Merlise achieves breakthrough results on SciFact-Open
Merlise
Editorial · Teardown

The byline was invented before the story was

20264 min read

WIRED said it retracted a freelance article after concluding that the piece was an AI fabrication and that the contributor's identity could not be adequately verified. This is a different kind of failure from a single unsupported quote. The problem sits underneath the sentences, in whether the reporting behind them ever happened at all.

The sentences in the piece could be checked. The person who supposedly wrote it could not be.

The deeper layer beneath fact checking

Traditional fact checking usually starts with a finished draft and works through it claim by claim. That approach assumes the reporting chain behind the draft is sound. When the contributor's identity is itself unverifiable, checking individual sentences may not be enough, because the entire premise of the piece, that a real person did real reporting, can be false.

A contributor can invent a pitch and a bio that both sound plausible, then fake the quotes to match, and the result can still read exactly like ordinary journalism. Surface fluency is not evidence of anything.

Checking provenance before checking sentences

For exactly this situation, intake level review outweighs sentence level fact checking. Identity checks. Source contact evidence. Interview records. Editorial signoff for a first time contributor, before a draft is treated as ready. Not all of it can be automated. The questions can still be structured, though, and the missing pieces made visible instead of assumed away.

The earlier a provenance gap is caught, the cheaper it is to fix. Catching it at pitch intake costs a conversation. Catching it after publication costs a retraction, a correction to the archive, and whatever trust the piece had already spent.

The evidence ledger

The same claim by claim view the product shows on a live document, built from this case.

VERIFICATION LEDGERContributor and story provenance
Contributor identity confirmed through independent contact before commissioning.Attribution
11%
Identity could not be adequately verified; contributor may not be a real, contactable person.Refuted
Publication's own account
Sourcing for the central claims in the piece traces to a named interview.Attribution
8%
No interview record located.Unsupported
Publication's own account

How it resolves

Two 2026 newsroom incidents, compared
MeasureArs Technica / ShambaughWIRED / freelancer
What failedA quote attributed to a real, named personThe contributor's identity itself
Caught byThe subject, publicly, within hoursThe publication's own review
Time to retractionUnder 2 hoursNot disclosed

The resolving record

WIRED's own account of the incident is the primary resolving record, describing both the fabricated content and the failure to verify the contributor.

WIRED's post incident statement