--- title: "How ReviShield works: from detection to decision" description: "How a review moves from detection to decision: who assesses it, who files, what is automated, and what happens when there is no basis for a request." url: https://revishield.com/how-it-works updated: 2026-11-06 --- How it works # From detection to decision, and who does what Automation collects and prepares. People assess, approve and file. The platform decides. Here is every step, with the module behind it. - 1 System ## Detected A new or changed public review appears on a protected profile. Shield Monitor picks it up from your own API access, a licensed data vendor or a record you supply. We do no scraping of our own. Attack Radar watches for waves of reviews and repeated wording across profiles. Shield Monitor, Attack Radar - 2 AI, checked by people ## Triaged AI proposes a ground and quotes the words it relies on, or abstains. Our code checks every quote against the original text. AI has no permissions and cannot change a case. Review text is treated as untrusted input. Policy Engine - 3 Named analyst ## Assessed A named analyst reads the review against the current rule and writes one of three verdicts, with a single ground and the reasoning. On Portfolio and up, ProofMatch privately compares the review with your records and returns a match, no match in scope, or not testable. A match closes the “no genuine experience” ground. No match is only a signal. Analysts are never paid per removal. Policy Engine, ProofMatch - 4 You ## Approved You state what you checked, see the success fee quoted for this review and approve the filing. If we know a statement is false, we do not file, whatever the form says. Evidence Vault & Case Docket - 5 Named person ## Filed A named person files the request in the platform’s own tool and stores the confirmation. Our software has no way to submit anything to a platform. One ground per review. One-shot appeals and batches are guarded. Evidence Vault & Case Docket - 6 The platform ## Decided Only the platform decides. We record the outcome on the timeline. If a review goes, we check it at 14, 30, 90 and 210 days. Outcome Intelligence Other ways a case can end ## Not every case goes to filing ### When the verdict is no You get a written closure citing the rule and a reply draft from Response Studio. On Growth and up, Recovery Router turns the complaint into a task for the right manager. None of this is conditional on the review being withdrawn. A case reopens only with new evidence. ### When it needs a lawyer Defamation or falsity claims, notices with legal content, pre-action letters, court orders and identifying an author go to a lawyer in every country. Legal Partner Desk hands over the documented case. The lawyer deals with you directly at a fixed price, and no commission flows either way. ### When it is an attack Attack Radar flags a spike, for example 15 or more 1-star reviews in 24 hours. It runs as one incident, outside all limits. We bundle the evidence for Google’s extortion report form: screenshots of demands, dates, sender, direct links and a timeline. Our advice is the same as Google’s: do not pay. Platforms ## Google and Trustpilot first, with their rules read by a person A case cannot move forward on a rule nobody here has read. Every rule in our register has a source, a date and an owner. ### Google - Grounds include fake engagement, conflict of interest, harassment, personal information, off-topic content and advertising. - Disagreeing with a review is not a ground. - Filed through the Reviews Management Tool. One appeal per review, in batches of up to 10, with the outcome by email. - Separate forms exist for legal removal and for extortion. ### Trustpilot - Grounds are harmful or illegal content, personal information, advertising, no genuine experience, and a different business. - Disliking the stars is not a ground. A critical tone is not either. - For “no genuine experience” the reviewer is first asked for details and has three days to respond. - Appeals run through the decision email. Trustpilot sanctions misuse of flagging, including bulk or one-sided reporting. Other platforms join in later phases, and each stays marked “unverified” until a person has read its official rules. Some sites can only be monitored, because ordinary reports do not work there. We say which when you ask. Checks on ourselves ## A success fee pulls towards filing, so we built the pull back **Precision brake** If rejections on a platform exceed 50% over 30 days, filing there pauses and every case needs a second person. **No bounty for analysts** Analysts are never paid per removal. We measure how often platforms agree with what they file. **One standard** We apply the same test to your suspicious positive reviews. **No detector scores** AI-detector scores help us sort work internally. They are never a ground and never go into a filing. The modules ## Ten modules, one record What each one does for you, and where it is included. ### Watch - **Shield Monitor** All plans Every new or changed review on your protected profiles, with alerts. It keeps running even if you go over a plan limit. - **Attack Radar** All plans Spots waves of reviews, repeated wording and patterns across profiles and platforms. An attack is one incident, outside all limits, and never charged per review. ### Judge - **Policy Engine** All plans Checks a review against the platform’s published rules for your country, names the one ground that may apply and lists the evidence it would need. - **ProofMatch** Portfolio and up Privately compares a review with your own records. A match closes the “no genuine experience” ground. No match is only a signal for the analyst and proves nothing about the reviewer. ### Record - **Evidence Vault & Case Docket** All plans Evidence, messages, filings, appeals and the outcome on one verifiable timeline. - **Outcome Intelligence** Growth and up Your results by platform and ground, with case counts and time to decision. Rates always come with their denominator. ### Respond - **Response Studio** All plans Reply drafts in your brand’s tone and the customer’s language, checked before anything goes out. You approve each one. - **Recovery Router** Growth and up Turns a negative review into a task for the right manager and tracks the fix. Resolving a complaint is never conditional on the review being withdrawn. - **Fair Review Capture** Growth and up Neutral review requests to all your customers, by QR code, NFC and direct links. No filtering of happy customers and no incentives. ### Escalate - **Legal Partner Desk** All plans Documented cases that need legal judgement go to an independent lawyer, by country and platform. The lawyer deals with you directly. No automatic letters. ## Questions about the process ### How long does a decision take? We set a target for our own first assessment: one or two working days, depending on your plan. The platform sets its own pace. Google says its evaluation typically takes several days. We cannot promise a date for a decision. ### Can I report a review myself? Yes, and it costs nothing. The free review check shows how. What you pay us for is the judgement, the evidence record and doing this consistently across every profile. ### What access do you need? Where access is needed, you invite us to a manager role. We never take passwords, we record your consent with its date and scope, and we remove access within 7 working days of the end of the service. ### Do you use bots or automated reports? No. No bots, proxy pools, browser automation of platform forms or CAPTCHA solving. A named person files each request.