ReviewProof ⚖

How do you read a ReviewProof risk report?

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

Read the report in the order it is printed. Totals first: imported reviews, flagged count, average rating. Then the risk score with its per-signal breakdown (gating, incentive, duplicate, AI-signal counts, plus burst status). Then provenance: verified versus unverified counts. Every flagged review in the detail traces to the imported row that triggered its signal, and the summary carries the standing caution in plain words: signals are heuristics for human review, not verdicts. The report is a worklist, not a judgement — your judgements, logged, complete it.

What is the risk report?

The risk report is the analysis summary produced from one workspace's imported reviews: a summary block of counts and the risk score, a per-review list of signals with the phrases or patterns that triggered the text checks, and the provenance counts. Each report is sealed with a certificate number (RP- followed by the date and a short code) tying it to the import that produced it, so two reports from different cycles can be compared without ambiguity about which file each describes.

The summary block, line by line

Total is the file you stand behind — if it does not match your source count, nothing below it can be trusted yet. Flagged is how many reviews carry at least one signal; it is not a fake count, and treating it as one is the first misreading. Average rating beside the negative count tells you the profile's shape: a 4.9 with 2 negatives in 300 has a suppression question attached, whatever the signals say. Verified and unverified counts are the provability line. Burst status names the day and its count when detected, inviting the artefact check first (export dates) and the cluster reading second. The note at the end — signals are heuristics — is not legal decoration; it is the instruction for how to hold every number above it.

The workflow is the same on every plan, and the free tier does the complete job for one location: import by CSV upload or paste (up to 500 reviews per import on Free), read the full risk analysis and report, open a provenance certificate for any review, and publish the verified-experience widget. Pro ($29/month, 3 locations, 5,000 reviews per import) adds the full provenance chain detail, the evidence export, AI reply drafts and the reviewer-fraud graph. Agency ($79/month, 50,000 reviews per import) is the multi-location tier. Nothing about the free audit is a sample or a teaser — the report you get free is the real analysis of your real rows.

From report to actions in one pass

Work the breakdown by weight. Gating rows: find the script or flow they quote and replace it. Incentive rows: find the offer and end it. Duplicates: open the clusters, weigh length and links. AI-signal rows: look for concentration in time and source before concluding anything. Bursts: artefact or surge, decided from the rows. Provenance: list the UNLINKED channels and fix the plumbing that loses references. Each line becomes either a solicitation fix, a supplier question, a data fix, or a logged flag — and the next import's report is where you verify the fix landed, which is why the certificate number and date on each report matter more than any single score.

ReviewProof's import is forgiving about format and strict about meaning. Column names are matched flexibly — body, review or text for the review itself; rating or stars; reviewer, author or name; a verified flag and a transaction reference when you hold them. Rows are capped per import by plan (500 on Free, 5,000 on Pro, 50,000 on Agency), which is generous for a single location's history. What the importer will not do is upgrade a review's status: a row that arrives without a transaction link is analysed fully and still labelled UNLINKED, because the label is the product.

Mistakes that undo the work

Screenshotting the score for the team chat without the breakdown invites exactly the misreadings the breakdown prevents. Comparing your score to a competitor's claim about theirs compares different files, baselines and honesty. And reading only the summary — never opening a flagged row — reduces an audit instrument to a weather report; the rows are where the truth of each flag lives.

A detail worth knowing before you start: ReviewProof analyses what you give it. The CSV importer reads reviewer, rating, title, body, date and source columns, plus a verified-transaction flag and a transaction reference when you have them. Reviews with a transaction link are marked verified; reviews without one are marked UNLINKED — kept in the audit, visible in the report, but excluded from the verified-experience widget, which only displays reviews that are both transaction-linked and in a kept or published state. That single rule is the difference between a widget that markets and a widget that proves.

Worked example: a first report, read properly

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

A landscaping firm's first report covered 129 reviews: flagged 26, risk score 18, gating 9, incentive 2, duplicate 6, AI-signal 11, no burst, verified 71, negatives 8. Read badly, it says '26 fake reviews'. Read in order, it said: the gating 9 all quoted one text message the office sent after good jobs — replaced that week. The duplicates were two clusters, one a bought-looking set of four long identical texts (flagged, supplier questioned), one a pair of short coincidences (kept, noted). The AI-signal 11 scattered across four years — a formal-writing customer base, no action. The 58 UNLINKED rows traced to one platform export without references — a plumbing note for next cycle. Four different actions, none of them 'delete 26 reviews'.

Your five-action checklist

  1. Reconcile the total against your source file before reading further
  2. Read flagged as 'carrying a signal', never as 'fake'
  3. Work the signal breakdown by weight: solicitation first, clusters second
  4. Answer the burst question from the rows: artefact, surge or purchase
  5. Finish with provenance: which channels lose transaction references, and why

Frequently asked questions

What does 'flagged' mean on a review?

The row triggered at least one signal — gating language, an incentive marker, duplicate text, AI-generation pattern, or membership of a burst day. It is a prompt to read the row, not a conclusion about it.

Where does the report's certificate number come from?

Each import's report is sealed with a number in the form RP- plus the date and a short random code, tying the summary to that import.

Does the report judge AI signals as proof?

No. The summary states it plainly: AI-generation signals indicate patterns associated with generated text; they do not prove a review was AI-written. That wording is part of the product.

What is the single most useful comparison for a report?

The same location's previous report. Solicitation counts falling after a script fix, verified share rising after a plumbing fix — trends are the audit working.

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.