Why Google deletes AI-generated customer reviews (and how detection actually works)

A restaurant owner messaged us last month, worried that three of his reviews had vanished overnight. He had not bought them and had not asked anyone to fake anything. What had happened was simpler and a little embarrassing: a well-meaning regular customer had used an AI tool to "polish up" a quick review before posting it, and the polish is exactly what gave it away.

Google never confirmed a single named "AI filter," and that is the point

A lot of what you will read elsewhere describes a specific new tool, with a name and a launch date, built solely to hunt AI-written reviews. Treat those claims carefully. Google's own published policy for Maps and Business Profile content does not name any such system. What it does say, and has said for years, is that content must reflect a genuine, actual experience, and that Google uses a mix of automated systems and human review to enforce that across an enormous volume of contributions.

That is not a small detail. It means the rule was never really about AI. Reviews get removed for failing the "genuine experience" test, and AI-written reviews fail it at a higher rate for reasons that have nothing to do with Google specifically hunting for them.

Why AI-written reviews get caught anyway

None of these require Google to have built anything exotic. They are the same signals that have caught fake and paid reviews for a decade, and AI-written text happens to trip several of them at once.

1. The phrasing repeats itself in ways real people do not

Ask five different AI tools to write a five-star restaurant review and you will get five variations of "exceeded my expectations," "attentive staff," and "will definitely be back." Real customers write about the specific thing that stuck with them, the mutton roll, the fifteen-minute wait, the uncle at the next table who wouldn't stop talking. Genuine reviews are lumpy and specific. Generated ones are smooth and generic, and smoothness at scale is exactly what content-moderation systems are built to notice.

2. The timing and volume look coordinated

A real business gets reviews in a messy, uneven trickle that roughly tracks its actual footfall. A sudden burst of similarly-worded five-star reviews within a short window, especially from accounts with little other activity, is one of the oldest fake-review signals there is. AI makes it trivially easy to produce a burst like that, but the burst itself is what gets flagged, not the tool that produced it.

3. The account behind the review has no real history

Google weighs a reviewer's overall activity, not just the one review. An account created recently, with no other reviews, photos or Maps activity, posting a suspiciously polished paragraph, reads very differently to a detection system than a regular Maps user leaving their fortieth review.

What this actually means for your business

You cannot control how a happy customer chooses to write their review, and you should not try to. What you can control is never being the one introducing AI-generated text into the process yourself, and steering customers toward writing quickly and honestly rather than "getting it right."

• Never use a "review writer" tool to draft text for a customer to post. This is the fastest way to get flagged, and it is happening to the customer's own account, not just your listing.
• Ask for reviews in a way that invites a quick, honest line, not a polished essay. "What's one thing you'd tell a friend about us?" gets truthful, specific, human answers. "Please leave us a detailed 5-star review" quietly nudges people toward writing something that sounds performed.
• A steady, uneven trickle of reviews over time is safer and more valuable than a burst. It also happens to be what actually helps your ranking, since Google weighs recency and consistency, not just count.