Google Research reports that, after a three-month randomized trial, senior patent lawyers with access to its AI drafting assistant scored higher than senior control-group lawyers on a later review task completed without AI. Lawyers with less than seven years of experience showed no discernible average gain on that task. The study used hypothetical patents, not client work.
A Google Research summary of the working paper describes 133 practicing lawyers at 11 U.S. intellectual-property firms. At each firm, two-thirds were assigned access to a then-unreleased Google Labs patent-writing assistant; the control group received AI training but could not use the tool during the study. The firms had regular, non-exclusive business with Google, the researchers note.
After 10 and 90 days, participants drafted patents from hypothetical inventor materials. Independent patent professionals, unaware of who had AI access, scored the drafts on quality. The access group scored 0.34 standard deviations above controls after 10 days and 0.38 after 90 days—roughly 10 and 11 percentile points in the authors’ comparison. The researchers say poor scores became less common and good scores more common, while the number of excellent scores did not rise.
At the 90-day mark, all participants also reviewed and corrected a flawed hypothetical patent without AI. The access group scored 0.32 standard deviations higher than controls on this task. Lawyers with at least seven years of experience scored 0.45 standard deviations higher; those with less experience showed no discernible average gain. Among junior lawyers, the researchers report more very low and good scores, fewer middling scores, and no increase in excellent scores.
The results separate better work with an assistant from performance after it is removed. They cover one patent-writing tool, test assignments using hypothetical inventions, and a three-month period; they do not show how client work or expertise changes over years. The researchers’ summary links the full working paper through the National Bureau of Economic Research.
Sources
- Google Research: the study design, results and limitations
- NBER Working Paper 35720
- Wikimedia Commons image record and rights
Illustrative image: 1895 U.S. patent drawing via Wikimedia Commons; public domain. It is not part of the study.