Free tools Windows power users keep installed
One-click scans. No signup required.
Estimate AI productivity at the level where work actually changes: define the task or workflow, compare output per unit of human effort, and check quality alongside speed. A faster task is not automatically a more productive business. The result depends on what was measured, who used the tool, how it was introduced, and what happened to time saved.
What counts as an AI productivity gain?
Productivity describes the relationship between inputs and outputs. In a workplace study, the input might be a worker’s time or effort; the output might be completed tasks, resolved customer issues, or work that meets a quality standard. Studies often measure only one part of that relationship, so name the outcome rather than treating every reported percentage as the same kind of gain.
As an Amazon Associate I earn from qualifying purchases.
- Time: minutes or hours required to complete a defined task.
- Quantity: units completed, such as customer issues resolved.
- Quality: accuracy, usefulness, or another scored measure of the work.
- Effort and work patterns: hours worked, time spent on particular activities, or work outside regular hours.
- Downstream outcomes: customer satisfaction, earnings, employment, or firm-level results.
For example, reducing time per task while maintaining quality may increase output per hour if demand exists and workers use the released time for more work. If quality falls, extra review is required, or saved time becomes slack rather than additional output, the time reduction alone does not establish a matching increase in overall productivity.
How do you measure AI productivity gains?
Start with a precise comparison: the same type of work performed with AI access versus under a clearly defined alternative. Choose a unit of analysis before collecting results—individual task, workflow, team, firm, or economy. A finding about one unit does not automatically answer a question about another.
#1 Best Overall
- Specify the work and output. Define the task, what counts as a completed unit, and the quality threshold. If tasks differ in difficulty, account for that rather than comparing raw counts alone.
- Describe the intervention and comparison. Record the tool or model, how it is integrated into work, access rules, training, and whether workers actually use it. State what the comparison group does without the intervention.
- Measure speed and quality together. Track completion time and an appropriate quality measure. Where relevant, include quantity, errors, rework, recipient or customer outcomes, and human effort.
- Choose a measurement window. State how long the study runs and whether the result reflects initial adoption or more settled use. If access is later removed, measure whether any gains persist.
- Separate access from use. Report the effect of being offered or assigned access separately from results among people who use the tool, when the study design permits. Access alone is not evidence of adoption.
- Check who benefits. Break results out by relevant, study-defined groups such as experience or skill when the evidence supports it. An average can hide large differences.
- Follow saved time into the workflow. Measure whether it becomes additional output, higher-value work, reduced overtime, or unallocated time. Do not presume the destination.
A simple task-level rate is quality-adjusted completed output ÷ human labor time. The exact quality adjustment depends on the work: it might be a pass threshold, a blinded rating, or a separately reported error and rework rate. If a study only reports task time or only reports output quantity, report that measure as such rather than converting it into a broader productivity claim.
Use a comparison that can support the claim
Random assignment can support a causal estimate for the people, tasks, tool, and conditions in that study. A staggered workplace rollout can show what changed as deployment occurred, but its design and context are not interchangeable with a randomized online experiment. A simple before-and-after comparison may also reflect changes besides AI use. State the design and limit the conclusion to what it can establish.
How much does generative AI improve productivity?
There is no single percentage that applies across jobs. Selected experiments and deployments report sizable task-specific effects, but their outcomes, populations, and methods differ. The figures below are examples of what studies measured, not a forecast for a new workplace.
| Setting and study | What was measured | Reported result and scope |
|---|---|---|
| Professional writing; Noy and Zhang, summarized by Stanford SCALE (2023) | Time taken and output quality in occupation-specific, incentivized writing tasks | Among 453 college-educated professionals, with half randomly given ChatGPT access, average task time fell 40% and output quality rose 18%. This applies to those study tasks and participants, not knowledge work generally. |
| Customer support; Brynjolfsson, Li, and Raymond (NBER Working Paper 31161, 2023) | Customer issues resolved per hour | The workplace rollout study of 5,179 agents found a 14% average increase, including a 34% improvement for novice and low-skilled workers and minimal effects for experienced and highly skilled workers. |
| Customer support; NBER Digest summary (2023) | Issues resolved per hour, time per chat, and customer satisfaction | The Digest reports 13.8% more issues resolved per hour, about 9% less time per chat, and no significant change in customer satisfaction. It describes a gradual rollout involving roughly 5,000 agents. |
| Knowledge work across firms; Dillon, Jaffe, Immorlica, and Stanton (NBER Working Paper 33795, 2025) | Email time and work outside regular hours | Across 66 firms and 7,137 knowledge workers, the second half of the six-month experiment saw the 80% of treated workers who used the tool spend two fewer hours on email each week and reduce work outside regular hours. Researchers did not detect shifts in task quantity or composition from individual-level AI provision. |
| Product innovation; Dell’Acqua and co-authors (NBER Working Paper 33641, 2025; published online in Organization Science in June 2026) | Performance on real product innovation challenges | In a preregistered field experiment with 776 professionals, individuals with AI matched the performance of teams without AI on the study challenges. The finding concerns those tasks and measures, not every form of teamwork. |
| Business problem-solving; Cruces, Fernández Meijide, Galiani, Gálvez, and Lombardi (NBER Working Paper 34851, 2026) | Performance on an incentivized workplace-style task, with a later module completed without AI | In a randomized online experiment with 1,174 adults aged 25–45, AI raised performance for all participants and narrowed the education-based performance gap on the task. Follow-up results indicate that gains were not purely delegation: participants did not perform worse than controls without AI, and lower-education participants retained part of their gain. Follow-up performance was higher when intensive AI use was combined with sustained task effort. This is a working-paper finding in a bounded experimental setting. |
The OECD’s 2025 summary gives a 5% to over 25% range across selected experimental studies in customer support, software development, and consulting. That is a range across different settings and measures, not a pooled estimate or general forecast. Its review also emphasizes that outcomes depend on task and user experience, and identifies long-term business effects and workers’ understanding of AI limitations as gaps.
Why do results differ between workers and settings?
AI changes some tasks more readily than others, and a tool that helps one group may add little for another. The customer-support rollout found a much larger measured improvement among novice and lower-skilled agents than among experienced and highly skilled agents. The writing experiment, support deployment, cross-firm study, and product-innovation experiment also differ in work, participants, intervention, and outcome.
To compare two estimates responsibly, check each of these dimensions:
- Task and industry: writing, customer support, email, software development, consulting, and product innovation are not equivalent workloads.
- Participants and baseline experience: occupation, skill, and familiarity with the task can shape results.
- Access and actual use: being assigned a tool, having it available, and using it are distinct conditions.
- Study design: randomized experiments and observed rollouts support different kinds of inference.
- Outcome and quality checks: time, volume, quality, resolution rates, and work hours answer different questions.
- Duration and persistence: short-run results do not establish that a gain lasts or survives tool removal.
The OECD’s 2025 review identifies human–AI collaboration as important and notes that user experience and task affect effectiveness. This makes “AI versus no AI” an incomplete description unless the study also explains what people did with the tool and how the work was evaluated.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteDo time savings translate into more output?
Not necessarily. In the multi-firm experiment, users spent less time on email and reduced work outside regular hours, but researchers did not detect a shift in the quantity or composition of tasks from individual-level AI provision. The distinction is central: a time-saving result is evidence about time, not proof of more total work completed.
The ILO’s June 2026 review examined experiments, firm-level and platform studies, and worker and firm surveys in Australia, Denmark, Germany, Korea, Kuwait, the United Kingdom, and the United States. It concludes that gains are real but uneven and often unverified. In the evidence it reviewed, worker-reported time savings of a few percent of working hours had not yet translated into higher measured output, earnings, or employment. This describes that body of evidence, not every employer or worker.
Rank #4
- Stay present in every scenario: Every conversation is covered, in person, on calls, and online. 4 MEMS + 1 VPU microphones with AI beamforming capture every voice across the room. Smart Dual-Mode Recording switches automatically between phone calls and in-person. The free Plaud Desktop captures online meetings without a bot
- Walk out of every meeting with notes ready to act on: Plaud Intelligence transcribes in 112 languages with speaker labels and turns each recording into action items, decisions, and follow-ups, structured and ready to use. Choose from 10,000+ customizable templates tailored to your role and industry
- AI summary ready before you reach your desk: Auto Transfer moves each recording to the Plaud app automatically, and AutoFlow transcribes and summarizes so your notes are ready before you are back at your desk. Upgrade anytime to Pro (1,200 min/mo) or Unlimited
- Access your AI workspace anywhere: One connected workspace across Plaud Desktop, Plaud Web, and the Plaud mobile app, so your conversations and finished work follow you everywhere
- Your conversations stay private and yours: Compliant with ISO 27001, ISO 27701, SOC 2, HIPAA, GDPR, and EN 18031, with zero data used to train AI models. Trusted by 2.5M+ professionals, including legal, medical, and business professionals handling sensitive information
What can current evidence say about economy-wide effects?
Task-level improvements do not by themselves establish an economy-wide productivity increase. To make that step, evidence would need to show how widely tools are adopted, how workflows and organizations change, whether quality is maintained, and whether local time savings become additional valuable output over time. The OECD identifies long-term business effects as an open evidence gap; current findings support bounded claims about studied settings, not a definitive long-term aggregate forecast.
When reporting an estimate, put the study population, intervention, comparison, metric, and period beside the number. Then say whether the finding is causal for the study setting or an observed association, and keep any extension to other roles or firms clearly separate from the measured result.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
