You keep meaning to try AI at work, but your calendar is already full. There is a workshop you have been meaning to book, a newsletter you have been meaning to read, and three Slack messages from a colleague who keeps swearing by ChatGPT. The irony is not lost on you — the tools that are supposed to give you time back are currently taking time away from the actual work.
I kept promising myself I'd "block an afternoon" to learn this. That afternoon never came. What actually worked was five tiny fifteen-minute experiments spread across two weeks — each one tied to a task I was already doing that day. None of them were clever. All of them saved real minutes.
This is the short, honest version of what worked. Five specific tasks you can try this week without learning anything new, with a starting prompt you can paste into ChatGPT, Claude, or Gemini and adapt. Each one maps to the Improve pillar of the ICE Method — taking something you already do and making it faster, not inventing new work.
Before we start — the only thing you actually need
One free account with one of the three tools above. That is genuinely it. No plug-ins, no automations, no prompt libraries memorised.
A small warning before we begin: do not paste anything into a public AI chat that you would not be happy seeing on a whiteboard in the lobby. Client names, salary figures, unreleased numbers — strip them out, or replace them with placeholders, before you paste. That habit takes about ten seconds and it is the single most important one on this page.
Way 1 — Turn your inbox into first-draft replies
This is the one we'd suggest trying first, because almost everyone has a long, awkward email sitting in their inbox right now. The kind where someone has asked three questions in two paragraphs, and a clean reply requires you to think about tone before you can even start typing.
Here is what we do instead. We paste the whole thread into the chat — every back-and-forth, headers and all — and ask for three reply options at different tones. Warm, neutral, and briefer-than-feels-comfortable. Then we pick the one closest, paste it back into our email client, and edit the bits that do not sound like us.
The honest time saving is about fifteen minutes per long email. Not because writing three paragraphs takes fifteen minutes, but because the staring-at-a-blinking-cursor phase evaporates. You are editing, not composing. Your brain finds that much easier.
Here is an email thread I need to reply to. Please draft three possible
replies: one warm and personal, one neutral and professional, one short
and direct (under 80 words). Keep my likely voice in mind — I am a
[your role] writing to [their role]. I need to [cover X, decline Y,
acknowledge Z]. Thread below:
[paste whole thread]
If you want to get better at this: after you pick one, ask the tool why it wrote what it did — the patterns you learn there carry across to every other writing task.
Way 2 — Compress long documents into 5-bullet decision briefs
You know the feeling — a 40-page PDF lands in your inbox with "thoughts by Friday?" written above it. Most weeks, you either skim it badly and contribute vaguely, or you block ninety minutes that you do not have.
Compression is a near-perfect fit because you are not asking the tool to invent anything — you are asking it to summarise what is already there. "Summarise this" produces wallpaper. "Give me the five decisions this document is asking me to make, one line of evidence for, one line against" produces a brief you can actually use in a meeting.
Rough time back: twenty minutes per document, sometimes more. The trick is being specific about what kind of summary you want — a decision brief is not the same as an overview.
Please turn the document below into a 5-bullet decision brief for me.
Each bullet should be a decision the document is implicitly or
explicitly asking me to make, one line of evidence for, one line of
evidence against, and a one-line recommended default. Assume I have
30 seconds to read this before a meeting. Document below:
[paste document]
If you want to get better at this: try the same document with two different "shapes" — a decision brief and a risk register — and notice how differently the same text reads depending on what you asked for.
🧠 Quick Challenge: It is 9:15am on Monday. You have a 60-page board pack to read before 11am, a long customer email you have been avoiding since Friday, and a messy spreadsheet of last week's figures that your manager wants "turned into a one-paragraph summary" by lunchtime. Based on the three tasks above, which is the right first use of AI this morning?
- A) The board pack — biggest document, biggest potential time saving
- B) The customer email — smallest task, lowest risk, fastest win
- C) The spreadsheet summary — most visible to your manager, best for credibility
Answer: B) The customer email. It is the smallest, lowest-stakes task on the list, which is exactly why it is the right first experiment. You will learn how the tool phrases things in your voice before you trust it with the board pack (where a bad summary is a bigger problem) or the spreadsheet note (where your manager will see the output). Way 1 specifically flagged the email as the one to try first — small, private, and forgiving if the first draft is mediocre.
Way 3 — Rewrite your own draft in a different voice or length
This one takes a moment to click because it feels like cheating. It is not. It is the opposite of blank-page generation — you have already done the thinking, and you are asking the tool to help with the revision.
You have probably felt this: you have written a paragraph, you know it is slightly wrong, but you cannot quite see what is off. Maybe it is too stiff. Maybe it is three sentences too long for the audience. Re-reading it for the fourth time rarely fixes it — and that re-reading is where the minutes disappear.
What helps is pasting your own draft in and asking for three alternative versions at different tempos or tones. You keep the one closest to what you wanted and ignore the other two. You are not outsourcing your voice — you are getting a second pair of eyes at 11pm. Rough time back: ten minutes per draft.
Below is a draft I have written. I think it is close, but something
is off. Please give me three rewrites of the same content: (1) the
same length, but a warmer, more conversational tone; (2) half the
length, same tone; (3) same length, slightly more formal. Do not
add new ideas. My draft:
[paste your draft]
If you want to get better at this: compare the three rewrites line by line against your original — you'll notice which of your own habits are working and which are not. Our guide on writing AI prompts that actually work covers the full four-part structure if you want to go deeper.
Way 4 — Turn an ugly spreadsheet note into a clean message
Every team has a person whose rows read like this: "Q3 L8 slippage — HR still chasing, est close F17, two reds amber by Mon — pls flag to SLT." It makes sense to them. It makes sense to approximately no one else.
This is a ridiculously good use of AI and it is wildly under-used. You paste the raw note — abbreviations, fragment sentences, the lot — and ask for a client-ready version, or a board-ready paragraph, or a Slack-ready update. The tool does not need perfect business context. It just needs enough to turn shorthand into full sentences.
We use this when a colleague sends us a raw note that needs to go to a client, and when we are tidying up the end-of-week summary nobody quite wants to write. Rough time back: ten to fifteen minutes per message.
⚖️ Raw note to client-ready message — rough minutes:
- Without AI — 15 minutes
- With AI — 3 minutes
Source: "Way 4" section above
The text below is a raw internal note with abbreviations and
shorthand. Please turn it into a clean, professional 3-sentence
message suitable for sending to a client. Keep all the facts intact.
Do not invent details. If anything is ambiguous, list the ambiguity
at the end rather than guessing. Raw note:
[paste the note]
If you want to get better at this: always keep the "list the ambiguity at the end" line in. It is the single most useful instruction for this kind of task, and it prevents the polite but wrong reply that is the classic AI failure mode.
Way 5 — Plan your week from a messy task dump
Sunday evening, Monday morning, or the quiet ten minutes after a chaotic meeting — most weeks, there is a moment where you scribble down a list of things to do and stare at it, slightly paralysed, unsure which to start.
This is a task where we expected AI to be useless and found it genuinely helpful. Not because it prioritises better than you do — it does not know your week. But because it forces the list into a shape. That structure is what you were going to do anyway, just badly, in your head, at a cost of fifteen minutes and a low-level sense of dread.
Here's what we do: paste the brain-dump, add any deadlines we know, and ask for a prioritised list with reasoning. We ignore about a third of the reasoning and accept the rest. The accepted portion is enough to get started — which is the whole point.
Below is a brain-dump of everything I think I need to do this week.
Please sort it into four groups: (1) do today, (2) do this week, (3)
delegate or delay, (4) probably does not need to happen. For each
item give one short sentence of reasoning. Here is extra context you
should weigh: [deadlines, meetings, known blockers]. Brain-dump:
[paste the list]
If you want to get better at this: after the list comes back, ask for "the one thing that, if I only did this, would make this week feel successful." That single follow-up is often more useful than the full sort.
Stacking these for about an hour back a week — the arithmetic (soft numbers)
Let's do the rough maths honestly. Not the marketing kind, where you add up best-case numbers and announce that AI saves you half a day.
One long email a week with Way 1: about fifteen minutes. One long document compressed with Way 2: about twenty minutes. One messy note cleaned with Way 4: about ten minutes. A Sunday-night sort with Way 5: about fifteen minutes. That is already an hour — and we have not touched Way 3.
Will you get every one of those wins every week? Probably not. Some weeks, nothing lands that needs this treatment. Some weeks, three things do and you claw back closer to ninety minutes. Roughly an hour a week back is a realistic outcome once you get past the first fortnight of clumsy prompts.
And that first fortnight is real. The outputs will be a bit off. You will phrase things awkwardly. You will occasionally get a reply so generic it feels worse than writing the email yourself. That is not a sign you are doing it wrong — it is a sign you are doing it.
A tiny habit that makes all five stick — the 15-minute rule
Here is the small change that turned these from "things I know about" into "things I actually do."
Once a day, for the next two weeks, take any task on your list that is likely to take about fifteen minutes — and before you start it, spend ninety seconds trying it with one of these five prompts. If the output is useful, use it. If not, bin it and do the task the way you always have.
You are not committing to a new workflow. You are running twelve tiny experiments at roughly zero cost. By the end of two weeks, you will know — from your own work, not from someone else's article — which two or three of these are worth keeping.
If your first attempt at any of these lands with a thud, our rundown of the mistakes almost everyone makes is probably the shortest fix. And if you want more ideas on where to start, our guide on finding your first real AI win at work is the natural next step.
You do not need a clear afternoon to start. You need fifteen minutes on a Tuesday, an email you were going to answer anyway, and a willingness to paste. You have got more than enough to get started.
Ready to try one of these prompts this week? The Prompt Library has ready-to-paste versions of all five, plus variations for different roles and industries. Pick the one closest to the task on your list right now and run it once — the rest gets easier from there.