How the score works
Every report on this site is produced by the same public method, with no manual adjustment and no input from the business being analyzed. This page documents it in full, including what it gets wrong.
In short
Revdict reads a place's publicly visible Google Maps reviews and tests each one against five manipulation signals — review bursts, review drives, low-activity reviewer accounts, near-duplicate wording, and generic accounts seen across our catalog. Each flagged review gets a 0–100 suspicion score; those are combined into a single 0–100 manipulation score for the place, weighted by what share of its reviews were flagged. Below 40 is Clean, 40 and above is Suspicious, 70 and above is Likely Manipulated — unless the flags are mostly review drives, in which case the verdict is Review Drive. The score is a statistical estimate of resemblance to manipulation patterns. It is not proof, and it is not an accusation against any business.
What each report examines
One crawl of the place's public Google Maps reviews, most recent first, plus the public profile of reviewers whose accounts look sparse enough to be worth checking. Nothing private, nothing purchased from a data broker, nothing supplied by the business.
The crawl is capped at about 1,000 reviews per place. That limit belongs to the data source, not to the scorer. For a place with more reviews than that, the report describes its most recent 1,000 reviews and says so — the "Reviews examined" figure on every report is the real denominator behind every percentage on the page. Of the 493 places in the catalog today, 430 have more reviews on Google than a single crawl returns.
The practical consequence: manipulation that stopped before the window opens is invisible to us. A place that bought reviews three years ago and never did it again will score Clean, correctly describing its recent record and saying nothing about its older one.
The five signals
Each signal runs independently over every review. A review can carry more than one. The weight is how much that signal counts once scores are combined.
Review burst
burst · weight 1- Fires when
- — A single day whose five-star count is at least 5 reviews AND at least 5× the place's average daily review volume over the preceding 30 days. The first 14 days of the fetched record have no baseline and are never bursts.
- Scores
- — Every five-star review posted on that day, at 80/100.
- Rationale
- — Organic reviews arrive at a rate set by footfall. Bought ones arrive in batches, because they are delivered as an order.
Review drive
review_drive · weight 0.5- Fires when
- — A burst day where at least 50% of the five-star reviews name a staff member ("Maria was amazing", "ask for Sam").
- Scores
- — Every five-star review posted on that day, at 40/100 — a lower score, and a separate verdict when it dominates.
- Rationale
- — Staff asking real customers to leave a review produces the same spike as a purchase, but the reviews are personal, varied and name different people. That is a review drive, not a review farm.
Low-activity account
low_activity_account · weight 0.8- Fires when
- — A one- or five-star review whose author has 2 or fewer reviews in total.
- Scores
- — 30/100, plus 15 if the account is not a Local Guide and 15 if it has never posted a photo.
- Rationale
- — Review farms burn through disposable accounts. A genuine reviewer usually leaves a trail — other reviews, photos, Local Guide status.
Duplicate text
duplicate_text · weight 0.9- Fires when
- — A review of at least 60 characters whose wording is at least 85% similar (trigram similarity) to another review of the same place.
- Scores
- — The similarity percentage itself, so a 92% match scores 92/100.
- Rationale
- — Bulk-written reviews are paraphrases of one brief. Two customers rarely describe the same meal in near-identical words.
Generic account
generic_account · weight 1- Fires when
- — An author who has left five-star reviews at 3 or more different places in our catalog, every one of them generic prose averaging 80+ characters, with no photo and no number (no price, date, dish count) anywhere.
- Scores
- — Every review by that account on this place, at 85/100.
- Rationale
- — Paraphrased review farms use seasoned accounts and reword one brief per venue, so no single review repeats and no single day spikes. What gives them away is the footprint across businesses: long praise that never mentions anything specific.
From signals to one number
- Each flagged review takes its highest weighted signal score, plus 15 points for every additional signal on the same review, capped at 100. Signals reinforce each other, but a review is not condemned three times over for one behaviour.
- The place's score is the average suspicion of its flagged reviews, scaled by how much of the review record they represent: at 40% of reviews flagged the average applies in full, and below that it is reduced proportionally.
- That is computed twice — over the whole record and over the last 180 days — and the higher result wins, so a campaign this season is not hidden by years of organic history.
That second step is the load-bearing one. Four suspicious reviews out of 2,000 is noise and scores near zero; four hundred out of 2,000 is a pattern. It is also why a place can have flagged reviews and still be graded Clean — the report shows both numbers so you can judge for yourself.
Verdicts
| Verdict | Score | Means |
|---|---|---|
| Clean | below 40 | Nothing in the public record stands out. |
| Review Drive | 40 and above | The flagged reviews are mostly from review drives — staff asking customers for named reviews. Real visits, solicited feedback, not a review farm. |
| Suspicious | 40 to 69 | Enough of the review record matches manipulation patterns to be worth a closer look. |
| Likely Manipulated | 70 and above | A large share of reviews carry strong signals. Still an estimate, not a finding of fact. |
What the score cannot tell you
- Who posted a review, or who paid for it. The signals describe patterns in a review record. They cannot attribute intent, and a business does not control who reviews it — a negative-review attack by a competitor raises the same flags as reviews a business bought itself.
- Whether a specific flagged review is fake. Every signal has innocent explanations. A local news feature produces a genuine one-day burst. A first-time reviewer with no photos is usually just a first-time reviewer. Short reviews resemble each other because there are only so many ways to write "great food, fast service".
- How accurate this is. We have not measured a false-positive rate, because doing so honestly needs reviews independently labelled as genuine or fake, and no such public dataset exists for Google Maps. Anyone quoting these numbers should quote that limitation with them.
- What Google already removed. Reports see only what was publicly visible at crawl time. Google blocks or removes most policy-violating reviews before anyone reads them, so a clean record may mean a working filter rather than an honest one.
- Anything about quality. A manipulation score is not a review of the business. Places with clean review records can be bad, and places with messy ones can be excellent.
Frequently asked questions
Does a high score prove a business bought fake reviews?
No. The score measures how closely a public review record resembles known manipulation patterns. It cannot establish who posted a review or why, and a business can score high because of a competitor attack, a viral moment, or a promotion it never asked for.
What is the false-positive rate?
We have not measured one. That would require a labelled set of reviews known to be genuine or fake, and no such public dataset exists for Google Maps. Treat every verdict as a statistical estimate with unquantified error, not a measurement.
How many reviews does each report examine?
Every review the scraper returns, most recent first. That is currently capped at about 1,000 per place by the data source, so places with more than that are scored on their most recent 1,000 reviews, not their full history.
Can Revdict see reviews Google has already removed?
No. Reports are built only from reviews publicly visible at the time of the crawl. A place that was cleaned up before we looked will read as clean, and Google removes the majority of policy-violating reviews before anyone sees them.
How often is a report updated?
A report is a snapshot of the day it was generated, shown on the page. Anyone can request a fresh crawl from the report page, and the score is recalculated from scratch each time.
Is Revdict affiliated with Google?
No. Revdict is independent, reads only publicly visible review data, and has no access to Google’s internal signals or enforcement decisions.
Citing this
The method and the catalog are free to cite. Please link the specific report rather than describing a business's score in isolation, and carry the limitations above with the number.
Revdict, "How the score works", August 2026. 493 places, 455,691 reviews examined. https://revdict.com/methodology
Run a place through the analyzer · Browse the catalog · Report an error in a report