Turn a Report You Dread Into a 10-Minute Task
Almost everyone has one: the recurring report nobody enjoys. The Monday status email. The monthly numbers recap. The “how did the campaign do” summary you rebuild from scratch every time. It’s not hard — it’s just tedious, and it eats an hour you’d rather spend elsewhere.
This is the single best first project for AI at work, because the format is stable and the judgment is repeatable. Here’s how to hand it off in one sitting.
Step 1: Do it by hand one more time, but narrate it
Open the report you already produce and write down, in plain words, what you actually do. Not the polished version — the messy internal logic:
- “I pull the numbers from the dashboard export.”
- “I flag anything that moved more than 10% from last week.”
- “I always call out wins first, then problems, then what I need from people.”
- “I keep it under 200 words because my boss won’t read more.”
That list is the instructions. Most people skip this step and then wonder why the AI’s version feels generic — you never told it the rules you keep in your head.
Step 2: Give it one real, finished example
Paste in an actual past report — a good one you were happy with. This is
the highest-leverage move in the whole process. The model learns your
structure, tone, and level of detail from a single strong example far
better than from any description of it. This is called few-shot prompting
Just a fancy name for “show it a couple of examples of what you want
before asking.” Models are dramatically better at matching a pattern they
can see than at following a description of one — the same way it’s easier
to copy a sample than to work from written instructions.
Step 3: Feed it the raw data and ask for a draft
Now paste in this period’s actual source material — the export, the figures, the notes — and say something like:
“Here’s this week’s data. Using the same format, structure, and tone as the example above, write this week’s report. Flag anything that moved more than 10% from the prior week. Keep it under 200 words.”
Give it the raw export, not your summary of it. The model does better work with the real material than with the version you’ve already trimmed — you’ve usually removed the exact details that would have mattered.
Step 4: Correct in place — don’t start over
The first draft will be 80% there. Resist the urge to open a fresh chat. Stay in the same conversation and fix it the way you’d redline a colleague’s draft:
- “Move the budget line to the top — that’s what she cares about most.”
- “You missed that the drop in signups was expected because of the holiday. Note that.”
- “Too formal. Write it like an internal Slack message.”
Each correction sharpens the next output. And here’s the payoff: those corrections are reusable.
Step 5: Save the recipe
Once you’re happy, collect the whole thing — your Step 1 rules, the example, and the corrections you made — into one block of text. Save it in a note. Next week, you paste that recipe plus the new data, and you’re done in the time it takes to read the draft.
If your AI tool supports it, this is exactly what a saved project or custom instruction
A place inside the tool where you store standing instructions and
reference files once, so you don’t re-paste them every time. Different
products call it different things — a “project,” a “GPT,” “custom
instructions,” a “gem” — but they all do the same job: remember your setup
so each new chat already knows the rules.
The takeaway
The goal isn’t to remove yourself from the report — it’s to remove the retyping from it. You still own the judgment: what matters, what to flag, what to leave out. You’ve just stopped rebuilding the scaffolding every single time.
Next up: how to build a reusable “assistant” for a task you do constantly — so you stop re-explaining yourself in every new chat.