A note before the line. This holds across ChatGPT, Claude, Gemini and Copilot, because it is about the shape of the task rather than the quality of any tool. The shape of tasks does not update with the models.
You needed to send a two-line email, and somewhere along the way you decided to be efficient about it. So you opened ChatGPT and explained the situation, waited, read a draft that was nearly right but not quite, explained what was off, waited again, and trimmed the result.
Roughly eighteen minutes after starting, you pressed send on a job your own fingers would have finished in four. Then you felt vaguely foolish, and possibly a little resentful of everyone who insists this technology saves them hours.
The foolish feeling is misplaced. You have simply discovered something the enthusiasts rarely say out loud, which is that for a whole category of work, AI genuinely is slower than doing it yourself. In one 2025 survey, 59 per cent of workers admitted spending more time wrestling with AI tools than the task itself would have taken. The trick is not trying harder with better prompts, it is knowing which side of the line a job sits on before you open the tool at all.
The short answer. AI pays for itself on tasks where the thinking or the volume is the hard part: synthesis, long documents, first drafts of substantial things. It costs you time on tasks where the doing is already quick and the shape is already in your head: short emails, small edits, anything where explaining the job takes longer than the job itself. Task shape decides it, not your skill with prompts, so the fastest AI habit is sometimes closing the tab.
Why the handover has a price
Every delegation carries overhead, and it is the same overhead whether the delegate is a person or a model. You explain the task, wait for the attempt, read what came back, correct what missed, and stitch the result into what you were doing.
When the task is large, that overhead is a rounding error against the hours saved. When the task is a two-line email whose exact wording you already know, the overhead is the whole cost, and you have paid it for nothing. The tool did not fail, the economics of small jobs did.
Which side of the line a task sits on
| Just do it yourself | Hand it over |
|---|---|
| Short emails and messages where you already know what to say | First drafts of anything substantial: reports, proposals, the long email you have been avoiding |
| Small edits to something nearly finished | Summarising documents too long to properly read |
| Tasks where explaining the context takes longer than the task | Turning a mess of notes into structure, or one format into another |
| Anything you do so often your hands know it | Volume work: ten variants, five options, the same thing reshaped for three audiences |
| Quick factual checks you can verify faster than you can prompt | Thinking work: options you had not considered, the flaw in a plan |
The pattern underneath the table is consistent. Complexity and volume reward the handover, because that is where the overhead disappears into the savings, while short, familiar, already-shaped work punishes it. Notice that skill does not appear anywhere in that sentence, a perfect prompt for a two-line email is still slower than the email.
The evangelist problem
Some of the pressure to use AI for everything comes from people who genuinely work on the other side of the line, in synthesis-heavy and volume-heavy jobs where the savings are real. Some of it comes from people performing enthusiasm.
Either way, their experience is not an instruction. If a task keeps coming out slower with the tool than without it, that is information about the task, not a deficiency in you. The professionals getting the most from AI are conspicuously willing to not use it, because the hour it saves on the report is only real if it is not being spent invisibly on eighteen-minute emails.
The habit
Before opening the tool, one question does the sorting. Is the hard part of this job the thinking or the volume, or is it just the typing? If it is just the typing, type.
There is a fuller list of the jobs worth handing over in what to actually use ChatGPT for, and a companion piece on the tasks you should never give ChatGPT for accuracy reasons. Between the three sits the whole judgement, and this one is yours to call in about three seconds once you know to ask it.
Not sure where to start?
The finder also knows when the answer is neither.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.