Fake reviews: how they work, why they spread, how to spot them

A fake review is any review that misrepresents the reviewer's relationship to the product or service. That covers more than obvious lies. It includes paid reviews, reviews exchanged for free products without disclosure, AI-generated reviews presented as human, reviews left by people who never bought the item, reviews posted by competitors to sabotage a brand, and reviews written by employees or their friends and family.

Genuine criticism — even harsh, even unfair — is not a fake review. The dividing line is honesty about the reviewer's actual experience, not the rating itself.

Independent estimates and regulator findings converge on a striking number: somewhere between 30 and 40 percent of online reviews show signs of manipulation. The UK's Competition and Markets Authority has investigated multiple platforms; the US Federal Trade Commission finalized a rule in 2024 banning fake and undisclosed-incentive reviews, with penalties up to $51,744 per violation. In the EU, the 2022 Omnibus Directive made similar practices illegal across all 27 member states.

Major platforms themselves report removing tens of millions of suspicious reviews every year. That number proves moderation is active — and also measures how much synthetic content open review systems attract in the first place.

There is no single source. The fake-review supply chain is layered, and each layer has different economics and detectability.

Most major review platforms operate an open model: anyone with an email can publish a review for any business. That design choice makes scale possible, and makes verification structurally impossible. Moderation is therefore reactive — content is reviewed after publication, often only when flagged.

Detection is improving. Platforms use behavioral signals (IP clusters, posting velocity, device fingerprints, account age) and increasingly machine learning to filter coordinated activity. But the offensive side is improving faster: AI lowers the cost of plausible content to almost zero, and review farms have moved to mobile networks and residential proxies that look indistinguishable from real users.

No single signal is conclusive. But the combination of several red flags is usually enough to discount a rating before it influences your decision.

Generative AI has changed the economics of fake reviews. What used to require a human writer in a low-wage market now costs fractions of a cent per review. Detection tools exist, but their accuracy on short, edited text remains limited, and adversarial techniques (paraphrasing, controlled noise, mixed human-AI editing) defeat most public classifiers.

The practical consequence: text-based authenticity signals are weakening, and the only durable signal left is whether the reviewer can prove they actually transacted with the brand. That is exactly the gap verified-purchase systems are designed to close.

Better moderation helps. Stricter laws help. But neither resolves the underlying issue, because both operate after the fact. The only structural fix is verification at the source: only count reviews from people who can prove they bought from the brand.

Verified-purchase models trade volume for signal. They accumulate reviews more slowly, but each review is anchored to a real, provable transaction — something open review systems cannot offer by design. This is the model Boxumer is built around: every review is tied to a purchase verified through the user's own connected email or order history, nothing else counts.

We don't claim this eliminates every form of dishonesty. But it eliminates the largest category — reviews from people who were never customers — and that alone changes what "4.7 stars" actually means.

Fake reviews are not going away. The supply side is faster, cheaper and more sophisticated every year, and the platforms most consumers rely on are structurally unable to fully solve the problem.

The pragmatic response is twofold: read open reviews critically using the checklist above, and weight verified-purchase signals more heavily for any decision that genuinely matters. Trust should be earned through provable experience, not borrowed from an aggregate score.