AI Reads the Reviews for You: Amazon Quietly Changed the Underlying Rules of E-commerce Shopping

A while back, I was buying a coffee machine on Amazon.

I opened the page — hundreds of reviews, packed wall to wall. I was about to do what I always do: scroll from the first one to the last, squinting through every single line.

Then I noticed a line of text.

"Customers say" — AI had already read through thousands of reviews for you, distilling them into a handful of keywords: easy to use, reliable, easy to clean, each with a little checkmark next to it. Click one, and it jumps straight to the relevant reviews.

Right then, I knew this was no small thing.

93% of People Are Doing the Same Thing

Let's step back for a moment.

In 2021, Kantar ran a survey. 93% of consumers read reviews before placing an order. 85% said reviews were a key factor in their decision-making.

See, reviews aren't reference material anymore — they ARE the decision.

But here's the problem. The more reviews there are, the murkier it gets. Amazon's search rankings are directly tied to review count and positive review rate. Some sellers got crafty, farming fake positive reviews, piling on fraudulent ones. Amazon handed out bans, cracked down, wave after wave. But as long as the ranking mechanism stays the same, review manipulation won't stop.

Amazon came up with a move that cut the problem off at the root: instead of making people read review after review, let AI read them for you.

What Is a Generative AI Review Summary?

That's what I saw on the coffee machine page.

It's not written by a real person, and it's not AI pretending to be human. What it does is feed hundreds or thousands of reviews into an algorithm, let the AI extract the common threads, and serve you a condensed version.

How's the performance, is it easy to use, is it stable — information that used to take you half an hour to piece together, AI lays out for you in three seconds. Negative and mediocre reviews that got buried are fished out from the corners and placed where you can spot them at a glance.

Amazon, Taobao, Dianping, Agoda — they've all rolled out similar features. The logic is the same: there are too many reviews, people can't keep up, so let the machines handle it.

But Does It Actually Work?

Which brings up a question.

Do AI-generated review summaries genuinely help consumers make better decisions, or do they just look impressive without delivering real value? Some people think they solve information overload. Others find them too mechanical, lacking warmth. Still others worry about AI hallucinations — confidently fabricating a product feature that doesn't exist at all.

Arguing about it won't settle anything. In 2024, a study set out to tackle this question empirically. 401 valid questionnaires, all from people who habitually read reviews when shopping online. The researchers used a classic theoretical framework — ELM, the Elaboration Likelihood Model — to break it down.

ELM says humans process information via two routes. One is the central route, which looks at the quality of the information itself; the other is the peripheral route, which looks at the credibility of the information's source. Applied to AI reviews, that boils down to two questions: is the review well-written, and is it trustworthy?

Four Things, All Indispensable

The researchers broke generative AI reviews into four dimensions, mapping neatly onto those two routes.

On the central route, there are two: review quality and emotional content. Quality refers to relevance, objectivity, and level of detail; emotion refers to whether the review contains something that resonates — cold or warm.

On the peripheral route, another two: review length and review credibility. Length is whether the AI summary is comprehensive enough; credibility is whether you believe it.

The result? All four dimensions had a significant positive effect on consumers' purchase decisions. And all of them worked through a single mediator — perceived usefulness.

Put simply: when consumers judge whether an AI review is good, it ultimately comes down to one question — "Is this useful to me?" High quality, emotionally resonant, right length, credible — consumers find it useful; and when they find it useful, they're more willing to place the order.

The numbers are solid. All four dimensions had correlation coefficients above 0.8, with p-values below 0.01.

A Surprising Finding: Trust Only Moderates Length

The most surprising part of the study, to me, was the set of hypotheses about "trust."

The researchers originally thought: whether a person trusts AI should affect their judgment across all dimensions of AI reviews, right? It doesn't.

Trust propensity only plays a role in one place: review length.

For people who trust AI, longer reviews feel more useful. They assume the length comes from algorithms crunching massive datasets, so the information is more complete, worth spending time reading carefully. For people who don't trust AI, long reviews feel excessive — they suspect it's fabricated, and perceived usefulness goes down instead of up.

But for quality, emotional content, and credibility, whether you trust AI or not makes no significant difference.

Why?

The researchers offered an explanation. Whether the quality is good or whether the emotion resonates — these are things people can sense using their own judgment, without needing to first trust AI. As for credibility, consumers mostly look to the platform for that — I trust Amazon as a platform, so I trust the AI reviews it produces.

In other words, some judgments rely on your own brain, and some rely on trust in the platform. Trust propensity toward AI is not a universal switch.

AI Reviews Are No Panacea

The study also flagged several unavoidable issues.

AI can carry bias. If the training data contains gender, racial, or cultural biases, those biases get amplified, and consumers receive distorted information. Worse, some might exploit the characteristics of generative AI to mass-produce fake reviews and inject them into the data source — far harder to detect than old-school manual review farming. Add AI hallucinations on top of that, and fabricated product features you didn't notice might only come to light after you've already placed the order.

Cognitive load is another factor. Too much information, too long, and consumers' brains can't hold it all — they end up unable to make a decision at all. If the solution to information overload is handled poorly, it just becomes a new round of information overload.

What Has This Actually Changed?

I thought about it, and I realized the real significance isn't in AI reviews themselves.

What's changed is the relationship between consumers and product information.

For the past decade-plus, e-commerce review systems have essentially been a UGC product — users writing references for other users. The system works, but it has inherent flaws: paid shills, review manipulation, information overload, extreme reviews drowning out measured assessments.

What generative AI review summaries do is add another layer on top of this UGC system: let AI read through the raw data for you first, then hand you the conclusion.

This is a shift in medium. It goes from "person-to-person conversation" to "AI reads everyone on your behalf."

So the most critical finding in that study is actually the "perceived usefulness" mediator. Consumers don't care how advanced the AI review's technology is or how complex the algorithm is. They care about one thing: does this actually help me make a decision?

If it helps, they buy. If it doesn't, they scroll past.

For e-commerce platforms, what does this mean? It means the AI review feature can't be window dressing. Quality must be high, information comprehensive, and length tailored to the individual — detailed versions for users who trust AI, condensed versions for those who don't. LazzieChat, built by Lazada, follows exactly this logic: it uses a conversational interface to understand user preferences, then delivers personalized recommendations, compressing the shopping flow.

Transparency has to keep up too. Label clearly that this is AI-generated, show the data sources, explain the algorithm's logic — the more you hide, the less consumers trust you.

In the End

I bought that coffee machine on Amazon.

Read the AI-summarized reviews, then clicked through to verify two of the original reviews. Placed the order — the whole thing took less than ten minutes.

Ten years ago, doing the same thing would have taken me at least forty minutes — scrolling through reviews, comparing, hesitating.

I don't know whether AI reviews are good or bad. But I do know one thing: when consumers' decision-making time gets compressed to a quarter of what it used to be, everyone in e-commerce and marketing needs to rethink where they stand.

The answer might be hiding inside another question: after AI has read all the reviews for you, what irreplaceable value do you have left?

That's a question worth serious thought for anyone in the consumer business.