Why every press release should pass through a Claim Safety Gate
FTC, ASA, and platform policies all converge on the same rule: don't claim what you can't prove. Here is how we encode it.
Press releases were once a low-risk format. Today they are scanned by AI distribution partners, surfaced inside answer engines, and cited as evidence in dispute filings. A single unsupportable claim — “world's leading platform for X” — propagates out to a hundred news aggregators and stays in the model's training set for years. The cost of the bad sentence has gone up.
The convergence
The FTC's endorsement guides, the UK ASA, EU UCPD, and every major search / AI-platform publisher policy now require the same thing: claims must be substantiated at the time they are made, with evidence the brand can produce on request. Penalties range from required corrections (mild) to platform de-listing (severe).
What the gate enforces
AuthoritySignal's Claim Safety Gate is a deterministic, rule-based classifier that runs every claim through three checks before a draft is allowed to render:
- Superlative without proof. “Best”, “largest”, “#1”, “most trusted” trigger a block unless the operator attaches an evidence URL whose content the gate verifies references the claim.
- Quantitative claim without source. Numbers — counts, percentages, year-over-year — require either a methodology link or an audited dataset reference.
- Implied endorsement. “Trusted by leading X”, “official partner of Y”, “recommended by Z” require named public evidence, not aggregate statistics.
Why deterministic, not probabilistic
We deliberately do not use a model to grade claims. The ruleset is a code-reviewable artifact in lib/claim-safety/rules.ts — operators can extend it for their industry, lawyers can audit it, and CI verifies that the rules behave the same way over time. A model would drift, fail silently when retrained, and be unjustifiable in a regulatory dispute. Claim safety is not the place for soft signals.
What it costs you
About four extra minutes per draft, the first time. Operators internalise the rules quickly and start writing claims that pass the gate by default. The output is shorter, more specific, and more citable — which is the actual goal of the press release.