How Our Fake-Review Score Works (and What It Can't Tell You)

Our fake-review score is a 0-100 number computed from a business's full public Google review history, combining four weighted signal categories: review-burst clustering, the share of reviews from low-activity accounts, duplicate or near-duplicate text, and mismatches between star rating and review content. Businesses land in one of three verdicts — Clean (under 30), Suspicious (30-59), or Likely Manipulated (60+) — based on the combined score, not any single signal.

The goal isn't to declare a verdict on any one review. It's to give you, in seconds, the same pattern-matching a careful reader would do manually across hundreds or thousands of reviews — something no one has time to do by hand. If you'd rather check a listing yourself first, see how to spot fake Google reviews for the manual version of the same checks.

#What signals make up the score?

Four categories feed the score, and they're weighted rather than simply added, because some patterns are stronger evidence than others on their own.

Review bursts. We look for clusters of reviews, usually 5-star, posted in a short window relative to a business's normal review pace. A burst that breaks sharply from the historical rate contributes more to the score than a gradual rating drift. Across our dataset, burst signals appear in 12,259 flagged reviews — the second most common pattern we track.

Low-activity accounts. We check each reviewer's visible history: how many reviews they've posted in total, and how varied those businesses are. Accounts with only one or two lifetime reviews, especially when several such accounts cluster on the same business in the same period, weigh heavily. This is the single most common signal in our data — 37,061 of the reviews we've flagged carry this marker, more than bursts and duplicate text combined.

Duplicate text. We compare review text across a business's history (and, at a lower weight, across other businesses we've analyzed) for exact or near-exact phrase matches. Reviews-for-hire operations frequently reuse short template phrases, which is a fingerprint organic reviews almost never share. 5,958 flagged reviews in our dataset carry this signal.

Rating-text mismatch. A 5-star rating paired with generic, non-specific text, or a 1-star rating that reads like a form complaint with no concrete detail, contributes a smaller weight on its own but strengthens the score meaningfully when it co-occurs with the other three signals.

#Why weight the signals instead of just counting them?

A single low-activity reviewer or one vague 5-star review happens constantly in real review histories and means nothing alone. What separates a Clean business from a Suspicious one is co-occurrence: multiple signals clustering on the same reviews, in the same time window, at a rate that breaks from the business's own baseline. Weighting lets a strong, rare signal like duplicate text move the score more than a weak, common one like a single vague review, and it's why the score is computed from the full review history rather than a handful of recent reviews.

#What does the score distribution actually look like?

As of July 2026, we've completed analysis on 242 places, covering 241,041 individual reviews. The average manipulation score across all of them is 32.1/100. Verdicts split as follows:

Most analyzed places score low. The score histogram shows a heavy concentration in the 10-19 range (120 of 242 places), with a second, smaller cluster in the 60-79 range — consistent with most businesses having ordinary review patterns and a distinct minority showing the clustered signals manipulation campaigns leave behind.

#What can't the score tell you?

The score is built entirely from public data: review text, star ratings, timestamps, and reviewer profile metadata visible on the listing. It cannot see what a platform's internal systems can — IP addresses, device fingerprints, payment records connecting a business to a review-selling vendor, or account-creation patterns behind the scenes. That means:

• It cannot identify who wrote a specific review, or prove any single review is fake. It flags statistical patterns across the review set.
• It cannot prove a business purchased reviews. A high score reflects patterns consistent with manipulation; it does not establish intent or a transaction. We report it that way for exactly this reason — see our note on writing about specific places below.
• It can be wrong in both directions. A business with an unusually enthusiastic, low-review-history customer base (a new location with a loyal following, for instance) could score higher than it deserves; a sophisticated campaign using aged, high-activity accounts and varied text could score lower than it deserves.
• It's a snapshot, not a verdict that ages well on its own. Review histories keep growing, so we recommend rechecking a listing periodically rather than treating one score as permanent.

Academic fake-review detection research reaches a similar conclusion: behavioral and contextual signals — burst timing, reviewer metadata, and text similarity — meaningfully improve detection accuracy over text analysis alone, but no published method claims certainty on an individual review; all of them describe probability, not proof.

The rule prohibits fake reviews and testimonials, including AI-generated reviews, suppression of negative reviews, and paying for positive or negative reviews.

That's the FTC's own framing of what its 2024 Consumer Review Rule bans — and it's a useful reminder that even regulators build enforcement around patterns of conduct, not a single review in isolation.

#How we write about individual places

Because the score can't prove intent, we never state that a named business bought or wrote fake reviews as fact. When we reference a specific place, we describe what the analysis flagged — a burst, a concentration of low-activity accounts — and link the full public report so readers can see the same evidence and draw their own conclusion.

This score is built entirely from Google Maps' public review data. For how Google's own moderation compares to Yelp's and Tripadvisor's approach to fake reviews, see Google vs. Yelp vs. Tripadvisor: who fights fake reviews best?

#Frequently asked questions

#Is a manipulation score the same as an accusation?

No. The score describes statistical patterns in a business's public review history that are consistent with manipulation. It is not a legal finding and doesn't establish that any specific review is fake or that the business paid for reviews.

#Can a business's score change over time?

Yes — new reviews keep arriving, so we recommend rerunning an analysis periodically rather than treating a score as permanent, especially for businesses near the Clean/Suspicious boundary.

#Why is "low-activity account" the most common signal instead of duplicate text?

Because it's the easiest signal to leave behind. Reviews-for-hire services can vary the wording of every review with minimal effort, but the accounts they use are, almost by definition, created or reused for the purpose and have little other review history — that pattern is much harder to disguise at scale.

#Does a low score guarantee a business has never had a fake review?

No. A low score means the review history doesn't show the clustered patterns our signals detect at a level that crosses the Suspicious threshold — it doesn't mean zero fake reviews exist, only that there's no statistical pattern strong enough to flag.

#Can I see which specific reviews were flagged?

Each public report page shows the evidence bullets behind the score — the signal types found and their relative weight — without exposing raw reviewer identities.

You can run this same scoring process on any Google Maps listing with our free fake-review analyzer — paste a Maps link and get the signal breakdown in seconds.