How Well Does AI Peer Review Work?

Cowen reports on an experiment that seeded 100 known errors into ten open-access psychology papers and tested several AI review tools against them, finding the best single system caught 71 percent of errors while combining every system together caught 93 percent. You get a concrete, sourced answer to how reliable AI-assisted error-checking currently is, and a reason to combine tools rather than trust any single one for a manuscript review.

Task

Research & literature

Time

6 min

Level

intermediate

Cost

Free — Nothing to pay. A free account at most.

Tools

None in particular

Published

2026-08-11

Voice

Tyler Cowen — Tests AI against the working tasks of an economist, including peer review and teaching material, and reports the results daily.