Google uses AI to fast-track fake review takedowns

If you run a local coffee shop or repair garage, few things are more frustrating than waking up to find a mysterious flood of 1-star reviews tanking your Google rating overnight.

To fight back against review bombing and paid astroturfing, Google rolled out a new machine learning model designed to catch fraudulent activity faster. The company announced it blocked or removed over 170 million policy-violating reviews across Search and Maps in 2023, representing a 45% increase compared to 2022.

Spotting coordinated review rings

Google handles around 20 million updates to local business profiles every single day, covering everything from holiday hours to customer photos and reviews. At that scale, manual moderation is out of the question.

Instead of just scanning individual reviews for obvious spam words, the new algorithm analyzes behavioral patterns across time. It flags sudden spikes in extreme sentiment, like an unusual burst of 1-star or 5-star ratings out of nowhere, and catches duplicate reviews copy-pasted across different business pages.

Google pointed out that the system was specifically able to uncover organized scam networks that paid users to generate fake positive ratings, shutting down campaigns before they spread further. Beyond text reviews, Google also caught 14 million more policy-violating videos than the year before and blocked more than 2 million fraudulent attempts to claim ownership of local business profiles.

Why it matters

For local business owners, faster fake review detection offers some genuine relief. Online reputation directly dictates foot traffic, and small shops rarely have the leverage to get wrongful negative reviews cleared quickly through support channels. Now, having a 45% more effective filter means fewer coordinated smear jobs slip through to damage a business's livelihood.

Looking at the broader industry, Google’s move is a necessary response to the reality of generative AI. Fabricating authentic-sounding reviews used to take human labor; now anyone can churn out thousands of unique, persuasive paragraphs for pennies. Google was smart to focus on temporal patterns, account networks, and distribution bursts rather than just textual analysis, as modern synthetic text can easily deceive basic sentiment tools.

However, we think this is only a temporary breather in an escalating arms race. If spam rings adapt by staggering their fake posts over weeks to avoid tripping velocity triggers, detection will get significantly harder. Google blocked 170 million bad reviews, which sounds impressive, but it also highlights the alarming scale of deception that local search still has to deal with every day.