Two very different people search for this, and they need different things.
The first is a shopper looking at a product with four and a half stars and eleven thousand reviews, wondering whether that rating is worth anything. The second is a business owner who just watched six one star reviews land on a Google profile in one weekend, none from anyone in the customer records.
This guide serves both. The first half is detection: the signals in language, timing, reviewer history and rating shape, how generative AI broke the old tells, and which checking tools still exist. The second half is response: reporting, evidence, and where a fake review becomes a legal problem.
Nobody knows the true rate. The only people who can count at scale are the platforms, and their numbers describe what they caught, not what got through. With that caveat:
Trustpilot’s Trust Report 2025 says it removed 4.5 million fake reviews in 2024, which it puts at 7.4 percent of all reviews submitted that year, and that 90 percent of those removals were automatic rather than manual. Tripadvisor’s 2025 Transparency Report says it caught 2.7 million fraudulent reviews in 2024, attributes 54 percent of total fraud to review boosting by owners, employees and affiliates, and says 214,000 removed reviews were AI generated. Google states on its official blog that it blocked or removed more than 240 million policy violating reviews from 2024, plus more than 12 million fake Business Profiles.
The most aggressive estimate comes from outside the platforms. The Transparency Company, a review fraud analytics firm, published research in December 2024 covering roughly 73 million reviews in Home Services, Legal and Medical categories across 100 US cities between February 2023 and February 2024. It concluded that 13.7 percent were actually or highly likely to be inauthentic, and 3.1 percent produced by AI or with AI assistance. That is one firm’s model applied to three sectors, not an industry measurement.
The honest summary: on well policed platforms, single digit to mid teens percentages of submissions get flagged or removed, and contamination looks worse in high value local service categories. BrightLocal’s Local Consumer Review Survey 2026, based on 1,002 US adults, found 97 percent of consumers think businesses should be punished for fake reviews.
The language tells
A real review is written by someone who had an experience. A fake one is written by someone producing something that looks like a report of one. That gap shows in the text. Read five or six reviews in a row rather than one, because most of these signals only appear across a batch.
Praise with nothing behind it
The most reliable single signal is enthusiasm that never touches a specific detail. “Great product, works perfectly, highly recommend” tells you nothing. A real buyer mentions the color they picked, the drawer that would not line up, the two weeks it took to arrive. Real experience is cluttered. Fake experience is clean.
The same test applies to negatives. “Terrible company, do not use, worst experience of my life” with no date, no service named and no description of what happened is the standard shape of a review attack.
The product name repeated like a keyword
People do not name the thing they bought over and over. They say “it” or “the chair.” A review reading “I bought the Acme Pro Ergonomic Mesh Office Chair and the Acme Pro Ergonomic Mesh Office Chair arrived quickly” was written from a brief instructing the writer to include the full product title for search visibility. For local businesses the equivalent is repeated use of the full business name plus the city.
Odd phrasing and translation artifacts
Review brokers operate globally, and much paid text is written by people not working in their first language, or is machine translated. Watch for missing articles, tense that slips, superlatives stacked oddly, idiom used slightly wrong. One awkward sentence proves nothing. A cluster sharing the same awkwardness is a signal.
A reviewer voice that does not match the product
This is the tell most people miss and often the strongest. Ask whether the writer sounds like the person who would buy this. A review of professional welding equipment that never uses a term a welder would use. A five star review of a pediatric dental practice that praises “the staff” but never mentions a child. A restaurant review that never names a dish. Related: reviews that read like ad copy. When a review lists features in the order the product page lists them, it was written from the product page.
The timing tells
Real reviews arrive the way customers do, unevenly but continuously. Manufactured reviews arrive in campaigns, because someone bought a campaign. Sort by newest and by oldest and look at the dates.
Clusters. Forty reviews in three days for a business that otherwise collects two a month. The cluster is the anomaly whether the reviews are positive or negative.
A launch burst with no runway. A listing that goes from zero to several hundred reviews within days of appearing, then flattens. Real demand ramps.
Negative bursts after a news cycle. After a recall, a viral complaint or a political controversy, people who were never customers pile onto the review pages. This is review bombing. The reviewers may be real, but the reviews describe no transaction, and Google, Yelp and Tripadvisor all treat off topic pile ons as policy violations. Yelp goes further, posting temporary public alerts on pages attracting sudden media driven attention.
One more pattern: a wave of five star reviews arriving right after a wave of one star reviews. That is usually a business burying genuine criticism with bought volume, and it is what detection systems catch best, because burying requires speed and speed produces clusters.
Reviewer history and profile tells
On every major platform you can click a reviewer’s name and see everything else they have written. This takes ten seconds and is more informative than the review itself.
Single review accounts. An account whose entire history is one five star review, or one one star review, is the commonest shape of a bought or malicious review. Not proof alone, since real people do review once. But a business whose positive reviews come overwhelmingly from single review accounts has a problem.
Only fives, or only ones. Look at the spread of ratings the reviewer has given. Real reviewers scatter. An account with sixty reviews that are all five stars is either paid or carries no information. An account that has only ever left one star reviews is usually a grudge account or part of an attack network.
Geographically impossible patterns. The cleanest structural tell in local reviews. An account that reviewed a dentist in Phoenix, a plumber in Newark, a locksmith in Tampa and a mover in Portland within one week is not a traveler. It is a for hire account working a queue. Google Maps profiles list the places reviewed, which makes this quick. The category version is the same tell: nine garage door companies in different states.
Profile thinness. Stock photo, first name and last initial, no photos on any review, account created recently. Any one is normal. All of them across several reviewers who hit the same business in one week is not.
Missing purchase verification. On Amazon, check for the Verified Purchase badge. A product whose positive reviews are overwhelmingly unverified while its negative ones are verified is telling you something.
What the rating distribution shape tells you
Most platforms show a histogram of how many reviews fell at each star level. Its shape is a fingerprint. A genuinely popular product usually produces a J: a large block of fives, a meaningful block of fours, a thinner tail of threes and twos, and a small but real block of ones. Even excellent businesses collect ones.
A flat top with almost no fours. Bought reviews are nearly always five stars, because that is what was paid for. Two thousand fives, forty fours and three hundred ones means two populations on one listing.
The U shape. A deep pile of fives, a deep pile of ones and almost nothing between usually means a genuine base plus either an attack or a bought layer. Read the ones to work out which.
Rating and text mismatch. Read the three and four star reviews first. Nobody buys a three star review, so that band is the most honest on the page. If it describes serious problems the five star band never mentions, trust the middle band.
How AI generated reviews changed the tells
Several classic tells were really tells about cheap labor. Broken grammar, translation artifacts and repeated typos existed because fakes were written quickly by underpaid people working from a template. Generative AI removed that constraint. A model writes fluent, plausibly detailed English at zero marginal cost, in any volume, with variation built in. Fluency is no longer evidence of authenticity, so watch for these instead:
Smoothness without grit. A 2025 study in the Journal of Retailing and Consumer Services by Zhao, Tang, Zhang and Lyu analyzed 714,016 reviews and reported that AI generated fakes showed higher comprehensibility and lower specificity and exaggeration than both human fakes and authentic reviews, with significantly higher mechanicalness and lower empathy than authentic reviews. In plain terms, AI reviews read better and say less.
A concession that costs nothing. Models are trained to sound fair, so they insert a token criticism: “the only minor downside is that it took slightly longer to arrive than expected.” The complaint is always safe. Real complaints have consequences attached.
Uniform length and tone. Human reviews run from four words to eight paragraphs. A batch where everything lands between 60 and 90 words in the same register is machine shaped: individually convincing, collectively identical.
Platforms adapted by leaning on signals text cannot fake: account age, device and network fingerprints, location consistency, the graph of which accounts review which businesses, and purchase records. Text is now the weakest evidence a detection system has, which is why manual reading still matters for you. You are assessing one listing, not a hundred million, and can weigh context a classifier cannot.
Regulators treat AI generated fakes as squarely covered. In December 2024 the FTC approved a final order against Rytr, which sold an AI testimonial and review writing service, on the basis that it gave subscribers the means to generate false and deceptive reviews.
Third party tools for checking reviews
This category has thinned badly, and most articles recommending tools are out of date. The two best known checkers are gone.
Fakespot was the most widely used review analyzer and was acquired by Mozilla, which built it into Firefox as Review Checker. Per Mozilla’s own May 2025 announcement, Review Checker shut down on June 10, 2025, and the Fakespot extensions, mobile apps and website stopped working on July 1, 2025. Any guide still telling you to install Fakespot has not been updated since mid 2025. ReviewMeta, the other long running Amazon analyzer, is no longer reachable as a working service.
What filled the gap is a churn of small extensions and web checkers, several ad supported, none with the track record Fakespot accumulated. We are not naming specific ones, because the names change faster than any published list can track. If you use one, check who operates it, what it collects, and whether it earns affiliate revenue on the products it grades.
Checking methods that do not depend on a vendor staying in business:
Sort by most recent and by lowest rating, and read both.
Open three or four reviewer profiles and check history and geography.
Paste a distinctive sentence into a search engine. Broker written text gets reused across listings.
On Yelp, open the “not currently recommended” list at the bottom of the page. Yelp shows you what its own software filtered out.
What the platforms actually do about it
Every major platform runs automated detection before publication, plus human review teams, plus enforcement against the accounts and businesses involved. The mechanics differ.
Google bans fake engagement under its Maps user generated content policy, defined to include paid or incentivized reviews, multiple accounts posting identical content at one person’s request, and emulators or device manipulation used to mimic genuine engagement. It also bans undisclosed conflicts of interest, covering current or former employment, contractual or consulting relationships and other affiliations. Enforcement runs from content removal up to posting restrictions on accounts and on the Business Profile. Google says its detection uses machine learning models trained with Gemini that learn suspicious patterns from a few examples and generalize across categories and languages.
Yelp is the most visible. Its recommendation software decides which reviews display by default and which drop to the “not currently recommended” list, applying, in Yelp’s words, the same objective rules to every business. The published criteria are conflicts of interest, solicited reviews, reviewer reliability, and usefulness. Yelp also posts public consumer alerts on business pages when it detects compensated or suspicious review activity.
Amazon screens before publication with systems it says analyze thousands of data points across billions of reviews going back to 1995, plus multimodal models weighing text, images, seller behavior and supply chain patterns together. It also sues review brokers rather than only deleting content. Trustpilot and Tripadvisor both publish annual transparency reports with removal counts and fraud breakdowns, and both lean heavily on automated detection.
Monitoring your own profile across these platforms is a separate problem from detection, and it is covered in our roundup of the best review management software.
The FTC rule on fake reviews and testimonials
In the United States there is now a specific federal rule, not just general deception law: the Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, codified at 16 CFR Part 465. The FTC announced the final rule in August 2024 and it took effect on October 21, 2024. Per the FTC’s own description, it prohibits six categories of conduct:
Fake or false reviews and testimonials. Creating, buying, selling or disseminating reviews that misrepresent who wrote them or describe an experience the reviewer never had. The FTC states this covers AI generated reviews.
Buying positive or negative sentiment. Providing compensation or other incentives conditioned on the review expressing a particular sentiment.
Undisclosed insider reviews. Reviews by officers, managers, employees or agents without clearly disclosing the relationship.
Company controlled review websites. Misrepresenting that a site the business controls provides independent reviews about that business.
Review suppression. Using unfounded or groundless legal threats, physical threats, intimidation or false public accusations to force removal of a negative review, and misrepresenting that the displayed reviews represent all or most of those submitted when negatives have been suppressed.
Fake social media indicators. Selling or buying fake followers, views or other influence indicators where the buyer knew or should have known they were fake.
Two points matter most in practice. First, the rule carries civil penalties. The FTC’s stated reason for issuing a rule rather than relying only on case by case Section 5 enforcement is that a rule lets courts impose civil penalties on knowing violators. In its December 22, 2025 announcement of warning letters to ten companies over possible Consumer Review Rule violations, the FTC referred to civil penalties of up to $53,088 per violation. That amount is adjusted for inflation annually, so check the current published figure before relying on it.
Second, the rule targets businesses, not reviewers. The FTC’s own questions and answers guidance states plainly that ordinary consumers cannot be liable under the rule for what they say or do not say in reviews. If you are an owner tempted to send a threatening letter to make a review disappear, the suppression provision is aimed squarely at you.
The FTC has enforced here beyond the rule itself: Fashion Nova agreed in January 2022 to pay $4.2 million over allegations it blocked negative product reviews from its own site. Outside the US, the United Kingdom’s Digital Markets, Competition and Consumers Act 2024 banned posting and commissioning fake reviews from April 7, 2025, with the Competition and Markets Authority able to fine up to 10 percent of global turnover.
For businesses: hit with fake negative reviews
A competitor, a fired employee, an extortion operation or an unrelated pile on has dropped reviews on your profile and you need them gone.
Confirm it is actually fake, not just bad
Platforms reject most removal requests, and the commonest reason is that the business is reporting a real customer it did not enjoy hearing from. Check your records first: the reviewer’s name, details in the text, the date, the service described, and ask the staff who were on shift.
You need one of the removable categories: the reviewer was never a customer, the review describes a different business, it came from a competitor or former employee, it is part of a coordinated burst, it contains no experience at all, or it breaches a content policy such as personal attacks or disclosure of personal information. “This review is unfair” is not a removal ground anywhere.
Gather evidence before you touch anything
Reviews get edited and profiles get scrubbed once a campaign ends. Capture first.
Full page screenshots of each review showing rating, date, reviewer name and text, with the URL and system clock visible.
Screenshots of each reviewer’s profile and full review history, which is where the geographic and category impossibilities live.
A timeline of the date and time each review appeared. A cluster persuades a moderator far more than any single review.
Proof of the negative: a note that you searched your CRM, booking system or point of sale for the name and the described transaction and found nothing.
Anything linking the reviewers to each other or to a competitor, and the trigger if there was one: a termination date, a dispute, a news story.
Archived copies through a web archiving service, so you hold timestamped third party evidence.
Report on the platform
Google. Report from your Business Profile or the Reviews Management Tool, choosing the specific violation rather than a generic one. Evaluation typically takes several days. If the report is rejected, Google allows a one time appeal covering up to ten eligible reviews, so use that slot deliberately and bundle your strongest evidence. Full walkthrough: how to delete a Google review.
Yelp. Report through the business account and cite the specific guideline breached, usually conflict of interest or lack of personal experience. Yelp’s software may demote rather than remove. See how to delete a Yelp review.
Trustpilot. Flag the review and supply the order or reference number you cannot match, since the process centers on whether a genuine transaction occurred. See how to remove Trustpilot reviews.
Amazon, Facebook, Tripadvisor and the industry sites. Each has a report function near the review, plus a seller or business support channel that usually outperforms the public button. Report every review in a burst at once and say that they are coordinated.
Respond in public, carefully
While the report is pending the review sits there being read. A short, calm public reply is written for the next reader, not the reviewer. Say you take feedback seriously, that you have searched your records and cannot locate a customer matching this account or transaction, and that you welcome direct contact to resolve anything genuine. Give a contact route and stop.
Do not accuse the reviewer of being a competitor in public, do not speculate about identity, do not disclose customer details while defending yourself, and do not threaten. Threatening a reviewer with legal action to force removal is the conduct the FTC rule’s suppression provision prohibits, and a screenshot of you doing it will travel further than the review did.
The structural defense is volume. A business with 40 reviews is badly damaged by six fakes. A business with 400 and a steady inflow absorbs them, which is why building that inflow through compliant, non incentivized requests is the best thing you can do before an attack happens.
When it crosses into defamation
Most fake reviews are a policy problem. Some are a legal one. The general line in US law is the difference between opinion and a false statement of fact. “The food was awful and the staff were rude” is opinion, protected however unfair it feels. “This restaurant gave my family food poisoning and was shut down by the health department” is a factual assertion. If it is false, was published to others and caused harm, it may be actionable. Fabricated accusations of crime, fraud, license violations or health violations are the usual candidates.
Two practical realities. The platform is generally not the defendant, because Section 230 of the Communications Decency Act protects interactive computer services from liability for content posted by users. The author is the defendant, which means anonymous reviewers must first be identified, usually through subpoena after a suit is filed. That takes time and money.
Before going near litigation, read our explainer on what defamation of character actually requires, because the gap between “this is defamatory” as people use the phrase and as courts use it is enormous. And note the asymmetry the FTC rule created: an unfounded legal threat sent to suppress a review is itself prohibited conduct, so any demand letter needs to come from a lawyer who has assessed the claim, not from a template.
Where a campaign is large and organized, both routes are slow, and a structured program running removal, suppression and review generation together is usually the realistic path. Our reputation management hub covers how those pieces fit together.
A two minute checklist
Check
What you are looking for
Weight
Read five reviews in a row
Shared sentence skeletons, uniform length, identical tone
High
Open three reviewer profiles
Single review accounts, all fives or all ones, impossible geography
Very high
Look at the date pattern
Clusters, launch bursts, spikes after a news event
Very high
Look at the star histogram
Missing four star band, U shape, implausibly clean average
Medium
Read the three and four star reviews
Problems the five star band never mentions
High
Cross check another platform
Large rating gaps between sites
Medium
Conclusion
For shoppers, the shift worth internalizing is that you can no longer judge a review by how well it is written. Generative AI made fluent, detailed fake text free. What it did not make free is a credible account history, a coherent geography, a purchase record, or a review pattern that looks like a human life. Stop reading individual reviews closely and read the population instead: the dates, the profiles, the distribution shape, the three star band.
For businesses the sequence rewards discipline. Confirm the reviews are fake against your own records. Capture the evidence before it disappears. Report through the correct platform category with the pattern, not an argument about fairness. Reply once in public, calmly, for the next reader. Escalate to legal advice only where a false statement of fact caused real harm, and never send a threat to make a review go away, because that is now specifically prohibited conduct rather than a gray area.
The durable protection is boring and it works: a steady, compliant flow of genuine reviews, so that when a handful of fakes arrive they land on a profile deep enough to absorb them.
