Ethics — what PlanOfAction will and will not do
PlanOfAction is an appeal prep desk. The seller supplies the facts; the tool structures, checks and phrases them. These rules are enforced in code and by tests, not just stated.
We never submit, log in or scrape
The tool never submits anything to Amazon, Google or any other platform, never logs in to a seller account, and never scrapes either. There is no SP-API connection and no browser automation. The seller copies the output and submits it themselves.
We never invent facts
Every sentence in an exported Plan of Action must be tied to one of the seller's own answers or to evidence they confirmed they have (the claim-trace rule), and every date, number and name in it must appear in those answers or in the evidence they hold. A sentence that fails either check is removed before export and counted on the draft page, and the removal is logged without its content.
We never create documents
No part of the product generates, drafts, templates, edits or "improves" invoices, receipts, supplier letters, letters of authorization, utility bills, IDs, screenshots, or any other document a platform could accept as proof. Requests to do so are refused, and so are requests to mention evidence the seller does not have. The evidence checklist only asks what the seller has; it never collects files.
We do not promise outcomes
We cannot and do not guarantee any outcome. Nobody outside the platform can. The product never shows the odds of an appeal succeeding, and never uses language that predicts the platform's decision. Refunds are never tied to the platform's decision.
Triage before money
Cases that need a specialist — counterfeit and intellectual property complaints without a retraction, forged or manipulated documents, compromised accounts, withheld funds, anything mentioning law enforcement, arbitration or litigation, and product safety complaints involving an injury, a recall or a regulator — get a plain explanation and a referral, and are never offered a paid build. Information-only notices (identity verification) get an information panel. If a referral trigger appears after payment, the build stops. Before the seller has opened the finished draft, the payment is refunded in full automatically. After that, during the included revision, the draft stays available, the revision stops, and a refund request goes to the founder, who decides it; the seller is emailed the decision.
Honest plans for deliberate violations
If a seller tells us a violation was deliberate, we will not help argue that it did not happen, blame someone else, or call it an accident or a misunderstanding. We will help them write an honest corrective plan: what happened, what they have fixed, and what they will do so it does not happen again.
Data minimization
Notice text, answers and drafts are encrypted at rest and deleted 30 days after the notice is pasted, or 30 days after payment for a paid build, screenshots after 24 hours, and anything at once when the seller deletes their case. Nothing is used to train models. Logs never contain notice text, answers or email addresses.
AI is labeled
Text written by an AI model is labeled where it appears: the "AI explanation" on the result page, the follow-up questions in the paid interview, the Plan of Action draft and the AI rubric score. Scores from the AI rubric are shown separately from the automatic checks and are described as our own rubric, not the platform's criteria.