Brands are now using AI-generated "influencers" and fake customer reviews that you genuinely cannot tell from the real thing. A 2025 study (University of Nottingham) found people spotted fake AI reviews only 50.8% of the time — the same as a coin flip — and the AI detectors built to catch them did no better. A peer-reviewed paper the same year reached the same conclusion. So stop trusting how "real" a review sounds. It tells you nothing.

Spotting a fake face (still doable, about 60 seconds)

Fake accounts are easier to catch than fake text. Before you trust an "influencer":

  • Reverse-image-search the profile photo (Google Lens, TinEye). If the same face appears under other names, walk away. (Caveat: AI-generated faces often return zero matches — a blank result isn't proof it's real, just one data point.)
  • Look for genuine live video. Real people go live and move in real time; many synthetic personas can't, or won't.
  • Check whether they ever reply to comments — or exist anywhere outside their own polished feed.
  • Old image tells (weird hands, mismatched backgrounds) are fading fast. The more durable giveaways are physics: lighting, shadows, and reflections that don't line up.

Defending against fake reviews (read the pattern, not the prose)

Since the words can't be trusted, use the structure around them:

  • Filter to "verified purchase" — but treat it as a filter, not a guarantee; verified-purchase fakes exist.
  • Read the 3-star reviews. The middle ratings tend to be the most honest.
  • Watch for clusters — a burst of 5-star reviews on the same day, or the same phrases copy-pasted across different products.
  • Cross-check a second site. A product adored in one place and ignored everywhere else is a flag.

The bottom line

You can no longer read your way to the truth — not a review, not a single polished image. What still works is reading the patterns: where a face turns up, when reviews land, whether anything exists outside the feed. Trust the pattern, not the polish.

Receipts: 2025 study, University of Nottingham (humans 50.8% / chance) plus a peer-reviewed paper in the Journal of Retailing and Consumer Services (2025); a Guardian investigation (June 2026) on AI influencers; Consumer Reports / BBB review-spotting guidance.