What workers say AI helps them do

The largest reported benefit is saving time at 90%. Focus, creativity and enjoyment follow at 85%, 84% and 83%. These measures describe workers' perceptions; they are not timed observations of work completed.

Citation's interpretation is that a useful evaluation needs more than one success criterion. A task can take less time while leaving the reader confused. Another can take the same time but produce a clearer explanation. The choice of outcome determines what a team should record.

Consider a research brief. Record the time spent gathering evidence, preparing a draft, checking claims and making revisions. If AI shortens drafting but creates extra verification work, both changes belong in the same assessment. Stopping the clock at the first response would miss part of the assignment.

Keep a small set of quality checks alongside the timing. Can a colleague locate the supporting evidence? Does the conclusion match the question? Are important disagreements represented fairly? These checks make the result interpretable without turning a subjective benefit into a productivity statistic.

A team can decide in advance what it wants to do with time it saves. It might compare an additional source, improve an explanation or reduce the amount of unfinished work carried into the next day. Naming that purpose helps connect a faster step with an outcome people actually value.

This is a proposed way to evaluate a workflow, not an intervention tested by the survey. The chart offers possible dimensions of value; a local trial would need its own observations and a clear definition of success.

The 2024 study surveyed 31,000 knowledge workers across 31 markets. These items concern AI users, whose subgroup size is not specified in the public summary. One person can report several benefits, so adding the bars would not count distinct people.

The original report provides the published figures and question wording. The chart and downloadable data use the same values; labels are shortened for readability.