How do I measure whether using AI is actually improving my productivity at work?

Measure at the task level, not the vibe level. Pick two or three recurring tasks where you use AI most, for example the weekly report, client emails, and meeting follow-ups. Write down the baseline: roughly how long each took before, and how often the output needed rework. After a month of AI-assisted work, compare three things: time per task, number of revision rounds, and whether quality held, judged by the reactions you can observe, such as fewer clarifying questions or fewer edits from your manager.

Then ask the question most people skip: where did the freed time go? Saving four hours a week is only a productivity gain if those hours moved into something more valuable, deeper work, more clients, earlier finishes. If you cannot name where the time went, it likely leaked into more email, more polishing, or more exploring the tools themselves. Productivity is the destination of the saved time, not the saving alone.

Watch two honest counter-metrics. First, rework created downstream: an AI-drafted document that colleagues had to fix or that caused a misunderstanding is negative productivity that never shows up in your own timing. Ask the people who receive your work whether it got better, worse, or stayed the same. Second, task inflation: AI makes producing so easy that people quietly produce more documents nobody needed. Fewer, better outputs beat more outputs.

Keep the measurement lightweight or it will not survive: a note per task per week is enough, and a calendar review at month end tells you where the hours moved. If after a month the numbers are flat, the problem is usually the workflow, not the tool: the tasks chosen were judgment-heavy, or the prompts were vague enough that editing ate the savings. Adjust the task list and re-run the month before concluding AI does not help your role.