the office tomorrow. Would she check reviews again? Or would she just buy the first brand it names?

That question sits behind our latest study on how Indonesian consumers use AI when they shop. We surveyed 180 consumers from Jabodetabek. Men and women, SES A and B, aged 25 to 45. We asked how they used AI when shopping in the past three months, how much they trust it, and what they did right after AI gave them an answer.

Consumers Trust in AI: The Numbers

•       62% already use AI for something related to shopping.

•       64% trust AI recommendations — the same as trust in other shoppers' reviews (68%).

•       Only 3% of AI shoppers said AI helped most at the final purchase decision.

•       80% check somewhere else before they buy — a review page, a store, or a person.

Consumers are giving AI the work. They are not giving it the final say.

Be Found, Be Verified, Be Chosen

Right after an AI suggestion, 44% check reviews or a marketplace, 22% visit a store, 14% ask someone they know, and only 18% buy directly. That gives brands three separate jobs:

•       Be Found — AI can find your brand and explain it.

•       Be Verified — your reviews, marketplace page and experts back up what AI said.

•       Be Chosen — the product is there, and the shelf, counter or staff close the sale.

The question is how many of these steps matter in your category.

Higher Involvment Products

For a relatively high-involvement category, it fits what AI is used for most. When we asked AI shoppers where AI helped most, 52% said learning about a brand or product, and 27% said discovering something new, e.g. ingredients, skin type, what not to mix in beauty category — these are learning questions.

In this beauty example, I would expect all three steps to matter.

Be Found: AI must describe your product correctly. A wrong answer about an active ingredient can become a safety or regulatory problem, not just a lost sale.

Be Verified: across all shoppers, 44% check reviews after an AI suggestion — 54% of women, against 31% of men. Beauty is female-skewed and priced high enough to check, so this step is likely to be at least as strong here.

Be Chosen: across all shoppers, store staff, pharmacists and experts still score higher on trust than AI (4.02 vs 3.76 out of 5). Beauty has real experts at the counter.

A weak AI answer could still be rescued by a strong review page or a good advisor. But if your brand is not 'found' - i.e. mentioned in the first stage, then you are not in the consideration set.

Low Involvement Products

Now think about snacks, instant coffee or tea bags — low-involvement categories.

Would anyone read three reviews before buying a Rp 12,000 pack of biscuits? Probably not. My reading is that the checking we measured is driven by risk. When the risk is almost zero, much of the checking may disappear.

That sounds like an easier market. It may not be.

If nobody verifies, Be Verified drops out — and Be Found becomes Be Chosen. The AI answer becomes the decision. No review page. No second chance.

And consumers are unlikely to ask AI about your brand. They will ask about the occasion: what to serve guests this weekend, what coffee to keep in the office. The brand AI names could take the whole occasion.

So here is the question for every brand team: does AI's answer get checked in your category? If yes, keep all three steps consistent. If not, your battle may be won or lost at Be Found — so check what AI says when someone asks about your occasion, not your brand.

What This Study Does Not Show

This is Greater Jakarta, SES A and B only — the most connected consumers in Indonesia. Read these numbers as the upper end of AI use, not the national average about consumers trust in AI

The beauty and snack sections are a hypothesis, built from what we measured plus what we know about how these categories behave. They are not a category finding. The next step is to test the same questions category by category.

Next in This Series

In the next article, I look at a decision with much higher stakes — buying a vehicle — and how far consumers let AI go there.

For now, the picture is clear. AI opens the file. Something else usually closes it. The open question is what happens when the purchase is so small that nobody bothers to check.

(Based on RB Consulting study, 2026: n=180 consumers, SES A/B, aged 25–45, Greater Jakarta. Journey and behaviour questions based on 112 consumers who used AI for shopping. Questions covered shopping in general, not specific categories.)