08 · Briefing it properly

The same "help me tidy this up" gets one person a document they can ship and another a plausible-looking mess. Almost all of the difference is in the brief.

This page is not prompt mysticism. It is six things you can apply immediately.

1. The three-part formula: what, with what, how

WorkBuddy cannot read your mind. The less you give it, the more chaotic the space it has to invent in.

Bad:

Tidy up the minutes from last time.

Which time? Where are they? Tidy into what? It has to guess all three, and one wrong guess wastes the run.

Good:

Turn the minutes in D:/meetings/0320-product-review.docx into a checklist with:

1) The conclusion for each agenda item
2) The owner and due date for each
3) Disputed or unresolved items flagged

Output as a table. No preamble.

All three parts present:

• What — turn into a checklist
• With what — that specific file, by path
• How — three sections, table format, no preamble

When you do not know what to give it, ask

I want to do X. What information do you need from me?

Let it enumerate the inputs before you start. That round costs almost nothing and saves several rounds of rework.

2. Small steps: one thing at a time

The classic beginner error is throwing an enormous requirement at it in one go.

Bad:

Read this 50-page industry report and write an internal deck for my boss, then email it.

That sentence hides four independent tasks: read, extract, build the deck, send it. If any link is misunderstood everything downstream drifts, and you cannot tell until the whole thing finishes.

Split into three:

Round 1:
Read D:/reports/2026-ai-industry.pdf and extract the five most important
conclusions, one sentence each, with the source page number.

Round 2:
Based on those five conclusions, write a 15-slide internal briefing outline
covering industry trends, focused on opportunities relevant to our business,
with recommendations and next steps.

Round 3:
The outline is right. Expand slide 2, "opportunity analysis", into full slide
content — 50 words or fewer per opportunity.

The benefit is that you confirm the direction at each step and can pull it back the moment it drifts, instead of scrapping the whole thing at the end.

3. Adjust rather than give up

The first answer is a starting point. Plenty of people get one unsatisfying result and stop, which is the real waste.

Four effective adjustments:

Approach	Example
Name the problem	"Too long." "Too formal." "Paragraph three does not follow — rewrite it."
Restate from a different angle	Replace an abstract requirement with a concrete business situation
Add constraints	Audience, setting, format, length
Switch perspective	Have it think as a boss, a customer, a PM, an analyst

A real three-round exchange:

Round 1: Write an apology email to a client — last month's report had bad data.

What it gave: A formal, long apology, tone far too grave.

Round 2: Too formal. This is a client we have worked with for three years. Lighter, more like a chat message, under 200 words. Focus on what went wrong and that it is fixed. Stop apologising repeatedly.

What it gave: Much tighter, much closer to how people actually write.

Round 3: Good. Add "I'll send you the version to check before the next delivery" at the end.

Think of it as directing a hard-working intern who needs feedback. The clearer you are about what is wrong and what you want, the closer the output lands.

4. Show it an example

Language struggles to convey a feel. The cheapest and most effective move is to hand it something you are happy with and say: match this.

Tone, structure, format, register — a clear reference nails all of them at once.

last-quarter-report.docx in the attachments is our team's standard format.
Follow that structure and register, using this quarter's data.

That single line outperforms ten lines of formatting instructions.

5. Give it an identity

WorkBuddy ships an Experts (专家) feature with preset roles across industries — legal advisor, product manager, data analyst, marketing copywriter. Pick one and it organises answers along that role's lines.

Underneath, this is a layer of role priming that aligns the framing, priorities and vocabulary to a field. For work needing specialist framing or a specific output style, it is noticeably more consistent.

You can also just write the role into the prompt:

You are a financial analyst with ten years in investment banking. Judging by
the best research you have seen, analyse this company's financials. Focus on
revenue structure, cash-flow risks, and competitive position against peers.

Give a clear recommendation. Where data is missing, make assumptions boldly
but label them as assumptions.

I want insight, not a data transcription.

Note the last two clauses: "label them as assumptions" stops it presenting guesses as facts, and "not a data transcription" sets the quality bar explicitly.

A role is not a qualification

Having it play a lawyer, doctor or financial adviser improves organisation and register. It does not confer professional standing. Medical, legal, investment and compliance decisions still need a qualified human — the official docs say the same.

6. Ask it to self-check

One extra line at the end of a complex task costs nothing and pays off visibly:

Self-check before output: confirm the structure is complete, there is no
duplication, and the conclusions are concise.

Especially useful for research work:

Complete this task independently: gather the significant developments in AI
office tooling from the past week, produce a briefing with source links, and
check for duplicate information before output.

A prompt checklist

Scan these before you hit send:

1. Is the goal explicit — what to produce, or what to process
2. Is the input named — a filename or path, not "that file"
3. Is the output format stated — Word / Excel / deck / Markdown / HTML, and where to save it
4. Are the constraints given — length, style, audience, required sections
5. Have you stuffed three jobs into one — if so, split them
6. Did you demand a preview for risky actions — deletion, overwriting, bulk renaming, always list first

When not to be specific

There is one case where detail hurts: when you do not know what you want either.

Piling on constraints then just locks it into the thinking you already had. Better to switch to Ask mode and have it offer three directions, then pick one and refine.

I want to run a quarterly retrospective with my team. What are the different
ways of structuring one, and what is each good for? Do not write the content
yet — give me three directions and their trade-offs.

Next

That is the end of Get started. Two routes from here:

• Go deeper on the product → Go deeper
• Find something to copy today → Playbooks

Reusable prompts are collected in the Prompt library.