What is in-place AI worth? The time math of copy-paste and app switching.
Copying text into an AI chat and pasting the answer back takes seconds. Seconds are easy to wave away, and easy to inflate into "hours saved". Here is the arithmetic in the open, with every number either sourced or marked as our assumption, so you can redo it with your own.
8 min read · · BeqSpark team
In short
- Measured step times put one copy-paste trip to an AI chat at about 5.8 seconds longer than doing the same thing in place, and about 10 seconds when the answer goes back into your text.
- Add the moment it takes to find your place after switching windows, and our middle case saves 12 seconds a use: about 11 hours a year at 15 uses a working day.
- The range is wide: from 1.8 hours a year (light use, steps only) to about 59 hours (heavy use, with an assumed detour now and then).
- Multiply by what an hour of your time costs and compare with a plan's price, but read the result as an upper bound: a few seconds saved don't always become useful work.
Why put numbers on it
AI chat is mostly used on text people already have. In a 2025 working paper on how people use ChatGPT, economists at OpenAI and Harvard found that writing was the most common use at work, about 40% of work-related messages in June 2025, and that about two-thirds of writing requests asked ChatGPT to change text the user provided, by editing, critiquing or translating it, rather than to write something new. (NBER)
Each of those requests starts with getting the text into the chat, and often ends with getting the answer back out. Tools that work on the text where it is, in place, skip that trip. It's tempting for anyone who makes such a tool, us included, to say it saves "hours". This post works out how many, with published research where something has been measured and clearly marked assumptions where it hasn't.
The result: low, middle and high
Three cases, from light use counting only the steps you skip, to heavy use counting everything we could reasonably add. How we calculated this explains every figure.
| Case | Seconds saved per use | Uses a working day | Hours a year | Working days a year |
|---|---|---|---|---|
| Low | 5.8 | 5 | 1.8 | 0.2 |
| Middle | 12 | 15 | 11 | 1.4 |
| High | 32.3 | 30 | 59 | 7.4 |
A working year here is 220 days and a working day 8 hours, both our assumptions.
What those hours are worth depends on what an hour of work costs. Two published averages: in 2025, labour costs in the EU averaged €34.9 an hour (Eurostat), and in June 2026, US employers' costs for civilian workers averaged $49.46 an hour worked (US Bureau of Labor Statistics). Both are what employers pay in total, not take-home pay.
| Case | Hours a year | At €34.9 an hour (EU, 2025) | At $49.46 an hour (US, June 2026) |
|---|---|---|---|
| Low | 1.8 | €62 | $88 |
| Middle | 11 | €384 | $544 |
| High | 59 | €2,067 | $2,929 |
These are values of time, not cash savings: see what this doesn't count.
How we calculated this
Time saved a year = seconds saved per use × uses a working day × working days a year. The seconds per use come from three parts: the steps you skip (measured), finding your place after a switch (derived from a published study) and detours (assumed). How often you use it, and on how many days, are assumptions too.
1. The steps you skip (measured)
To time the steps, we used the keystroke-level model, a method from 1980 that predicts how long a skilled user takes to do a task by adding up the times of its steps (Card, Moran and Newell). In 2009, Jeff Sauro measured the times of common combined steps with 26 users (Sauro, HCII 2009): a keyboard cut and paste, including highlighting the text, took 4.51 seconds; clicking a link or a button, including the wait for the screen to respond, 3.73 seconds; and each typed character 0.23 seconds.
| Step | Copy-paste into a chat | In place |
|---|---|---|
| Select the text (for the chat, also copy it and paste it there) | 4.51 | 4.51 |
| Go to the chat, or open the tool | 3.73 | 3.73 |
| Type a request of 25 characters, or pick an action | 5.75 | 3.73 |
| Go back to your work | 3.73 | — |
| Total | 17.72 | 11.97 |
So working in place skips 5.75 seconds per use. Two of our choices make that figure smaller, not larger: we give the in-place path the whole cut-and-paste time just for selecting the text, and we leave out keyboard shortcuts, which would shorten both paths.
When the answer goes back into your text, the loop adds copying the answer (a click, 3.73 seconds) and selecting your text again and pasting over it (4.51 seconds). In place, one click on Replace (3.73 seconds) does it: another 4.51 seconds saved. The low case assumes no use writes back, the middle case half of them, the high case all of them.
2. Finding your place after a switch (derived)
In a study published in Harvard Business Review in 2022, workers on 20 teams switched between apps and websites about 1,200 times a day, and spent just under four hours a week getting reoriented after switching (Harvard Business Review). Dividing one by the other, 240 minutes over about 6,000 switches in a five-day week, gives at most 2.4 seconds per switch; we use 2. A trip to a chat and back is 2 switches, so 4 seconds. That's our arithmetic on the study's averages, not a figure its authors give per switch, and it may overlap with the measured step times, so the low case leaves it out.
3. Detours (assumed)
Sometimes a trip to another window turns into something longer: a new message waiting in the chat app, or another tab that catches your eye. Research on interruptions shows that getting back to work can cost far more than the switch itself (our summary of the research), but we found no measurement of how often a trip you start yourself becomes a detour. So the high case simply assumes that one trip in 10 turns into a 3-minute detour: 18 seconds per use, on average. If that's not you, use the middle case.
4. How often, and for how many days (assumed)
How often you'd use it matters most, and only you know that. For scale: in July 2025, about 700 million people sent 18 billion ChatGPT messages a week (NBER). That averages about 26 messages a week per weekly user, or about 5 a day over a five-day week, and in June 2025 nearly three-quarters of all messages weren't about work. We use 5 uses a working day for the low case, 15 for the middle and 30 for the high. Count translations and look-ups too: they're the same trip.
| Input | Value | Kind | Source |
|---|---|---|---|
| Cut and paste, including selecting | 4.51 s | Measured | Sauro, 2009 |
| Click a link or a button | 3.73 s | Measured | Sauro, 2009 |
| Each typed character | 0.23 s | Measured | Sauro, 2009 |
| Length of a typed request | 25 characters | Assumed | "Explain this in plain words" is 27 |
| Finding your place, per switch | 2 s | Derived | Harvard Business Review, 2022, our division |
| Uses that put the answer back | None, half or all | Assumed | Low, middle and high case |
| Detours | 1 trip in 10, 3 minutes | Assumed | High case only |
| Uses a working day | 5, 15, 30 | Assumed | For scale: ChatGPT's average use, NBER, 2025 |
| Working days a year | 220 | Assumed | A five-day week, minus about eight weeks of holidays and leave |
| Hours in a working day | 8 | Assumed | Only to turn hours into days |
| What an hour costs | Your choice | Examples | Eurostat, 2025; US Bureau of Labor Statistics, June 2026 |
Work out your own numbers
Start from a case, then change any number. To compare with a price, enter what an hour of your time costs and a plan's price for a year from our pricing page: the calculator works out how many uses a day it takes for the time saved to be worth the price.
What this doesn't count
- The value of the AI itself. In a nationally representative US survey in November 2024, workers who used generative AI reported saving 5.4% of their work hours on average (Federal Reserve Bank of St. Louis). That's the gain from using AI at all. This post counts only the trip around it, which costs the same whether the answer is good or bad.
- Waiting for the answer. It takes as long in a chat as in place, so it's left out.
- Smaller costs we couldn't measure, like fixing the formatting of a pasted answer, or telling the chat where the text came from. Leaving them out keeps the estimate low.
- Whether seconds become money. Twelve seconds saved don't turn into twelve seconds of paid work. The money figures are the value of the time at a given hourly cost: an upper bound, not cash in your pocket.
- The price of the tool. That's what the comparison with a plan's price is for.
So, is it worth it?
On time alone, it depends almost entirely on how often you'd use it. In the low case, a year adds up to 1.8 hours. In the middle case, 11 hours, about 1.4 working days. In the high case, 59 hours, about 7.4 working days, though 56% of that rests on the detour assumption.
In our view, the seconds are the smaller part. What you keep by staying in place is your train of thought, and the research on context switching suggests that's where the real cost of switching lies. The seconds are simply the part that can be counted.
Sources
- NBER Working Paper 34255: How People Use ChatGPT (Chatterji, Cunningham, Deming, Hitzig, Ong, Shan and Wadman, 2025)
- Eurostat: EU hourly labour costs ranged from €12 to €57 in 2025 (31 March 2026)
- US Bureau of Labor Statistics: Employer Costs for Employee Compensation, June 2026 (9 September 2026)
- Communications of the ACM: The keystroke-level model for user performance time with interactive systems (Card, Moran and Newell, 1980)
- Human-Computer Interaction International (HCII 2009): Estimating Productivity: Composite Operators for Keystroke Level Modeling (Sauro, 2009)
- Harvard Business Review: How Much Time and Energy Do We Waste Toggling Between Applications? (Murty, Dadlani and Das, 2022)
- Federal Reserve Bank of St. Louis: The Impact of Generative AI on Work Productivity (Bick, Blandin and Deming, 2025)
Keep reading
- The real cost of context switching, and what helps
People switch apps about 1,200 times a day. What research on context switching really shows, which famous numbers hold up, and what you can do about it.
- How to use AI in your Windows apps without copy-pasting
Use AI on text in Word, PowerPoint, your browser and other Windows apps without copy-pasting: what Windows has built in, how to choose, and tips that help.