Fact-Checking Policy

Every claim on this site is sourced. This page describes our sourcing standards.

Source hierarchy (strongest to weakest)

1. Primary peer-reviewed: PMC (PubMed Central), Nature, NEJM AI, JAMA, npj Mental Health Research. Used for all clinical-efficacy claims.
2. Regulatory primary: FDA AI/ML-enabled device list, DOJ/FTC filings, state medical and psychology boards. Used for clearance and compliance claims.
3. Vendor primary: vendor pricing pages, vendor docs, vendor-published security pages (SOC 2 attestations, BAA availability). Used for pricing, feature, and compliance claims.
4. Verified user data: G2, Capterra, Trustpilot — current counts, linked directly. Used for rating claims.
5. Tier-1 journalism: MobiHealthNews, STAT News, Fierce Healthcare, NYT, WaPo, WSJ. Used for industry-context claims.
6. Tier-2 journalism: Behavioral Health Business, Healthcare IT News, Verywellmind. Used as secondary corroboration.

Vendor blogs and PR material are never used as primary sources for factual claims. They are cited only when reporting what the vendor says about itself.

Clinical-evidence verification

Every clinical-efficacy claim links to a specific peer-reviewed study. When we state effect sizes (for example, "51% reduction in depression symptoms"), we link directly to the study and cite the sample size and methodology constraints. Our clinical reviewer verifies clinical-evidence claims before publication.

Pricing verification

Pricing is re-verified against vendor pricing pages quarterly (Q1/Q2/Q3/Q4). Every review carries a visible last-reviewed date. Pricing changes discovered between cycles are corrected within 72 hours.

Regulatory verification

FDA clearance status is verified against the FDA's AI/ML-enabled device list with an access date. Telemedicine state coverage (Talkiatry, Cerebral) is re-verified quarterly against vendor-published state lists, cross-referenced against state licensing boards.

What happens when sources conflict

When vendor-reported data conflicts with third-party data (for example, vendor claims 95% accuracy but a peer-reviewed study reports 82%), we cite both and explain the discrepancy. We do not pick the higher number.