How to spot fake reviews: a practical detection checklist

Volume anomalies

Genuine review streams follow a slow, transaction-paced rhythm. A 12-month-old business that suddenly receives 80 five-star reviews in 14 days is almost always being boosted. Check the histogram, not just the average.

Language fingerprints

AI-generated reviews tend to over-use balanced sentence structure, generic praise ('great service, highly recommend') and zero specifics. Real customers name a staff member, a product, a date or a problem.

• No proper nouns or named staff
• No mention of price, time or location
• Identical opening across multiple reviews
• Reviewer profile with one review and no photo

Reviewer-graph signals

When ten reviewers who all joined the platform in the same week have only ever reviewed the same three businesses, you are looking at a coordinated network. ScoreReview UK calculates this graph daily and quarantines suspect clusters before they reach public profiles.

Sentiment-rating mismatch

A four-paragraph complaint about delivery time ending with five stars usually means a copy-paste mistake by a broker. Our classifier flags any review where sentiment polarity diverges from the rating by more than two standard deviations.

Verification provenance

The single highest-signal field is provenance. Reviews tied to an invitation token, a receipt or an identity check are dramatically harder to fake. Always check the verification tier before trusting a star count.

What a consumer can do in 30 seconds

Sort by lowest rating first — businesses that suppress negatives will show a strangely empty tail. Read the business replies, not the reviews; thoughtful replies correlate with real operations.

Why platforms have a duty here

The FTC final rule and the UK Digital Markets, Competition and Consumers Act both place direct liability on platforms that fail to take reasonable steps to detect fake reviews. 'We just host them' is no longer a defence.