What the research shows AI delivers, and which popular claims do not hold up.

The best free resources

• Generative AI at Work
  Solid proof: an AI assistant made 5,179 support agents 14% faster, newcomers 34%.
  Erik Brynjolfsson, Danielle Li, Lindsey Raymond, NBER · 2025

• Still Waters, Rapid Currents (first titled "Large Language Models, Small Labor Market Effects")
  Counter-evidence: across 25,000 Danish workers, chatbots saved about 3% of time, with little pay effect.
  Anders Humlum, Emilie Vestergaard, NBER · updated 2026

• The GenAI Divide: State of AI in Business 2025 (the "95% of pilots fail" report)
  The source of the "95% fail" headline; it covers big-company pilots, not small firms.
  MIT NANDA · Jul 2025

• MIT's 95% AI failure rate is wrong
  Walks through the "95%" study's small sample and its broad definition of failure.
  Arnon Shimoni · Aug 2025

Curated by effectiveAI. Every link checked November 2026. Know a better free resource? Tell us.