How fake reviews are made, and how to spot them

The thread that gives them away is not the wording, it is the shape. Faking a real reputation means faking the pattern of real custom, and that leaves marks, in the timing, the spread and the reviewer profiles. The newest twist is that you can no longer tell a fake from a real one by reading it, so the defence has moved entirely to the pattern.

How they are produced

• Review farms: organised, often offshore networks of paid posters, recruited through social and messaging apps, that sell reviews in bulk with discounts and "replacement guarantees" if some get removed.
• Rebate-for-review: a real person buys the product at full price, earning the "verified purchase" badge, posts a glowing review, and is quietly reimbursed afterwards, which launders the payment off the platform.
• AI-written: language models now generate convincing reviews, with plausible personal detail, for products no one ever bought.

Why you cannot just read for it

The single most important finding here is that reviews written by current AI are, in one 2025 preprint (not yet peer-reviewed), hard to tell apart from real ones to both ordinary readers and the automated detectors built to catch them. So any tool that scores a review as "fake" or "real" is an input to judgement, never a verdict, and the reliable signals are structural: a sudden cluster of reviews from a standing start, a flawless spread with no honest middle, thin or brand-new reviewer profiles, and vague praise that could describe anything. Each of those is a question, not proof, and several of them together, across different categories, is when you slow down.

The scale, and the limit

The platforms now remove fakes at industrial scale, hundreds of millions a year between them, much of it caught automatically, which tells you both how big the problem is and how seriously it is being policed. The limit on all of it: detection is imperfect in both directions, real businesses sometimes get caught and fakes sometimes survive, so treat every signal as a prompt to look closer, and lean on the harder thing software cannot fake, the pattern over time and the proof a business can put in your hand.