Who's Reading the Reviews? The AI, Not You

Published October 4, 2026

There used to be a ritual: open a product page, scroll past the star average, and start reading. The glowing five-stars, the furious one-stars, the honest three-stars where the truth usually hides. For twenty years, that was how online shopping worked. Increasingly, nobody does the reading anymore — the reading is done for you, by the platform itself. Amazon stamps an AI-written paragraph at the top of its product pages summarizing what thousands of reviewers said. Google's AI answers quote Reddit threads directly in search results. Yelp built an assistant that reads 330 million reviews on your behalf. The reviews haven't gone away. They've been promoted — the crowd still speaks, but an AI now decides what you hear.

From star averages to AI verdicts

The shift started in August 2023, when Amazon rolled out AI-generated customer review highlights: a short paragraph on the product page summarizing the sentiment and features reviewers mention most, with key product attributes — "ease of use," "battery life" — as clickable buttons that surface the matching reviews. Amazon's director of community shopping, Vaughn Schermerhorn, framed it as answering a long-standing customer need: nobody wants to read 4,000 reviews to learn whether a blender is easy to clean.

Google and Yelp took the same logic further in 2026. In May, Google updated AI Overviews and AI Mode with an "Expert Advice" section that surfaces quotes from Reddit and forums directly inside the AI's answer — so a buyer may never click through to the original discussion. Reddit, in turn, began testing its own AI shopping experience in February, with product carousels and buy buttons drawn from community discussions. Yelp launched Yelp Assistant in April: a chatbot that reads across the company's 330 million reviews, recommends local businesses with explanations of why each fits, then lets you book or order inside the app. CEO Jeremy Stoppelman's pitch was blunt: the bot reads 500 reviews in a second; a human might read the first five.

The numbers behind the shift

PYMNTS Intelligence found that 48% of online shoppers used AI during their most recent purchase. AI-referred traffic to US retail sites grew 393% year over year in the first quarter of 2026. ChatGPT's share of product research climbed from 2% to 30% in two years. The review economy became the recommendation economy: the AI's synthesized answer is the last stop before the buy button.

The platforms can feel the ground moving. Reddit's chief operating officer, Jen Wong, told Cannes Lions 2026 that "reading honest, real-world experiences is now a more important factor in the purchase decision than reviews from professional critics or influencers" — and 71% of its users visit during the consideration phase of a purchase. Meanwhile, Google's earlier data-licensing arrangement with Yelp collapsed after Google began summarizing restaurant recommendations in ways that gave consumers less reason to visit Yelp. Whoever controls the AI's answer controls the sale, and everyone knows it.

The laundering problem

Here is the catch, and it is a big one: an AI summary is only as honest as the reviews it summarizes. And the reviews are, increasingly, not honest.

Researchers at Pangram Labs analyzed nearly 30,000 reviews across 500 Amazon best-sellers and found about 3% were AI-generated — concentrated exactly where parents spend the most trust: baby products had the highest share, at 5.2%. And 74% of the AI-written reviews gave five stars, compared with 59% of human reviews. Separate Cornell research found that machine-generated fake reviews defeat human detection and — the part that should alarm you — score higher on perceived usefulness than authentic reviews in controlled tests. The fakes don't just slip through. They read as the best reviews.

And the fakes are getting industrial. In September 2026, Singapore's Competition and Consumer Commission announced its largest fake-review enforcement ever: a provider called Reputifly that trained its AI specifically on real Google reviews so its output would defeat platform detection, recruited human proxies through Telegram to post from genuine accounts, and sold packages like 35 five-star reviews over seven days for S$219 (about $172 USD). The scheme ran across six platforms at once. Forty-five businesses admitted buying the service and must now publish public apologies for six months — on their websites, social media, and physical storefronts — while Reputifly donates its proceeds to charity.

The point is that every fake review it seeded now lives inside the AI summaries and recommendations that shoppers trust. A fabricated five-star doesn't just fool one reader anymore — it gets laundered into the platform's official verdict. For the FTC angle, see our fake-review crackdown explainer.

What the summaries hide — and what the counters are doing

Even with perfectly honest reviews, summaries flatten. An AI paragraph will tell you reviewers love a vacuum's suction; it may not tell you that the dozen people who bought the second-generation model say the battery dies in a year. Minority complaints — the specific, experience-based warnings that made reading reviews worthwhile in the first place — are exactly what summarization smooths away.

The counters are moving, slowly. In an April 2026 safety update, Google said it had acted on 292 million reviews, 79 million edits, 782,000 accounts, and 13 million business profiles during 2025. In May 2026, University of East London researchers published a "hybrid fusion" detection model that combines language analysis with behavioral signals — like whether a review's emotional tone matches its star rating — hitting 93% accuracy on Amazon review data and 91% on Yelp. The long-term direction, per Trustpilot's own reporting, is away from judging review text and toward verifying that a reviewer actually bought the thing.

But notice who benefits from the AI summary layer even when it works as intended: the platform. Summaries keep you on the platform's page instead of clicking through to the originals. And the incentives are crooked — Amazon's AI writes the verdict from reviews Amazon hosts, on pages where Amazon sells the product.

How to read reviews in the AI era

Use the summaries — but treat them as a first draft, not a verdict:

Click through to the underlying reviews. Amazon's highlights link attributes to the reviews behind them. If the summary glows and the recent reviews don't, believe the recent reviews.
Read the 3-star reviews. They're where balanced buyers describe trade-offs instead of emotions. Our guide to spotting AI-generated fake reviews has more tells for the bad ones.
Sort by most recent. A 4.8 average built in 2022 can mask a product that changed suppliers in 2025. Everything drifts toward inflated averages — see our rating inflation explainer — so recency is your best filter.
Cross-check across platforms. The Singapore case showed one provider seeding six platforms at once. Independent voices tend to show up in more than one place.

Go deeper with AI

This article is the starting point. Copy any of these prompts into your favorite AI assistant to learn more about AI review summaries and fake reviews:

🔎 Stress-test a product's reviews before you buy

I pasted below the most recent 20 reviews for [product name] from Amazon, plus the platform's AI summary. Act as a skeptical auditor: flag any reviews that look AI-generated or fake (watch for generic praise with no specifics, suspiciously similar phrasing, five-star ratings that never mention a downside), then tell me what the real complaints are and whether the AI summary matches them or hides them.
[PASTE REVIEWS HERE]

🧭 Compare two products without trusting the star average

I am choosing between [product A] and [product B], both rated around [X] stars. I pasted the 1-star, 2-star, and 3-star reviews for each below. Ignore the star averages entirely — summarize what actually goes wrong with each product in everyday use, quote the most specific complaint you find for each, and recommend the better pick for someone who uses it [describe your use case].

💡 Cross-check an AI summary the way researchers do

Below is an AI-generated review summary from [Amazon / Google / Yelp] for [product or business], followed by 30 of the underlying reviews. Check the summary for accuracy: list each claim it makes, mark which claims are supported by multiple real reviews and which appear in none, and flag any polished-sounding five-star reviews that might be AI-generated fakes laundering the summary.

Tip: replace the bracketed parts with your own situation — the more specific your prompt, the more useful the answer.

Summary

Reviews haven't disappeared — they've been absorbed into an AI layer that reads them, summarizes them, and hands you a verdict. In 2026, nearly half of online shoppers used AI during their most recent purchase, and platforms from Amazon to Yelp now compete to own the synthesized answer. The problem: summaries are only as honest as their inputs, and AI-generated fake reviews are rising, score higher on perceived helpfulness than real ones, and get laundered straight into the platform's official answer.

The single most important takeaway: never let the AI's verdict be your last look — click through to the reviews behind the summary, especially the recent and middle-star ones, because a machine that reads 500 reviews for you can't tell which of them were never written by a human.

FAQ

When did Amazon start summarizing reviews with AI?

August 2023, with "AI-generated customer review highlights" — a short paragraph on product pages summarizing reviewer sentiment and frequently mentioned features, plus clickable attribute buttons (like "ease of use") that surface the matching reviews.

Do AI review summaries include fake reviews?

They summarize whatever reviews exist on the platform, fakes included. Cornell research found AI-generated fake reviews score higher on perceived usefulness than authentic ones, so a summary can end up built on the most polished fakes.

Can I still read the original reviews?

Yes — Amazon's highlights link to the underlying reviews by attribute, and Yelp and Google reviews remain browsable beneath their AI layers.

What was the Singapore Reputifly case?

In September 2026, Singapore's competition watchdog announced its largest fake-review enforcement: provider Reputifly trained its AI on real Google reviews to beat detection, posted fakes through human proxies across six platforms, and sold 35 five-star reviews over seven days for S$219. Forty-five businesses admitted buying fakes and must display public apologies for six months.

Are AI shopping assistants biased toward their own platforms?

Structurally, yes. Google's AI answers keep you on Google; Yelp's assistant keeps you on Yelp; Reddit's shopping experience keeps you on Reddit. The summary is a convenience, but it also removes your reason to visit the original sources.

Sources

← Back to the rating wall