A randomized trial with 133 patent attorneys at eleven US intellectual property firms found AI assistance sharply improved drafting, with the largest gains going to the least experienced lawyers. The advantage then behaved differently for juniors and seniors.
The working paper from the National Bureau of Economic Research, funded by Google, gave two-thirds of participants access to InFlow, an unreleased assistant from Google Labs. The rest worked unaided.
Blinded attorneys at an outside firm scored the output on enforceability, technical accuracy, strategic ambiguity, completeness and clarity. The assisted group beat the control group in drafting at both the 10-day and 90-day marks.
Gains concentrated among lawyers with fewer than seven years of experience, consistent with studies in writing, customer service and software development that show AI compressing skill gaps.
The redlining task is where the trial earns its keep. Every participant, AI or not, had to critique an existing patent draft without help. The assisted group scored better overall, but the edge narrowed with each round, and only senior attorneys held theirs.
The authors read that as a training problem rather than a tooling one. Junior lawyers learn judgment by struggling through first drafts, and handing that work to a model removes the reps.
