Evidence · searched to 2 Nov 2026
What works: AI at work, rated by the evidence
Does AI make people faster or better at their work? Here is what published studies show, use case by use case, and how sure we can be.
Each rating comes from the studies by fixed rules: a verdict, the size of the effect, the strength of the evidence from 0 to 5 and the cost. Where studies are few or weak, the rating says so.
Pilot rating under our v0.3 rules. The rules are published on our method page but not yet independently reviewed, so ratings may change as the method is checked and new studies arrive. Our method. New finding: most used vs best studied.
Rated use cases, by what the evidence says
Show
Evidence points to a gain 4
Strength 2 or more, with studies pointing the same way. Read who it works for, and the risks.
Scientific researchDoes AI improve research papers and peer review?2 of 3 studies found a clear gain in quality ratings.May help7 studies · cost $Research and analysisDoes AI help with consulting and analysis work?Across 2 studies, time per task changed by −25.1% to −16.4%.May work for tasks that sit inside what the AI does well2 studies · cost $$Legal workDoes AI help with legal work?3 of 4 studies found a clear gain in time per task.May help4 studies · cost ?WritingDoes AI help professionals write faster and better?3 of 3 studies found a clear gain in quality ratings.May help3 studies · cost $$
Nothing in this group for that job.
No clear effect so far 3
Enough studies to rate, but they do not show a clear gain or loss on the main outcome.
Clinical workDo AI scribes reduce doctors' paperwork?3 of 6 studies found a clear gain in time on notes.No clear effect so far8 studies · cost ?Research and analysisDoes AI help people analyse data more accurately?3 of 7 studies found a clear gain in accuracy and 1 a clear loss.Mixed results7 studies · cost $$TeachingDo AI tutors help students learn?2 of 6 studies found a clear gain in test scores.No clear effect so far6 studies · cost ?
Nothing in this group for that job.
Early signals 14
Strength 1: a few studies, too weak or too few for a verdict. The direction is shown, not a promise.
Clinical workDo AI-drafted replies save clinicians time on patient messages?Across 4 studies, time on each message changed by −64.2% to +8.6%.Not enough evidence yet5 studies · cost ?TeachingDoes AI feedback on students' work help them learn?0 of 2 studies found a clear gain in learning, tested without AI.Not enough evidence yet3 studies · cost $Clinical workDoes an AI chatbot help doctors diagnose and plan treatment?4 of 5 studies found a clear gain in reasoning scores.Not enough evidence yet5 studies · cost $$Software engineeringDoes AI make code review faster or better?1 of 4 studies found a clear gain in review time and 2 a clear loss.Not enough evidence yet4 studies · cost $$Research and analysisDoes AI make financial analysis better?2 of 3 studies found a clear gain in quality and accuracy and 1 a clear loss.Not enough evidence yet3 studies · cost ?Scientific researchDoes AI speed up systematic reviews without more errors?0 of 2 studies found a clear gain in time per study.Not enough evidence yet2 studies · cost $TeachingDoes AI guidance for tutors and teaching assistants help students?1 of 2 studies found a clear gain in student learning.Not enough evidence yet2 studies · cost $Hiring and recruitmentDo AI-written job posts and applications lead to more hires?0 of 2 studies found a clear gain in callbacks or hires.Not enough evidence yet3 studies · cost $Software engineeringDoes AI make software developers faster?Across 3 studies, time per task changed by −42.4% to +19%.Not enough evidence yet6 studies · cost $$TeachingDoes AI save teachers time on lesson planning?Across 2 studies, preparation time changed by −31% to −24%.Not enough evidence yet3 studies · cost $Software engineeringDoes AI help people write requirements and designs?0 of 3 studies found a clear gain in expert-rated quality.Not enough evidence yet3 studies · cost $$Creative and design workDoes AI help people make better slides?0 of 2 studies found a clear gain in quality of slides.Not enough evidence yet2 studies · cost $$Research and analysisDoes AI help people understand long documents?2 of 5 studies found a clear gain in quality of understanding.Not enough evidence yet5 studies · cost $$TranslationDoes AI help professional translators?Across 2 studies, time per task changed by −31.1% to +3.5%.Not enough evidence yet4 studies · cost $
Nothing in this group for that job.
Too few studies to rate yet 10
Fewer than two strong-enough studies, or problems with the ones there are. Each page lists what exists.
Does AI help IT and security staff resolve problems faster?IT and security support · 3 studies countedDoes AI make creative and design work better?Creative and design work · 3 studies countedDoes AI make better hiring decisions?Hiring and recruitment · 3 studies countedDoes AI-written marketing content sell more?Marketing · 4 studies countedDoes AI business advice help small businesses?Running a small business · 1 study countedDoes AI help customer service agents?Customer service · 2 studies countedDoes AI help staff get rules and policies right?Office and admin work · 3 studies countedDoes AI help people find information for their work?Research and analysis · 1 study countedDoes AI save time on notes and routine records?Office and admin work · 3 studies countedDo AI assistants save time on everyday office work?Office and admin work · 4 studies counted
Nothing in this group for that job.
The evidence map
How studies are graded →Each cell: how many studies measured that outcome, and their grades (A strongest, D weakest). A dash means no study we found has measured it. "Nobody has tested this" is a finding too.
Download the evidence map
Every study in the map, one row per outcome: 145 studies, 460 rows, with the grade, whether it counts, the effect as the paper reports it and where in the paper it appears. Free to reuse under CC BY 4.0; please cite “AI Adoption (2026), Adopt evidence map, pilot of 2 Nov 2026, aiadoption.org”.
When AI made things worse
Some studies found AI slowed people down or made results worse, often on tasks just outside what the AI does well. What hasn't worked →
How up to date is this?
Evidence searched to 2 Nov 2026: 114 studies across 31 use cases. We add new studies as they are published and log every rating change on the method page.
Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. No company pays for or has any say in a rating.