Pangram AI detector review

Pangram Labs, founded as Checkfor.ai by Max Spero and Bradley Emi, has become the detector researchers cite when they want the best available result. It is also the one humanizer vendors least like to be tested against.

The accuracy evidence

• Research from the University of Chicago and University of Maryland measured a false-positive rate of about 1 in 10,000 documents.
• It tied for first place in the COLING 2025 detection benchmark.
• A June 2026 peer-reviewed study of 160 academic papers found it was the only tool tested that reliably caught both fully AI-written and humanized text.

That last result matters most for this site's readers. Most detectors struggle once text has been run through a humanizer; Pangram is the main exception on current evidence, which is one more reason synonym-swapping tools are a poor bet.

Pricing

The limit that still applies

Princeton's Arvind Narayanan made the point that sticks: a 1-in-10,000 false-positive rate sounds negligible until you multiply it. A student who submits hundreds of pieces of work over four years, at a university running every one through a detector, gives the detector a lot of chances. Across a whole student body, a small per-document rate still produces real students falsely accused. Pangram being the best detector does not make any detector a verdict.

Who it is right for

• Editors and newsrooms screening submissions, where a flag starts a conversation rather than a penalty.
• Institutions that want the strongest available signal and have a process for appeals.
• Writers who want to know whether their own careful prose is at risk before someone else checks it.

What Pangram does not give you is the reason. A score tells you that a document looks machine-written, not which sentences carry the signal or what to change. That is the part our free analyser covers: every sentence marked, the signal named, and a rewrite toward your own style.