A three-month randomized trial involving 133 patent lawyers found that an AI writing assistant improved immediate drafting quality but produced uneven learning after the tool was removed. The study covered 11 intellectual-property law firms and used independent legal experts to grade the work.
Two-thirds of lawyers received early access to a Google Labs patent-writing tool, while a control group received AI training but no access during the study. After 10 days, assisted drafts scored 0.34 standard deviations higher; after 90 days, the gap was 0.38. Junior lawyers also finished the first task 18 minutes faster than the control group’s 124-minute average.
Researchers then tested unassisted judgment by asking participants to correct a flawed patent without AI. The treatment group scored 0.32 standard deviations higher overall, but the improvement came from senior lawyers, whose gain was 0.45. Junior lawyers showed no average increase; their results spread toward both stronger and weaker scores.
The experiment concerns one profession, one tool and a limited sample, so it cannot establish a universal effect. It does show why measuring output while AI is available is not enough to determine whether workers are building durable expertise.