How we work
Agents draft. A gate checks. A person reviews.
Cross-referencing recall databases and reading ingredient panels for every product is a second job. So we built a pipeline that does the reading — and a human still has to approve what it finds before it reaches you.
What the agents actually do
Every page on SafeMama starts as a draft written with AI research agents, pointed at the same named public sources listed in our Sources registry — the FDA, EPA, CPSC, EWG, and the rest. The agents pull the sourcing together and write the first pass. They do not decide what publishes.
What has to pass before anything goes live
Every draft runs through a deterministic sourcing check before it can publish — not an AI judgment call, a fixed set of rules. It fails the page if a product claim has no FTC disclosure, if a negative claim about a product or brand has no citation, or if a source we're not allowed to reproduce shows up in a table. None of that is optional, and none of it is graded by the same system that wrote the draft.
After that, a person reviews the page before it ships. We're building out a named, credentialed reviewer panel next — see sources & contributors for where that stands today.
What we don't claim
- No named reviewer panel yet.We won't say a page was reviewed by a specific named person until that's actually true.
- No medical claims. Nothing here diagnoses, treats, or replaces a conversation with your pediatrician.
- Affiliate status never decides a pick. See our disclosure page for the full accounting.