Pangram Labs

Category: AI Writing

AI detection with a published false-positive rate of 1 in 10,000.

Pangram does AI content detection, and the thing separating it from the many free detectors is published research and a measured false-positive rate rather than a confidence score with no working shown. Its accuracy has been independently verified by researchers at the University of Chicago and the University of Maryland, and it publishes a model card benchmarking against 26 frontier models. Submit text and you get a verdict with segment-level highlighting showing which passages are AI-generated and, separately, which look AI-assisted, which matters because mixed human-written and AI-edited content is the normal case rather than the exception. It detects text across more than 20 languages, holds up against text run through humanising tools, includes a similarity and plagiarism checker in the same pass, and now detects AI-generated images too. Access is via a web dashboard, a Chrome extension that works inside Google Docs and Gmail, and an API for volume. Twenty free checks a day on the free plan.

Our verdict

The strongest option if you need detection you can actually stand behind. Most AI detectors are unreliable enough to be dangerous, and accusing a freelancer wrongly is a worse outcome than missing a case, so a documented 1-in-10,000 false-positive rate and third-party verification from Chicago and Maryland is the whole point here. Segment-level highlighting is the practical feature, since it lets you have a conversation about specific passages rather than a blanket accusation, and separating fully AI-generated from AI-assisted text reflects how people actually work. The API makes it usable as a gate in a publishing workflow rather than a manual check. Worth being clear-eyed: no detector is infallible, so treat a result as evidence for a conversation and not a verdict, and detection is an arms race, so accuracy needs rechecking over time. Also remember Google does not penalise AI content for being AI, it penalises unhelpful content, so use this for quality control rather than SEO panic.

Best for: Editors, content leads and agencies who commission writing from others and need a defensible answer on whether a deliverable was machine-generated.

Pros

  • Published 1-in-10,000 false-positive rate, verified by Chicago and Maryland researchers
  • Separates fully AI-generated text from AI-assisted, with segment-level highlighting
  • Holds up against text run through humanising tools
  • Plagiarism and similarity checking in the same pass, plus AI image detection
  • Chrome extension works inside Google Docs and Gmail, and there is an API for volume
  • Detects across more than 20 languages, and 20 free checks a day

Cons

  • No detector is infallible, treat a result as evidence for a conversation
  • Detection accuracy is an arms race, worth rechecking over time
  • Narrow by design, it does one job rather than several
  • Remember Google penalises unhelpful content, not AI content as such

Pricing: Free, 20 checks a day, then paid plans and API (Paid)

Rating: 4/5