Correcting green patent labels reduces the count by 25.5% — the errors are systematic.

Governments and climate investors rely on official green patent tags to track clean technology. However, no one had checked whether these tags accurately label inventions across large patent databases. Researchers developed an approach called Error-as-Signal to find mistakes. They compared official tags on millions of patents against an independent model. When the two disagreed, they had two large language models evaluate if each invention directly mitigates or adapts to climate change. They discovered that mistakes are systematic rather than random noise. The official system falsely tagged many non-green inventions as green, while missing many truly green patents. Fixing both types of errors reduced the total count of green patents by 25.5%. Instead of inventors trying to game the system, these mistakes mostly happen because complex, unconventional patents are hard for patent offices to classify correctly. Highly complex inventions are often overlooked and left untagged.

Why it matters

Researchers and investors rely on green patent counts to track climate innovation. This study shows those counts are off by a quarter, and the errors follow predictable patterns — some green inventions are systematically overlooked while others are wrongly included. That means policy and investment decisions based on these numbers may be misdirected.

Questions answered