How Much Time Will AI Save You at Work? Calculator for 30 Jobs

👁 115 reads

With regular AI use in typical tasks, you save between about 2.5 hours per week (physician, documentation only) to about 10 hours (copywriter, social media) on a 40-hour workweek. Large surveys of people already using AI show less: 2.8% of work time in Denmark and 5.4% in the US. The calculator below counts your scenario task by task, with a source for every number.

[pw_kalk_zawody]

How We Calculate Time Savings

For each of 30 professions, we broke down 5-8 typical tasks and assigned them a share of the work week. The share is our estimate from typical job descriptions, not a research result, so you can adjust it in the calculator. The formula:

• task savings = hours per week × task share × AI savings percentage × frequency of use,
• month = week × 52/12, or 4.33 weeks,
• dollar value = hours × your hourly rate; we count a year as 46 work weeks.

We take the savings percentage from studies where someone measured work time with and without AI. We matched each task to the closest type of work:

• writing text - Noy and Zhang, Science 2023: 40% faster for short professional texts; we use 25-35% because real texts are longer and need checking,
• email - Copilot experiment in 66 companies (NBER w33795): about 2 hours per week less in the inbox, 12-17% of time in Outlook; we use 15%,
• analysis and reports - BCG consultants with GPT-4 worked 25.1% faster (HBS 24-013); we use 25%,
• programming - from 55.8% faster on a simple task (Peng et al.) to 19% slower for experienced programmers (METR 2025); we use 20% for new code,
• customer service - 14% more cases per hour (Brynjolfsson et al.),
• legal writing - 11.8-32.1% faster (Choi, Monahan, Schwarcz),
• lesson preparation - 31% faster in a randomized study of 259 teachers (EEF),
• medical documentation - 9.5% faster with an assistant that listens to the visit (NEJM AI 2025),
• translation - 10%, explained below.

Where there’s no research (photo selection, captions, home staging, automation), we write “estimate” explicitly and explain how we calculated it. Tasks requiring physical presence, such as teaching a class, examining a patient, or supervising a construction site, get 0%.

Results for 30 Professions at 40 Hours

The table assumes you check all tasks and use AI for each one. This is the upper limit of a realistic scenario, not a market average.

We compared results with surveys where profession-specific data exists. Teachers come out at 6.1 h per week in our calculator, while US teachers using AI weekly report 5.9 h (Gallup 2025). Lawyers come out at 6 h, while lawyers and accountants in the Thomson Reuters survey expect 5 h. UK civil servants in a Copilot experiment reported 26 minutes per day, or about 2.2 h per week (GDS 2025) - less than our result, because not everyone used the tool for every task.

Why Surveys Show Less

• St. Louis Fed: AI users saved 5.4% of work hours, and counting all workers 1.4% (St. Louis Fed).
• Denmark, 25,000 people in 11 professions: average 2.8% of time and no impact on wages (NBER w33777).
• McKinsey estimated in 2023 that it’s technically possible to automate activities taking up 60-70% of workers’ time. That’s the technology’s potential, not measured savings (McKinsey).

The difference comes from frequency: surveys count people who open a chat once a week. That’s why the calculator has a “How often will you use AI” toggle. The second issue is learning: in the EEF study, teachers had 5 weeks to get comfortable with ChatGPT before time measurement began.

Where AI Slows You Down

Three results lowered our numbers:

• METR: experienced programmers on their own repositories took 19% longer with AI, though they themselves believed AI sped them up.
• English-Polish translation: in data from 90 million words (Terribile 2023), the average suggested 18% savings, but after removing outliers, post-editing machine translation was slower than translating from scratch in 89% of assignments. That’s why we give translators only 10%.
• BCG: on a task outside the model’s capability, consultants with AI had 19 percentage points fewer correct answers than the group without AI.

If you want to check which activities AI can handle at all, without counting hours, we also have a task calculator and a profession table with AI limits.

Frequently Asked Questions

How much time does AI really save at work?

On average, not much: 2.8-5.4% of hours in large surveys, or 1-2 hours per full-time job. With regular use in text-based tasks, it comes out to 5-10 hours per week.

Does saved time mean higher earnings?

Not automatically. In the Danish study, time savings didn’t translate into either wages or hours worked. The dollar amount in the calculator is the value of time at your hourly rate.

Can I paste client data into ChatGPT?

Not into free accounts. Personal, medical, financial data and confidential information only in tools with a data processing agreement. We show warnings for each profession under the result, and write more about this in our article on data security.

Where do the task shares in the week come from?

That’s our estimate from typical job descriptions. You can change each share in the calculator, and the “Copy link to result” button saves your settings in the URL.

Does the calculator send my data anywhere?

No. It calculates in the browser, and keeps settings in the part of the URL after the # symbol, which the browser doesn’t send to the server.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources

We verified all numbers on October 3, 2026.

• Noy and Zhang, Science 2023 - experiment with 453 people: short professional texts written with ChatGPT 40% faster, quality 18% higher.
• Dillon, Jaffe, Immorlica, Stanton, NBER w33795 (2025) - Copilot in 66 companies, 7,137 people: active users spent about 2 hours per week less on email (in the 11.2025 version about 12-17% of time in Outlook).
• Microsoft Work Trend Index, special report 11.2023 - Microsoft study of 147 people: search, meeting summary and writing together 29% faster; catching up on a 35-minute meeting 11 instead of 43 minutes.
• Dell’Acqua et al., HBS 24-013 (2023) - 758 BCG consultants with GPT-4: tasks within AI’s reach 25.1% faster and 12.2% more tasks; task outside reach - 19 percentage points fewer correct answers.
• Dell’Acqua et al., NBER w33641 (2025) - 776 people from Procter & Gamble working on product: with AI 16.4% shorter (individuals) and 12.7% shorter (teams).
• Peng et al., arXiv 2302.06590 (2023) - GitHub Copilot: simple, closed task (HTTP server in JavaScript) 55.8% faster.
• METR, Becker et al. (2025) - experienced open source programmers on their own repositories: with AI tasks took 19% longer, though they themselves rated that AI sped them up.
• Brynjolfsson, Li, Raymond, QJE 2025 (NBER w31161) - 5,179 customer service consultants: +14% resolved cases per hour, +34% for new employees, almost no change for the best.
• Choi, Monahan, Schwarcz, Minnesota Law Review 2024 - GPT-4 in 4 legal tasks: time shorter by 11.8% (memo) to 32.1% (contract), quality almost unchanged.
• Terribile, Translation Spaces 2023 - 90 million words, 879 translators: post-editing machine translation on average faster, but for English-Polish after removing outliers it was slower in 89% of assignments.
• EEF and NFER, Teacher Choices (12.2024) - 259 teachers: lesson planning 56.2 instead of 81.5 minutes per week (-31%), no drop in material quality.
• Gallup and Walton Family Foundation (06.2025) - survey of 2,200+ US teachers: those using AI weekly report 5.9 h savings per week (self-reported, not measured).
• Lukac et al., NEJM AI 2025 - 238 physicians, 72,000 visits: Nabla assistant shortened note writing by 9.5% compared to control, difference for DAX not statistically significant.
• Choi and Xie, Journal of Accounting Research 2026 - 277 accountants and data from 79 firms: about 9% of time shifted from data entry to client work, month-end close 7.5 days faster; excessive trust in incorrect suggestions occurred.
• Thomson Reuters, Future of Professionals 2025 - 2,275 lawyers, accountants and auditors expect about 5 h savings per week within a year (respondent forecast).
• GDS, M365 Copilot experiment (06.2025) - 20,000 civil servants in 12 UK institutions: average 26 minutes per day (participant reports).
• Bick, Blandin, Deming, St. Louis Fed (02.2025) - AI users in the US saved 5.4% of work hours (counting all workers: 1.4%).
• Humlum and Vestergaard, NBER w33777 (2025) - 25,000 people in 11 professions in Denmark: average 2.8% of work time, no impact on wages and hours.
• McKinsey Global Institute, The economic potential of generative AI (06.2023) - potential to automate activities taking up 60-70% of work time.

Kinetyra