Choice Science · established evidence
The Reviewers Who Never Bought: What Anderson and Simester Found About Deceptive Reviews From Loyal Customers
Deceptive reviews are usually imagined as competitor sabotage: a rival posts fake one-star ratings to drag a business down. A 2014 study by Eric Anderson and Duncan Simester found something stranger and more uncomfortable. Looking at a large private-label retailer's own site, they identified reviews written by accounts with no record of ever buying the product being reviewed, roughly five percent of all the reviews on it. Those unverified reviews were systematically more negative, described the actual product experience less, and carried more of the linguistic markers researchers associate with deception. The part that breaks the folklore is who wrote them. Many were not saboteurs at all, but the retailer's own best and most loyal customers. Deceptive reviewing, it turns out, is not only an attack from outside. It can come from the people who like a business most, which is exactly why a raw star average cannot be read as a clean record of real experience.
The folklore says sabotage. The evidence says something odder.
Ask most owners why a fake review appears under their name and they will point outward, at a competitor. The intuition is reasonable and partly correct: review manipulation is a real, strategic behavior, and it does rise where a business is under competitive pressure. But it is not the whole picture, and treating every suspicious review as an outside attack misses where a large share of the problem actually comes from.
In 2014 Eric Anderson and Duncan Simester published a study, in the Journal of Marketing Research, that looked directly at the reviews on a large private-label retailer's own website and matched each review against the retailer's purchase records. That matching is what makes the paper unusual. It could separate reviews written by someone who had actually bought the product from reviews written by someone who had not, which is a distinction a star average never shows you and most platforms cannot enforce.
What Anderson and Simester actually measured
The headline number is that about five percent of the reviews on the retailer's site came from accounts with no purchase record for the product being reviewed. On its own, five percent might sound like noise. The importance is in how those reviews behaved compared with the verified ones written by confirmed buyers.
Because the study joined review text to transaction data at the level of the individual account, it was not inferring fraud from a platform's filter or from a suspicious pattern. It was reading the reviews of people the retailer could confirm had, or had not, bought the item. That is a stronger footing than most review-fraud work, which has to use proxies like a site's own recommended-review flags to guess which reviews are fake.
Reviews without a purchase looked different in three measurable ways
The unverified reviews, the ones from accounts with no purchase on record, were systematically more negative than the reviews from confirmed buyers. They were also less likely to describe the concrete experience of using the product, the fit and feel detail that someone who had actually handled the item tends to include. And they contained more of the linguistic markers that deception researchers associate with fabricated accounts.
Put together, those three signals describe a recognizable thing: a review written by someone reconstructing an opinion rather than reporting an experience. More negative, thinner on sensory specifics, and worded in the way people word claims they did not live. None of those signals is individually conclusive, but as a pattern separating no-purchase reviews from verified ones, they are consistent and they point the same direction.
The reviewers who never bought were often loyal customers
Here is the finding that unsettles the sabotage story. When Anderson and Simester looked at who was writing these no-purchase reviews, they did not find only competitors and strangers. Thousands of the reviewers with no purchase record for the product were among the retailer's own most loyal, highest-value customers, people with long, real buying histories at the business.
That is a genuinely strange result, and it matters because it dismantles the clean mental model most reputation advice runs on. In that model there are two populations: real customers who write honest reviews, and bad actors who write fake ones. Anderson and Simester show the line is not that clean. A single person can be a devoted, high-spend customer of a business and also the author of an unverified, more-negative, deception-marked review of a product they never bought. Loyalty and deceptive-style reviewing are not mutually exclusive.
Why a real customer would review something they never bought
The study documents the pattern more than it settles the motive, and the careful reading keeps those apart. What the data establishes is that loyal customers were writing no-purchase reviews. Why they did it is the more interpretive layer, and it should be held more loosely.
Several plausible mechanisms are consistent with the evidence and worth naming carefully. A committed customer may review a product they considered and rejected, or one they returned, or a variant of something they own, treating their relationship with the brand as license to weigh in beyond a single verified transaction. Some may be voicing a grievance about the company through whichever product page is in front of them. The point for a business owner is not to diagnose any one reviewer, but to accept that the population writing about you is broader and less verifiable than the population that actually bought from you.
This is why the star average is a weaker signal than it looks
The Anderson and Simester result does not stand alone. It joins a body of evidence showing that the number buyers lean on most, the average star rating, is a noisier and less trustworthy quality signal than its prominence suggests.
De Langhe, Fernbach and Lichtenstein examined 1,272 products across 120 categories and found that average user ratings did not line up well with independent quality scores, were often built on too few ratings to be statistically informative, failed to predict resale value, and ran higher for pricier and premium-brand items regardless of actual quality. Buyers nonetheless weight the average heavily, more than smarter cues like the number of ratings behind it. Read alongside Anderson and Simester, the conclusion is not that reviews are worthless. It is that an unexamined star average silently blends verified experience, unverified opinion, and a measurable slice of deception into one confident-looking number.
Deception is a predictable, and now regulated, system
If some deceptive reviews come from loyal customers rather than rivals, that does not make the adversarial kind imaginary. Luca and Zervas, using a platform's own filtered-review flags as a fraud proxy, found that fake reviews are more common for businesses with weak existing reputations, few reviews or low ratings, and rise when a business faces more direct competition. Manipulation is a strategic response to reputational and competitive pressure, not a random nuisance. Both things are true at once: deception is partly a competitive weapon and partly a behavior of a business's own base.
Since October 2024 the response side is no longer voluntary. The Federal Trade Commission's Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, prohibits reviews from reviewers who do not exist or never used the product, insider reviews that are undisclosed, conditional incentives tied to a positive rating, and selectively suppressing negative reviews while showing positive ones, with civil penalties up to 51,744 dollars per violation. The practical implication is direct. A business cannot answer a review-integrity problem by quietly generating its own favorable reviews or scrubbing the unfavorable ones. The lawful path is verification and honest response, not counter-manipulation.
How far the finding travels
Anderson and Simester studied one large private-label retailer with its own site and its own purchase records. That is the source of the study's strength, the ability to match reviews to real transactions, and also the limit of its reach. It is an established finding about that setting. Applying it directly to a med spa's Google reviews or a contractor's Yelp page is a reasonable extension, not a proven transfer, because local review platforms rarely expose verified-purchase status and the buyer relationships differ.
What travels cleanly is the structural lesson, not a specific percentage. On any review surface, some share of what is written comes from people who did not have the experience they describe, that share skews more negative and thinner on real detail, and some of it comes from inside your own customer base rather than a competitor. A star average shows none of this. Reading a reputation accurately means treating the average as a starting question, not an answer, and knowing which profiles carrying your name you actually control and can verify.
What verification changes
The operational takeaway is to read an aggregate rating as a starting question rather than a settled fact about real experience, and to build a reputation you can actually see into and stand behind. That starts with ownership: knowing every profile carrying your name across Google, Yelp, Apple, Bing and the directories your field trusts, claiming and verifying the ones that should be yours, and correcting the ones telling a different story.
From there, the disciplines that hold up are watching those surfaces continuously rather than discovering a damaging review weeks late, responding to criticism in the open, and earning reviews only from real customers. None of that removes the unverified or the deceptive review from the internet. What it does is give you a reputation you understand well enough to manage, one where you can tell the signal from the noise instead of trusting a number that quietly contains both.
The evidence
Key findings, with their sources
-
About 5% of the reviews on a large private-label retailer's site came from accounts with no purchase record for the reviewed product, and those reviews were systematically more negative, less likely to describe the product experience, and more likely to carry linguistic deception markers.
established Anderson, E.T. & Simester, D.I., "Reviews without a Purchase: Low Ratings, Loyal Customers, and Deception", Journal of Marketing Research, 51(3), 2014.
-
Thousands of the reviewers with no purchase record were among the retailer's own best and most loyal customers, showing deceptive-style reviewing is not confined to competitor sabotage.
established Anderson & Simester, Journal of Marketing Research, 2014.
-
Across 1,272 products in 120 categories, average user ratings did not converge with independent quality scores, were often based on too few ratings to be informative, and ran higher for pricier and premium-brand items regardless of actual quality, yet buyers weight the average heavily.
established de Langhe, B., Fernbach, P.M. & Lichtenstein, D.R., "Navigating by the Stars", Journal of Consumer Research, 42(6), 2016.
-
Fake reviews are more common for businesses with weak existing reputations and rise when a business faces more direct competition, making manipulation a strategic response to reputational pressure rather than a random nuisance.
established Luca, M. & Zervas, G., "Fake It Till You Make It: Reputation, Competition, and Yelp Review Fraud", Management Science, 62(12), 2016.
-
Since October 21, 2024, the FTC's 16 CFR Part 465 prohibits reviews from reviewers who never used the product, undisclosed insider reviews, conditional positive-only incentives, and suppressing negative reviews, with civil penalties up to 51,744 dollars per violation.
established Federal Trade Commission, "Trade Regulation Rule on the Use of Consumer Reviews and Testimonials", 16 CFR Part 465, effective Oct 21, 2024.
-
Applying the no-purchase-review finding directly to local review platforms is a reasonable extension but not a proven transfer, because those platforms rarely expose verified-purchase status.
emerging Interpretation of Anderson & Simester 2014 scope; flagged as an extrapolation, not a direct finding.
Calibration
What is proven, what is promising, what is unproven
| Evidence tier | Tactics | What the evidence says |
|---|---|---|
| established | No-purchase reviews exist, skew more negative, carry deception markers, and include loyal customers (Anderson & Simester 2014); the star average is a weak quality signal (de Langhe et al. 2016); fraud tracks competitive pressure (Luca & Zervas 2016); 16 CFR 465 is binding law. | Peer-reviewed causal and empirical studies; binding US federal regulation. |
| emerging | Transferring the exact no-purchase share and pattern from one private-label retailer to local review surfaces (Google, Yelp) for med spas, home services, dental and legal. | Structurally reasonable extension; local platforms rarely expose verified-purchase status, so the specific magnitude is not established for these settings. |
| contested | Reading a raw star average as a reliable proxy for real quality or real experience. | The evidence runs against it: averages blend verified experience, unverified opinion, and a measurable slice of deception into one number. |
Reference
Glossary
- Review without a purchase
- A review written by an account with no record of buying the product being reviewed. In Anderson and Simester's data these were more negative, thinner on real detail, and more deception-marked than verified-buyer reviews.
- Deception markers
- Linguistic patterns that deception researchers associate with fabricated rather than experienced accounts, for example fewer concrete sensory details and wording typical of claims a person did not live.
- Verified-purchase status
- Whether a platform can confirm the reviewer actually bought the item. Most local review sites do not expose this, which is why a star average cannot separate real experience from unverified opinion.
- 16 CFR Part 465
- The FTC's 2024 rule making fake, undisclosed-insider, conditional-incentive, and suppressed reviews federal violations, with civil penalties up to 51,744 dollars each. It makes counter-manipulation an unlawful answer to a review problem.
Straight answers
Frequently asked questions
Are deceptive reviews always written by competitors?
No. That is the central surprise of Anderson and Simester's 2014 study. While competitors and outside actors do post fake reviews, the researchers found that many of the unverified, more-negative, deception-marked reviews on a large retailer's site came from the business's own best and most loyal customers. Deceptive-style reviewing is not only an outside attack.
Does this mean online reviews are unreliable?
Not worthless, but weaker than they look. The evidence shows an unexamined star average blends verified experience, unverified opinion, and a measurable slice of deception into one confident number. Separate research across 1,272 products found average ratings often fail to track actual quality. The accurate read is to treat the average as a starting question, not a settled fact.
How much of a business's reviews are written by people who never bought?
In the setting Anderson and Simester studied, a large private-label retailer with its own purchase records, about five percent of reviews came from accounts with no purchase on record. That figure is established for that setting. It should not be assumed to be the exact rate on a local Google or Yelp page, where platforms rarely expose verified-purchase status, so the pattern transfers more safely than the precise number.
If I have fake or unfair reviews, can I just post my own positive ones to balance them?
No, and since October 2024 that path is unlawful. The FTC's 16 CFR 465 prohibits reviews from people who never used the product, undisclosed insider reviews, and suppressing negative reviews, with penalties up to 51,744 dollars per violation. The lawful response to a review-integrity problem is verification and honest reply, and earning reviews only from real customers, never generating or scrubbing them.
What can a business actually do about reviews it did not write?
You cannot remove every unverified review from the internet, but you can build a reputation you can see into and stand behind: own and verify every profile carrying your name, correct the listings telling different stories, watch those surfaces continuously so a damaging review is caught early, and respond in the open. That turns a star average you cannot interpret into a reputation you can manage.
Provenance
Sources
- Anderson, E.T. & Simester, D.I., "Reviews without a Purchase: Low Ratings, Loyal Customers, and Deception", Journal of Marketing Research, 51(3), 2014, 249-269 (established)
- de Langhe, B., Fernbach, P.M. & Lichtenstein, D.R., "Navigating by the Stars: Investigating the Actual and Perceived Validity of Online User Ratings", Journal of Consumer Research, 42(6), 2016, 817-833 (established)
- Luca, M. & Zervas, G., "Fake It Till You Make It: Reputation, Competition, and Yelp Review Fraud", Management Science, 62(12), 2016, 3412-3427 (established)
- Chevalier, J.A. & Mayzlin, D., "The Effect of Word of Mouth on Sales: Online Book Reviews", Journal of Marketing Research, 43(3), 2006, 345-354 (established)doi.org
- Federal Trade Commission, "Trade Regulation Rule on the Use of Consumer Reviews and Testimonials", 16 CFR Part 465, effective Oct 21, 2024; FTC press release, Aug 14, 2024 (established, binding regulation)ecfr.gov
Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.