Editorial Standards

Where findings come from

Most findings start from monitored feeds: government health and safety agencies, recall databases (UK OPSS and FSA, US FDA and CPSC), and established science and health publications. When someone requests research on a specific product we don't already cover, we also query PubMed and OpenFDA directly for relevant published studies or agency records.

How a finding is drafted and rated

We use AI assistance (Anthropic's Claude) to draft the plain-English write-up of each finding in its own words, and to propose an initial severity rating. The AI is explicitly instructed not to reproduce source text verbatim — every finding links back to its original source so you can read the primary material yourself.

Severity is one of three levels, applied against fixed criteria:

• HIGHActive recalls, confirmed deaths or hospitalisations, banned substances found in consumer products, or immediate danger to vulnerable groups such as children or pregnant women.
• MEDIUMPeer-reviewed research linking product ingredients to long-term health risks, regulatory warnings, or products under active investigation.
• LOWObservational studies, precautionary advice, minor reformulations, or general wellness concerns.

Review before and after publication

Every AI-drafted finding is checked against its source by our AI editorial review process. Findings it rates as high-confidence, from established agencies and health publications, can be published without waiting for a person; everything else sits in a review queue until a member of our team approves it. Our team reviews auto-published findings afterwards, and an automated fact-check re-reads published findings against their sources. We're a small, actively-developed platform, not a newsroom with multiple layers of editorial sign-off, so treat this as a real but proportionate check, not a guarantee of exhaustive fact-checking on every word.

Corrections

Every finding page has a “Report an issue” option. Reports go to our team for review, and can result in an edit or the finding being taken down. If you spot something wrong, please tell us — that's the fastest way to get it fixed. Automated compliance checks also re-read published findings, and when they find urgent problems in wording (for example an unsupported claim about a named brand, or language that reads like medical advice) they can replace that phrase with more cautious wording straight away. Our team is notified of every such change and reviews it.

Keeping revenue separate from ratings

ProductSafer earns money from advertising, sponsorships, affiliate links, and paid products including the ProductSafer Verified certification and B2B recall monitoring. None of these influence which findings get published, how they're rated, or any product's safety score — those are separate systems with no code path connecting payment to editorial outcome. Full detail on each revenue source is in our disclosure policy.