marlow checks the articles and documents you read — and the ones you author, before they leave the building. Whether the quoted experts exist, whether the cited studies are real, whether the numbers trace back to a real source. It shows you what it found, what it couldn't check, and why. The conclusion stays yours.
And it isn't a few headline cases. An audit of 2.5 million biomedical papers found fabricated references have climbed roughly twelvefold in two years — reaching one paper in every 277 by early 2026.
Receipts, naturally — The Guardian · Fortune · International Accounting Bulletin · Semafor · STAT / The Lancet
Every check runs against a registry of authoritative sources — 15 live today, with 34 more mapped and licence-assessed, with industry‑specific source packs on the roadmap.
So who checks them? marlow does.
Scan a report in your browser today. Open the Documents tab in the marlow panel and drop in a PDF or Word document — currently in beta — and every checkable claim gets the same treatment as a web page: checked against external sources, each landing in a named, auditable state. A report you commissioned, a supplier's white paper, research you're about to act on — checked before you rely on it, whether AI wrote it or not.
The check has to be independent of whatever produced the document — the system that writes a claim is the wrong system to certify it, the same reason the auditor is kept separate from the accounts. As more documents are drafted by machine, an independent, claim-level evidence trail is what lets a team show its work held up before anyone relied on it.
A pre-send verification API is in development. Submit a document's text and get back the same per-claim evidence cards as a structured record: what was checked, against which sources, and which of seven honest states each claim landed in. Not a score you have to trust — receipts you can audit. We're onboarding pilot partners first.