We read the reviews, run the stats, and tell you whether the rating is earned, with the evidence to back it up.
We don't scrape Google. Your AI agent already has a browser; we have the scoring. It collects a listing's reviews, we grade them, and the place appears here for everyone.
check a place with your ai agent
1.
Connect the legitreviews MCP. In Claude Code:
claude mcp add --transport http legitreviews https://legitreviews.ai/mcp
In claude.ai: Settings → Connectors → Add custom connector → https://legitreviews.ai/mcp. Any other MCP client (Muse, Cursor, your own agent) takes the same URL, Streamable HTTP, no key needed to try it.
2.
Ask it about a place:
“Are the Google reviews for the Blue Lantern Cafe on Main Street legit?”
At least 29 reviews were posted by low-history accounts across 2 days with statistically unlikely review volume. The worst was on May 4, 2024 which received 19 reviews (about 1 in 1014 by chance based on normal volume). This looks like a coordinated 5 star campaign by the business.
11 reviewers independently report the business rewards customers for reviews: "Writing the review gives you a free coconut ice cream, which is worth getting!" · "They encourage you to write a review for a free ice-cream maybe that's why."
30% of reviews from single-review accounts, skewing the average up.
base score
every place starts at 100
100
rating inflation
headline 4.73★ is 0.25 above the 4.48★ that credible reviewers (detailed Local Guides) give
-15
coordinated review injection
29 reviews (1.94% of all) from low-history, non-Local-Guide accounts, across 2 statistically-impossible days. The worst, 2024-05-04, drew 19 reviews (about 1 in 1014), 16 of them injected; 16 of them predate the 24-month window and weigh a third
-3
review solicitation
11 reviewers independently report the business rewards customers for reviews, e.g. "Writing the review gives you a free coconut ice cream, which is worth getting!", "They encourage you to write a review for a free ice-cream maybe that's why."
-18
throwaway single-review accounts
30% from 1-review profiles, pushing the average up
Reviewers describing the business asking for, steering, or rewarding reviews, classified from the review text by an LLM. Only firsthand reports of policy-violating behavior (rewards, positive-only asks, gating, pressure) are counted. The penalty slides with the number of independent witnesses, so a single account only nudges the score. Tap a reviewer to read their full review.
★★★★☆Youjin Lee · local guide · 27 reviews · Edited 5 months ago
↳ “Writing the review gives you a free coconut ice cream, which is worth getting!” reward offered · counted
★★☆☆☆Awen Chen · local guide · 13 reviews · a year ago
↳ “They encourage you to write a review for a free ice-cream maybe that's why.” reward offered · counted
↳ “all the high reviews here are due to free ice cream offered for review” reward offered · not counted: hearsay, only firsthand reports count
evidence: burst days (2)
Days with statistically-improbable review volume for a place averaging 1.67/day. Flagged = reviews posted that day by low-history, non-Local-Guide accounts: a statistical flag, not proof they’re fake (a real group visit or promo can look identical). Days that kept the place’s normal share of Local Guides read as genuine surges.
day
flagged
odds
2025-09-06
13 of 15
1 in 109
2024-05-04
16 of 19
1 in 1014
historical
organic: kept the place’s normal share of Local Guides, reads as a genuine surge. historical: older than 24 months. The odds of each day are in the downloadable report.
how it works
1
Step 1: your agent collects the reviews
It opens the listing in its own browser, sorts by newest and collects them: all of them, or the 500 most recent at a big listing. We don't scrape Google.
A to F, plus the rating credible reviewers actually give, with the evidence one click away. The place then appears here for everyone.
Google usually asks the browser to be signed in before it will sort or load more reviews. The agent reads Maps through your signed-in browser, much like scrolling the listing yourself. Heavy use could trip Google's rate limits; if you plan to check many places, a secondary Google account is a sensible choice.
No agent? Scripts can POST the same review JSON to /api/submissions; the format is at /api/sources/google-maps. Everything required is visible on the review card; review ids and exact timestamps are optional. Results from a single unconfirmed submitter are labeled as such until a second source agrees.
methodology
what signals do we look for?
We read a place's full review history and weigh the signals that are costly for a review farm to fake:
Burst campaigns: clusters of 5 star reviews dumped on a single day by low-history accounts. A genuine surge keeps a place's normal share of Local Guides, so it isn't counted against it.
Drip-feed farms: reviews arriving nearly every day at a pace no organic business sustains, a paid stream spread thin to dodge obvious spikes.
Cross-posted text: the same review copied word-for-word onto other businesses, a farm seeding a whole chain.
Throwaway accounts: single-review profiles padding the average up.
Review solicitation: reviewers themselves reporting that the business rewards, pressures, or steers reviews (“free dessert if you leave 5 stars”). An LLM reads every such mention; only firsthand reports of policy-violating practices count, and the penalty deepens with each independent witness. Politely asking for a review is allowed by Google and never counted.
Too-perfect records: a spotless, unanimous 5 star history often signals curated or incentivized reviews, so it can't earn a top grade, even when we can't prove fraud.
what's the “adjusted rating”?
What hard-to-fake reviewers actually give, vs the headline number. A gap means the rating is inflated. Two groups qualify:
Credible reviewers are Local Guides who also wrote a substantive review. The badge alone is gameable (farms run Local Guide accounts that drop bare 5 star ratings), so we only count ones that put real words behind the stars.
Established accounts have 30+ lifetime reviews, regardless of badge or text. Farms pad accounts with a handful of reviews to look plausible, but aging thousands of accounts past 30 is the expensive part.
Each place is anchored to whichever group gives the more conservative rating. When a place corrects the headline number, the “per credible reviewers” / “per established accounts” label tells you which.
where does the review data come from?
From the agents and scripts that submit it. Every submission is checked for structure and consistency (Google's own review and profile ids, timestamps against displayed dates, the rating average against the displayed rating, the count against the listed count, overlap with what we already hold), and a place submitted by one unconfirmed source carries an unverified source note until a second, independent submission agrees. Places we pulled ourselves before October 2026 need no such note.
what does a high score mean?
It means we found no evidence the business is manipulating or inflating its own reviews. It does not promise the rating is accurate: a clean place can still be mis-rated by too few reviews, reviewers who self-select, or honest reviews that just don't match your experience. We measure whether the number looks gamed, not whether it's right.