Google's Moderation: Are certain reviews more susceptible to purgatory than others?

This month we continue to examine the new practices by Google on their review moderation. You can view our previous blogs on the subject in the previous Insights from Turner and Brewing Data blogs.

To briefly review: Google uses a hybrid approach of manpower and machine-learning algorithms to detect what they deem as prohibited & restricted content. And lately, they've been taking a heavier-handed approach in regulating reviews. They are with greater frequency withholding reviews for further evaluation, which consequently leads to historic reviews being publicly posted weeks or even months after being originally written.

J Turner Research has been monitoring this behavior since late July of this year and found that 28.6% of reviews being fed through our Google API are appearing with historical dates attached to them. 25% of all reviews are backdated 30 days or more. Google themselves also boast that they're faster than ever in stopping bad actors and better at preventing fake reviews ever being public.

It's become apparent that rather than specific topics or content (for our purposes pests vs safety, for example) — it is behavior patterns that get flagged. For instance, a sudden spike in reviews. 5-star reviews are seemingly targeted more.

This monitoring behavior actually coincides with how renters think too. A sudden spike in 5-star reviews is one of the most obvious signals to prospective renters that a community's profile is not to be trusted. From our latest Internet Adventure Study:

A separate study from Localo shows that from November 2023 through January 2026, 89.1% of deleted Google reviews were 5-stars.

5-star reviews are the most likely to be flagged as inauthentic. You might also expect that reviews without any text are lower effort and thus more likely to be withheld/deleted/moderated in some other way. However, 80.6% of reviews contained text. Speculatively, this could be because reviews without text don't provide content that might be flagged by Gemini. Common phrases of generic praise, e.g. "highly recommend and excellent service" are seemingly more suspect than a 5-star review without any accompanying commentary at all.

As methodologies to illicit inauthentic reviews become more sophisticated, Google's countermeasures and LLM training become more heavy-handed in turn. J Turner will continue to monitor trends and patterns on what types of reviews receive moderative action and how long they remain in purgatory.