Treat ratings as a starting point. Match reviews to the exact product and use you need, examine incentives and relationships, and trace important claims to independent evidence. Preserve the listing, review context, and written seller promises. Polished writing does not establish fabrication, and a disclosure does not prove the experience happened. An evidence worksheet organizes unanswered questions; it cannot calculate the probability of fraud or decide a legal violation.
Last updated: 2026-10-15
A holiday gift has a 4.8-star rating, pages of confident praise, and several photographs. What you actually need to know is narrower: did anyone use the exact model for the task your recipient will perform? A fluent paragraph cannot answer that question by itself. Nor does awkward language prove that a reviewer is dishonest.
This guide is for U.S. shoppers evaluating consumer reviews and promotional testimonials before buying ordinary goods. It explains evidence checks, not a method for identifying an author’s software. It does not certify a product, rank review platforms, or assume every missing disclosure establishes a violation. The practical objective is to identify what remains unverified before a sale countdown makes the decision feel urgent.
Separate the writing tool from the claimed experience
16 CFR 465.2 prohibits specified business conduct involving material misrepresentations about a reviewer’s existence, use of a product, or actual experience. We checked the eCFR compilation displayed as current through October 13, 2026. The focus is the misrepresented experience, rather than whether the sentence is grammatically polished.
There is a relevant update to older AI-review headlines: on December 22, 2025, the FTC set aside its 2024 Rytr consent order. The later order found the complaint insufficient to support the alleged FTC Act violation and recognized legitimate uses of an editable review draft. That case-specific reversal did not repeal Part 465. An AI-assisted account of real use and an invented owner describing a nonexistent test are different situations.
As an editorial practice, ask what claim you can check. “The ear cushions stayed comfortable during my four-hour train trip” contains a use condition. “An exceptional, life-changing sound experience” supplies little information for checking fit. Either sentence could have been written by a person or edited with software. The relevant question is whether there is credible support for the experience and whether it answers your needs.
Read incentives and relationships as separate evidence
The FTC’s rule Q&A distinguishes sentiment-conditioned incentives from rewards for honest reviews without that condition. A business cannot require positive or negative sentiment, expressly or implicitly; disclosing the reward does not cure that condition. Other FTC Act and disclosure issues can still apply.
Separately, the Endorsement Guides, particularly section 255.5, address unexpected material connections that can affect an endorsement’s credibility, including payment, free products, or relevant relationships. The FTC’s explanation of the Guides says they interpret how the FTC Act applies; the Guides themselves do not have the force of law. Avoid treating every disclosure question as the same violation of the review rule.
For a purchase decision, record the relationship separately from the result. A clearly disclosed sample can still produce useful testing. A reviewer who paid retail can still perform a poor test. A missing disclosure means the commercial context is unresolved, rather than proving no connection exists. Look for plain explanations of who supplied the product, paid for the post, or benefits from the purchase link, rather than inferring everything from a badge.
See how an average can change without the product changing
Consider an entirely hypothetical, unweighted rating system. Twenty reviews have a total of 80 stars, so their average is 80 ÷ 20 = 4.0. Add twenty five-star reviews and the new total is 80 + 100 = 180 across forty reviews: 4.5. The displayed average rises by half a star without any change in the product.
For this illustration only, assume the additional reviews were manufactured. That is an assumption supplied to demonstrate arithmetic, not a finding inferred from a high rating. Real platforms may weight, filter, or round ratings differently. A sudden burst could also follow a promotion or a product launch. The calculation shows sensitivity to the added observations; it does not estimate the prevalence of manipulation or tell you which reviews to remove.
The FTC’s shopping guidance advises checking several sources rather than relying on stars, and notes that misleading reviews can be favorable or unfavorable. Reading only one-star comments is therefore not a complete antidote. Look for repeatable, specific issues across independently sourced accounts, and distinguish a seller-service complaint from a product-performance claim.
Start with a real purchase scenario
Suppose you are considering a fictional $120 wireless headset, model H2, as a gift. The recipient needs four hours of comfortable listening and reliable calls with a particular phone. These are invented purchasing facts, not a recommendation or an account of our testing.
Before reading reviews, write those requirements down. Match the model, generation, size, seller, and date. A review of an earlier H1, or of an order arriving quickly, cannot establish H2 call quality. Open enough of the review to see whether the writer describes unboxing, brief listening, repeated use, or a measured test. A claim copied across several websites is still one originating account until you find separate evidence.
Treat “verified purchase” according to what that platform actually says it verifies. It should not substitute for checking test duration, exact variant, or incentives. Likewise, a photograph of a box supports a narrower observation than a demonstrated four-hour listening session. More detail creates more claims to check; it does not automatically create more truth.
An eight-point information worksheet
The table below is our editorial checklist. It measures the availability of information useful for this decision, not authenticity, safety, legal compliance, or a probability of fraud. The four dimensions are not statistically independent, and the total has no validated purchasing threshold.
| Dimension | 0 points | 1 point | 2 points |
|---|---|---|---|
| Product match | Product unclear | Brand/family or variant uncertain | Exact model and relevant variant identified |
| Use context | No meaningful use described | Brief use or unboxing | Actual use conditions and duration match the question |
| Commercial context | Relationship unresolved | Partial or ambiguous explanation | Product supply/payment/relationship explained sufficiently to assess |
| Corroboration | Only copied praise | One relevant external fact checked | Key claim checked with separate, relevant evidence |
Assume listing A’s evidence identifies H2 but consists of brief unboxing, unresolved commercial context, and reposted praise. Its information total is 2 + 1 + 0 + 0 = 3/8. Assume listing B identifies H2, contains a relevant four-hour use account, clearly explains a manufacturer-provided sample, and has a manual confirming phone compatibility but no independent call-quality test. Its total is 2 + 2 + 2 + 1 = 7/8.
Neither total proves truth. B still leaves call quality unresolved; its manual cannot validate the listening account. A’s low score could reflect sparse documentation rather than fabrication. If reliable calls are essential, that unresolved requirement can outweigh the higher total. Seek a relevant independent test or a meaningful opportunity to assess the product under the actual purchase terms. Do not turn the worksheet into another star rating.
Use the signal to choose the next step
| Observation | Next action | Conclusion to avoid |
|---|---|---|
| Many similarly worded reviews appear together | Check dates, model references, and distinct originating accounts | “This timing proves AI fraud” |
| Review discusses checkout before use | Use it only for the experience actually described | “It proves long-term reliability” |
| Free sample or affiliate arrangement is disclosed | Check use conditions and independent support | “Disclosure guarantees honesty” |
| Ratings are high but your required feature is untested | Ask a specific question or postpone reliance on the claim | “High average means it fits my needs” |
| Incentive requires a five-star rating | Save the incentive terms and report the concern | “Adding an ad label cures the condition” |
Before assuming negative reviews were suppressed, check the sorting, filters, and selected variant. Preserve what you actually saw rather than asserting why absent content disappeared. A complaint is more useful when it identifies observable behavior and lets the platform or regulator investigate intent and responsibility.
Preserve enough context to make the evidence usable
The FTC’s online-shopping records checklist recommends retaining seller information, order details, policies, promises, and communications. For a review concern, add a dated screenshot showing the product URL, model/variant, review text, author identifier, publication date, rating, and any relationship disclosure. Save the sort/filter setting and the review’s direct link where available. Capture the incentive message or seller answer in context, not just the most alarming sentence.
Keep the original locally. If sharing a copy publicly or with others, redact unrelated names, addresses, order numbers, and account details. A screenshot records what was displayed; it does not independently establish who wrote the review or that the test occurred. An edited page later may require explaining the two dates rather than treating either image as the complete history.
Use this question template before checkout:
Does this review concern model [number/variant]? How long was it used, under which conditions, and was the product supplied or the review rewarded? Please confirm whether [specific feature] works with [my use condition], and identify the source of that claim in writing.
Record who made the claim separately from who confirmed it. A seller forwarding another review has not necessarily promised the feature; “usually works” does not resolve a particular phone, software version, or use condition. Ask for the scope. If the seller cannot verify it, leave the result unresolved instead of filling a worksheet cell with an assumed yes. Preserve the original question and complete answer so you can distinguish advertising, personal experience, and formal product documentation.
A short action sequence
- Write the recipient’s two or three essential uses before looking at the average rating.
- Trace each decisive claim to the exact model, an actual use account, and its commercial context.
- Find independent support for the important uncertainty; count repeated copies once.
- Save the listing, relevant review, disclosure, and written promise before paying.
- Report concrete concerns through the platform’s process and FTC ReportFraud, distinguishing observation from suspicion.
The FTC’s December 2025 consumer alert recommends reporting suspicious reviews to the hosting platform and including “fake review” when reporting to the FTC. Reporting does not resolve a purchase dispute automatically. For an actual product failure, use the warranty and repair records guide; for a separate billing problem, consult the credit-card billing-error guide.
Frequently asked questions
Can I ask AI to decide which reviews are fake?
It can help organize claims and missing information, but a label is not proof of the author’s identity or experience. Require links to the original reviews and check the underlying evidence yourself before making an accusation or purchase decision.
Is an incentivized review necessarily useless?
No. Separate the incentive’s terms, the disclosure, the actual test, and corroboration. A useful disclosed test and an undisclosed relationship are different issues; neither should be reduced to a single trust badge.
Does the FTC rule guarantee my money back?
No. The rule Q&A says it creates no private right of action. Transaction-specific rights and policies determine remedies.
Sources checked October 15, 2026. The rating arithmetic, headset scenario, and worksheet are hypothetical editorial examples. The Rytr update describes a specific December 2025 order; it is not a claim that all AI-assisted reviews are unlawful or that all generated reviews are permitted.
Sources and review
Sources reviewed:
- eCFR — 16 CFR Part 465, Consumer Reviews and Testimonials Rule
- FTC — Consumer Reviews and Testimonials Rule: Questions and Answers
- eCFR — 16 CFR Part 255, Endorsement Guides
- FTC — Endorsement Guides: What People Are Asking
- FTC — Online Shopping
- FTC — Warns businesses about fake reviews, December 2025
- FTC — Rytr Order Reopening and Setting Aside Order, December 22, 2025
General U.S. consumer education, not a legal determination, product certification, or individualized purchasing advice. The worksheet is an editorial aid, not a fraud detector. Applicable law, platform policies, product facts, and available remedies can change.
Frequently asked questions
Does polished or AI-assisted writing prove a review is fake?
No. Investigate the claimed experience and supporting evidence. Editing or translation can coexist with real experience; detailed prose can also describe an experience that never happened.
Does a free-product disclosure make every claim trustworthy?
No. It explains a relationship, not the truth of the test results. Check whether the reviewer actually used the exact model in the conditions relevant to you.
Does reporting a fake review automatically entitle me to a refund?
No. The rule creates no private right of action. Transaction-specific rights and policies determine purchase remedies.
