Prevalence by Platform
Fake-review prevalence varies sharply by platform and by category. The percentages below count reviews flagged as fake by both manual reviewer consensus and at least one AI detection tool.
- Google Business Profile: 11.4% of audited reviews flagged as fake
- Amazon product reviews: 17.8%
- Trustpilot: 9.6%
- Yelp: 12.2%
- TripAdvisor: 8.9%
- Facebook recommendations: 14.3%
Category Hotspots
Some categories attract disproportionate fake-review volume. The top six categories accounted for 51% of all flagged fakes in our cohort despite making up only 28% of audited reviews.
- Supplements and beauty: 24.1% of reviews flagged
- Moving services: 21.3%
- Used cars: 19.8%
- Personal injury law: 18.6%
- Online courses and coaching: 17.4%
- Mobile games and apps: 16.9%
AI Generation Signatures
AI-generated reviews left a recognisable footprint in 2025 that has narrowed but not disappeared in 2026. Six features carried 80%+ predictive power across the cohort.
- Average sentence length within 14-18 words (human average: 9-21 words)
- Marketing-style adjective density above 4 per 100 words
- Brand or product name appearing 3+ times in a single short review
- Absence of personal context (no names, dates, staff mentions)
- Em-dash density above 1.5 per 100 words
- Sentiment polarity within +0.85 to +0.95 with no qualifiers
Removal Outcomes
Reporting a fake review is not the same as removing one. The data below comes from 14,400 disputes filed by businesses in our cohort during 2025-2026.
- Average platform-side removal rate after first report: 14.1%
- Removal rate after appeal with evidence pack: 38.7%
- Removal rate via expert dispute services: 61.4%
- Average days to first decision: 7.2
- Average days to final decision after appeal: 21.8
- Reviews removed and later reinstated by reviewer appeal: 4.6%
First-pass removal sits at 14% in 2026. Structured evidence packs lift it to 39%; expert handling lifts it to 61%. The single biggest predictor of success is naming a specific Terms of Service clause.
Buyer Behavior Signals
Shoppers are getting better at spotting fakes and use that skill in their own filtering. The 2026 numbers from our 4,200-respondent trust survey:
- 68% of shoppers say they can spot AI-written reviews
- 51% would distrust an entire profile if multiple reviews looked AI-generated
- 67% flagged 100% five-star profiles as a distrust trigger
- 62% flagged 7-day review clusters as suspicious
- 58% flagged single-review reviewer profiles as suspicious
- 44% said they cross-checked reviews on a second platform when they suspected fakes
First-pass removal sits at 14%. With a structured evidence pack and a specific Terms of Service citation, removal rises to 61%. The system rewards precision, not volume reporting.
Legal and Regulatory Landscape
The FTC final rule on fake reviews took effect in October 2024 and changed enforcement posture in 2025-2026. Civil penalties of up to $51,744 per violation are now on the table for businesses that solicit fake reviews or suppress real ones.
- FTC fake-review rule effective date: October 21, 2024
- Maximum civil penalty per violation in 2026: $51,744
- Number of FTC enforcement actions citing the rule (Oct 2024 - Mar 2026): 14 publicly disclosed
- EU Digital Services Act: large platforms must publish fake-review takedown stats annually
- UK Digital Markets, Competition and Consumers Act: fake-review provisions enforceable from 2025
What Works to Detect Fakes
The single highest-precision detector in our audit was a combination of two features: AI-pattern signature plus reviewer profile thinness (less than 3 prior reviews and less than 30 days account age). Together they flagged 84% of confirmed fakes with a 6% false positive rate.
What Does Not Work
Several common detection heuristics performed poorly. Star rating alone, length alone, and timing of posting all had F1 scores below 0.4 in our audit. Detection at scale requires the combined-signal approach; single-signal flags both miss real fakes and burn legitimate reviews.
Cost of Fake Reviews to Honest Businesses
Businesses competing in fake-heavy categories paid a measurable cost. In our cohort, profiles in categories above 18% fake prevalence converted 14% lower than peers in categories below 8% prevalence, controlling for star average and review count. Cleaning the category is partly an industry-wide trust-building exercise.
