Reviews, ratings and the verified transaction: reputation after generative text
Platforms are fighting generated text with text analysis, which is the one contest they cannot win. The signal they actually need — that a real person who really transacted wrote this — exists in their own systems and is not bound to the review.
How does review integrity work after generative text?
Not by examining the review. Text analysis and behavioural patterns were calibrated against humans writing at scale, and generated text defeats both. The durable signal is not the words but whether the reviewer actually transacted — which the platform knows and never binds to the review.
- Reputation systems infer authenticity from content and behaviour. Neither establishes that a specific human who transacted wrote this review.
- Verified purchase badges are platform-internal assertions, not verifiable artefacts, and they do not travel to aggregators or regulators.
- Linking reviews to transactions reduces reviewer pseudonymity, which is a real cost that selective disclosure partially addresses.
Part of Gaming and consumer marketplace identity
The two signals and what happened to them
Review integrity has historically rested on content analysis — linguistic patterns that distinguish fabricated reviews from genuine ones — and behavioural analysis — posting velocity, account age, rating distributions.
Generated text eliminated the first. A model produces a review indistinguishable from a genuine one because it was trained on genuine ones, and detection of generated text has proven unreliable enough that several vendors have withdrawn such claims.
Purchased accounts eliminated the second. A genuine account with real history, real purchases and normal behaviour, rented for a review, produces no behavioural anomaly because there is no anomaly to produce.
What a verified purchase badge is
The closest existing mechanism, and worth crediting properly: it links a review to a transaction on the platform, and it does raise the cost of fake reviews materially.
Its limits are structural rather than implementational:
| Property | Verified purchase badge | Portable receipt |
|---|---|---|
| Links review to a transaction | Yes | Yes |
| Verifiable by the platform | Yes | Yes |
| Verifiable by an aggregator | No | Yes |
| Verifiable by a regulator | No, without platform cooperation | Yes |
| Survives platform data loss or exit | No | Yes |
| Distinguishes reviewer from account holder | No | Yes, if the reviewer signs |
Rows three and four matter increasingly. Regulators pursuing fake review enforcement need evidence they can evaluate, and aggregators republishing reviews currently pass along a badge they cannot check.
The receipt
{
"type": "manav-stmt/1",
"action": "review_submission",
"transaction": "[order reference, hashed]",
"item": "[product or service identifier]",
"review_hash": "[hash of the review text as submitted]",
"transacted_at": "[date]",
"reviewer": "[credential assertion]",
"platform": "[countersignature]"
}The review hash is what prevents a second failure mode nobody discusses: a genuine review being edited after publication, either by the reviewer under pressure or by the platform. The hash fixes the text at submission.
The pseudonymity cost, stated properly
This is the honest objection and it deserves the body of the article rather than a footnote.
Reviewers rely on pseudonymity. A person reviewing a medical service, a legal service, a dating platform or an employer has legitimate reasons not to be identifiable. Linking reviews to transactions and to a durable credential erodes that.
Three partial mitigations, none complete:
- Selective disclosure. The receipt proves a transaction occurred without revealing which transaction or which reviewer credential, using a zero-knowledge construction. Technically feasible, operationally heavier.
- Categorical scoping. Apply receipts to product reviews and not to service categories where pseudonymity has protective value.
- Reviewer choice. Let the reviewer decide whether to attach a receipt, and display the difference. That preserves pseudonymity at the cost of weaker signal for those who choose it.
Option three is probably right for most platforms and it is the one that looks weakest in a product review, because it does not eliminate unverified reviews. It does mark them, which is what a reader actually needs.
Where this matters commercially
Regulatory enforcement on fake reviews has increased, and platforms are being asked to demonstrate the integrity of their review corpus rather than assert it.
A platform that can report the share of its reviews carrying verifiable transaction linkage has an answer. One relying on internal badges is asking a regulator to take its word, which is a weaker position each year.
Why the verified-purchase badge is not enough
| Claim | Established? |
|---|---|
| An order exists on this account | Yes |
| The account belongs to a distinct person | No |
| The reviewer is that person | No |
| The review is portable to another platform | No |
The last row is the commercially interesting one. A review is the most valuable asset a reviewer builds and it cannot leave the platform, which is why reputation is a lock-in mechanism rather than a portable credential.
Objections and honest limits
“Fake reviews are a text problem.” They were, while writing was the expensive part. Generation removed the cost, so the scarce thing is now the transaction rather than the prose.
“A receipt reveals purchase history.” Only what the reviewer chooses to present, and only that a qualifying transaction occurred — not the order, the price or the history.
Binding a review to a transaction
- Sign at the transaction, not at the review. The purchase is the scarce fact.
- Let the reviewer present it later. Including on another platform.
- Reveal the minimum. That a qualifying transaction occurred, not the order details.
- Keep text analysis as a secondary signal. Useful, and no longer the primary control.
Terms used here
- Verified purchase
- A platform flag indicating an order exists on the reviewing account — a claim about the account, not the person.
- Portable reputation
- Review history a person can carry between platforms, which no major platform currently supports.
- Incentivised review
- A review exchanged for value, which is a disclosure problem distinct from authenticity.
Frequently asked questions
Does this eliminate fake reviews? No. It eliminates reviews from accounts that never transacted. A genuine purchaser paid to write a favourable review produces a valid receipt.
What about pseudonymous reviewers? That is the central cost and the article treats it as such. Reviewer choice with visible marking is the most practical mitigation; selective disclosure is the technically stronger one.
Is this not just a verified purchase badge? The badge is a platform assertion. A receipt is verifiable by aggregators, regulators and readers without the platform's cooperation, and it survives the platform.
Why hash the review text? To fix the text at submission, so post-publication editing — by anyone — is detectable.
Why doesn't text analysis work any more? It was calibrated against humans writing at scale. Generation removed the cost of good prose, so the text carries much less signal.
What does a verified-purchase badge prove? That an order exists on that account. It says nothing about whether the account is one distinct person or who wrote the review.
Does a receipt expose purchase history? No. It can reveal only that a qualifying transaction occurred, without the order, price or history.
Where this fits in Manav
Manav proves a specific person authorised a specific action, without a vault, a token or surveillance. The biometric never leaves the device and the platform receives a signature rather than a profile.
Sources and further reading
- Regulatory rules and enforcement actions addressing fake consumer reviews.
- FTC — business guidance on marketplaces and consumer protection
- Published research on the limits of generated text detection.
- FCC — protecting consumers from SIM swap and port-out fraud
- FTC Rule on Consumer Reviews and Testimonials (16 CFR Part 465)