ReviewProof ⚖

Why is a star rating without provenance not enough?

By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026

Run the free review audit See pricing · See a sample provenance report

Short answer

Because a star average answers none of the questions a sceptical reader asks. Which transactions stand behind these reviews? How were they solicited — universally, or through a filter? What happened to the negative ones? A rating is a summary of opinion with the evidence discarded; provenance is the evidence. That is why ReviewProof's products are a report, a certificate and a widget rule rather than a score: the report shows the patterns in your own rows, the certificate shows the record behind one review, and the widget displays only reviews whose transactions can be shown. Stars are the headline; provenance is the article, the sources and the corrections log.

What is a rating without provenance?

A rating without provenance is an aggregate — average stars across a set of reviews whose origins, solicitation and moderation are undocumented. Aggregates are not lies; they are compressions, and like all compressions they discard exactly the information a challenger needs. Provenance is the uncompressed record: row, transaction link, chain of custody. The argument of this page is not that ratings are useless but that they are a claim, and claims need the record behind them at the moment they are doubted.

The three questions that dissolve a bare rating

Doubt arrives in three standard forms, and a bare rating survives none of them. The customer's: 'these could be anyone's words' — unanswerable from an average, answerable from certificates behind a widget. The platform's or regulator's: 'show us how these were obtained' — unanswerable from an average, answerable from the audit (solicitation counts at zero, cohort documented, moderation logged, negatives kept). The counterparty's — buyer, franchisor, lender: 'what is this reputation asset actually made of?' — unanswerable from an average, answerable from era maps and linked shares. Notice that in all three the rating's number is not even disputed; it is simply set aside as the wrong kind of object, and the conversation moves to the layer the business either has or has not built.

Keep the negative reviews. That instruction surprises people mid-crisis, but suppression is itself a violation under the FTC rule, and a profile scrubbed clean during an attack reads as manipulation to anyone who later examines it. ReviewProof has no silent-delete path: moderation decisions are kept, published or flagged internally, each one is logged in the chain of custody, and the evidence export counts the negatives you kept. The strongest response to a suspicious review is a boring, complete record — the row, the dates, the signals, the decision, all preserved while you dispute through the platform's own process.

Keep the stars — demote them

The practical resolution is not to abandon ratings (customers will use them whatever you publish) but to demote them to their proper rank in your own house: an output you observe, never an input you manage. Manage the inputs instead — solicitation universality, linked share, signal lines, moderation quality, reply quality — and let the rating be whatever honest operations make it. Publish accordingly: your platform profiles will show what they show, but the surface you control, your own site at the decision moment, carries the widget: fewer reviews, each with its papers in order, under an average computed over the provable set. You are not hiding the stars. You are finally showing the working.

After the first audit, the habit that matters is cadence. Re-import on a regular rhythm — monthly suits most locations — so new reviews enter the provenance record while the underlying transactions are still easy to match. Each import logs provenance events, moderation starts from kept (no suppression), and the certificate for any review shows its chain: imported, transaction-linked or not, analysed with its signals, moderated with its decision. A record built gradually like this is worth far more in a dispute than one assembled in a panic the week the dispute arrives.

Mistakes that undo the work

Managing to the rating — setting score targets, celebrating movements, tasking staff with the number — reliably regenerates every violation this library documents, because the number cannot be raised honestly at the speed management wants. Quoting your platform rating in contexts that imply verification ('our verified 4.9') borrows provenance language for an unprovenanced object. And disparaging a competitor's higher rating without evidence is merely envy with a theory.

Triage starts from your own data, not from the public profile. Import the reviews, look at which rows carry which signals, and separate three questions that are often mashed together: is this review about a real transaction, was it solicited in a way the FTC rule prohibits, and what record do we hold if anyone asks? ReviewProof answers the first two from the imported rows — transaction link present or UNLINKED, gating or incentive language present or absent — and answers the third with the provenance chain. What it will not do is declare a review fake: signals are heuristics for human review, and acting as if they were verdicts creates a new problem while solving none.

Worked example: two competitors, one careful customer

A fictional worked example, written to show the numbers and the decisions in practice.

A couple choosing a builder for an extension compared two firms. Firm A: 4.9 stars, 212 reviews on a platform, nothing behind them anywhere. Firm B: 4.6 on the platform — and on its own site, a verified widget showing 81 reviews at 4.5, each opening into a certificate with a masked invoice reference, kept one- and two-star reviews included (three of them, answered with specifics). The couple chose Firm B at a higher quote, and said why at the first meeting: 'We could check his worst reviews were real jobs. We couldn't check anything about the other one's best ones.' The stars were never the product. The checkability was.

Your five-action checklist

  1. Write down the three doubt questions and test your current presence against them
  2. Demote the rating internally: observe it, never target it
  3. Manage the inputs: ask universality, linked share, signal lines, replies
  4. Publish the verified widget at your own decision points
  5. Use rating language precisely; never let 'verified' touch an unlinked number

Frequently asked questions

Are star ratings useless, then?

No — they are a discovery signal and a summary. They are insufficient as proof, which is the moment businesses actually need: doubt, dispute, due diligence.

What average does the verified widget show?

The widget displays the transaction-linked, kept or published set; any summary alongside it is computed from verified reviews only and the AI summary is labelled as AI-generated.

Can provenance raise our rating?

Indirectly and honestly: universal asks usually collect more reviews overall, and provable quality converts careful buyers. Targeting the rating itself is how violations happen.

What should we quote in a tender or a sale?

The record: linked share, era map, signal lines at zero, negatives kept, evidence packs by date. Let the examiner compute whatever average they trust from documents.

Keep reading

Related guides from the ReviewProof library:

Start with the tool itself: open the free audit desk, see how ReviewProof works, compare Free, Pro and Agency plans, inspect the sample provenance report, or browse all ReviewProof guides.