You’re about to buy a £200 air fryer. It has 4.8 stars from 3,000 reviews. Sounds safe — except a growing share of those reviews may have been written by AI: fake, fluent, and utterly convincing. So are AI-generated product reviews trustworthy? Increasingly, the answer is: less than you think. Here’s how to spot the fakes and find opinions you can actually rely on.
How Big Is the Problem?
Fake reviews aren’t new — paid review farms existed for years. What’s new is scale and quality. AI can generate thousands of unique, natural-sounding reviews in minutes, each slightly different, each hitting the right keywords, each free of the broken English that used to give fakes away. Sellers buy them in bulk; competitors buy negative ones to attack rivals. Studies and investigations (including by the consumer group Which?) repeatedly find significant proportions of reviews on major platforms are suspicious — and AI has made the problem worse while making detection harder.
The UK has started fighting back: since 2024, fake reviews are explicitly illegal under consumer protection law, with the Competition and Markets Authority able to fine offenders. But enforcement against thousands of anonymous AI-generated reviews is like bailing the ocean with a teacup. Self-defence remains essential.
Why AI Reviews Are So Convincing
Modern language models write reviews that hit every trust signal humans look for:
- Specific details: “the non-stick coating survived six months of daily use” — plausible specifics that were never experienced.
- Balanced tone: “four stars because the beep is loud, but otherwise perfect” — the minor criticism that makes it feel authentic.
- Personal voice: first-person anecdotes, casual language, even typos inserted deliberately.
- Variety: no two alike, defeating simple duplicate detection.
They’re also cheap enough to deploy at volumes that drown out genuine reviews. A product with 200 real mixed reviews and 3,000 AI-generated five-star reviews looks like a 4.8-star winner. The maths of trust breaks down.
How to Spot AI-Generated Reviews
No method is foolproof, but these signals help:
Check the review distribution. Genuine products usually show a natural curve — mostly 4–5 stars, a meaningful tail of 1–3 stars. Suspicious patterns: overwhelmingly 5-star with almost nothing else, or a suspicious cluster of reviews posted in a short window (launch-day review bombs).
Read the 3-star reviews. Fake-review operations focus on 5-star praise and 1-star attacks. The middle reviews — balanced, specific, boring — are most often genuine. They’re also the most informative.
Look for generic vagueness. AI reviews often describe the product in terms anyone could write from the listing: “great quality, fast delivery, highly recommend!” Genuine reviews mention specific, lived details: “the left hinge squeaks after two weeks”, “doesn’t fit under my 72cm counter”.
Check reviewer profiles. Click through: does this reviewer only review one brand’s products? Dozens of 5-star reviews posted on the same day? No profile history? All red flags.
Use detection tools. Browser extensions like Fakespot (now part of Mozilla) analyse review patterns and give adjusted ratings. They’re imperfect but useful as a second opinion.
Be wary of “review for discount” schemes. Reviews incentivised by sellers skew positive by nature — and since 2024, undisclosed incentivised reviews are illegal in the UK. Treat clusters of gushing reviews mentioning “I received this at a discount” with scepticism even when disclosed.
Where to Find Trustworthy Opinions Instead
- Professional reviewers: Which?, Wirecutter, Trusted Reviews, RTINGS — organisations that buy products anonymously and test methodically. Not infallible, but accountable.
- Long-form user reviews: detailed reviews with photos of actual use, especially critical ones, are hard to fake convincingly at scale.
- Forums and communities: Reddit, specialist forums, Facebook groups — real users discussing real experiences, with community BS-detection.
- Friends and family: unfashionable, but still the most trustworthy source for many purchases.
- YouTube reviews from established creators: harder to fake than text, and reputational incentives keep most honest.
This connects to the broader challenge of AI-generated misinformation — reviews are just one front in a larger authenticity crisis.
What Platforms Should Do (and Are Doing)
The responsibility shouldn’t rest entirely on shoppers. Platforms are deploying AI detection of their own — the irony of AI fighting AI — alongside verified-purchase requirements, review velocity monitoring and legal action against review sellers. Amazon has sued fake-review brokers; Google has taken action against businesses gaming reviews. Progress is real but slow, and the economic incentives (more reviews = more sales = more platform fees) create a permanent conflict of interest.
Regulation helps at the margins: the UK’s fake-review ban gives the CMA teeth, and the EU’s Digital Services Act imposes transparency obligations. But technology will keep evolving on both sides — which is why understanding how recommendation systems work matters too, since those systems decide which reviews you see first.
The Takeaway
AI-generated product reviews are increasingly common, increasingly convincing, and not trustworthy at face value. Protect yourself by reading mid-range reviews, checking reviewer histories, using detection tools, and leaning on professional testers and real communities for big purchases. A healthy scepticism about star ratings isn’t cynicism — it’s just accurate. The five-star average means less than it used to; the detailed three-star review means more.
AI-Generated Product Reviews: FAQs
What percentage of online reviews are fake?
Estimates vary wildly and depend on platform and product category — studies have found anywhere from 5% to 30%+ suspicious reviews in sampled products. High-competition categories (supplements, electronics accessories, beauty) tend to be worst. The honest answer: enough that you should never rely on star ratings alone for significant purchases.
Are incentivised reviews always bad?
Not necessarily — a reviewer who got a free product but writes honestly is different from a paid five-star farm. The key is disclosure (legally required in the UK) and balance: incentivised reviews skew positive even when honest, because dissatisfied recipients often just don’t review. Treat them as marketing-adjacent, not independent opinion.
Should I trust reviews on the seller’s own website?
Less than independent platforms. Sellers control which reviews appear on their own sites — negative ones mysteriously vanish. Third-party platforms (Trustpilot, Google, Amazon) have their own problems but at least the seller doesn’t hold the delete button. Cross-reference: if a product has 500 five-star reviews on the seller’s site and 3.2 stars elsewhere, believe elsewhere.
What’s the single best review habit?
Read the three-star reviews first. They’re where the honest trade-offs live: “good but…”, “works well except…”. Five-star reviews tell you what’s best; one-star reviews tell you what can go wrong; three-star reviews tell you what it’s actually like to own. Ten minutes with the middle of the distribution beats an hour with the extremes.
