Toybird Labs Blog
AI at work

Use ChatGPT at Work | Build an AI Work Brief Before the Prompt

Instead of opening ChatGPT and inventing the request from scratch, define the work first. An AI Work Brief captures the task, source material, audience, constraints, unresolved items and finished output.

5 min

This guide is a route into the task-specific articles, not a collection of supposedly universal prompts. You can use the examples without installing an application.

Choose the job before choosing the wording

Writing a reply is creating a document. Extracting an owner from notes is organizing evidence. Comparing quotes is checking conditions. Rehearsing a sales conversation is practising a skill. These jobs need different materials and different checks.

Job Bring this material Aim for this result Check before using it
Reply to a customer Received message, confirmed facts, next action A reply draft No invented cause or promise
Decline a request Request, firm limits, approved alternatives Clear refusal and next step Start date is not mistaken for completion
Organize meeting notes Notes of statements and decisions Actions with owners and dates Questions are not turned into agreements
Summarize a document Source, reader and decision Evidence-linked decision note Numbers and footnotes are retained
Compare options Quotes and mandatory conditions A like-for-like table Unknown does not mean zero or absent
Clarify a vague brief What is known and what is difficult A few useful questions Already answered questions are not repeated
Rehearse a conversation Role, stage and objective Dialogue followed by feedback Roleplay and coaching stay separate
Explain to a new audience Accurate original and reader's knowledge An accessible explanation Simpler wording has not changed conditions
Draft instructions Actual steps, exceptions and checks A reproducible first draft No imaginary buttons or steps
Organize research Sources, dates and a question Facts separated from interpretation Sources support the individual claims

Choose one row. You do not need to put every possible safeguard and output type into one enormous prompt.

A small change in purpose changes the output

Suppose you have a proposal. “Summarize this” is not the same job as “identify what a manager must decide before adoption.” The latter may need a cost calculation, a rollout contradiction and missing responsibilities rather than a page-by-page recap.

Or suppose you have three quotes. Start by normalizing participant count and billing units. Ask for a recommendation only after distinguishing mandatory requirements from preferences. This makes the request useful even when the result is “clarify these two points first.”

A reusable request structure

Task: [one specific action]
Purpose: [what the reader will decide or do]
Source material: [the message, notes, document or data]
Confirmed conditions: [dates, limits, required facts]
Unknowns: [what must not be guessed]
Output: [reply, table, questions, checklist or dialogue]
Before finishing: [the particular mistake to check]

These fields are prompts to think, not a mandatory form. A simple task may need only three of them. When a response goes wrong, check which decision-making information was absent rather than simply asking for more detail.

Continue with a complete example

Read next

Ask ChatGPT Questions First: Turn a Vague Task into a Usable BriefYou do not need a perfect prompt before asking for help. But an unlimited interview can consume the time you hoped to save. Ask questions that change the work, then define when to stop and draft.Compare Quotes with ChatGPT: Put Three Proposals on the Same BasisA per-session price, an all-inclusive quote and a per-person fee cannot be compared by choosing the smallest number on the page. First calculate the same required service, then check its mandatory conditions.Write a Customer Reply with ChatGPT: Facts, Promises and the Next UpdateA customer needs an answer before their meeting. Your technical team has not established the cause. You can still write a useful initial reply without turning the customer's deadline into a recovery promise.Summarize a Long Document for a Decision, Keeping Numbers and Footnotes“Annual cost: JPY 360,000; launch in November” is a clear summary. It is also wrong for the proposal below because it drops setup costs and launch conditions. A decision summary must stay connected to its source.Turn Meeting Notes into Action Items Without Inventing Owners or DatesA person's name in a meeting note does not necessarily make them the task owner. “Ask Tanaka” may be Sato's action. A reliable follow-up list distinguishes decisions, proposals, cancellations and unresolved assignments.Practise a Sales Conversation with ChatGPT When the Customer Is Not Looking to Change“Be a difficult customer” often creates an argument rather than a useful rehearsal. In an early sales conversation, practise finding out whether there is a reason to change before rushing to a demonstration.

The detailed guides contain the actual input material and an edited completed example. Read the finished result alongside its source; do not treat an attractive answer as proof that the source was understood.

Fix one meaningful difference at a time

After a first answer, give a specific correction: “Keep the amounts and conditions. Move the conclusion first,” or “The date is a communication deadline, not a completion promise. Correct only that sentence.” If the issue is factual, return to the original material before adjusting tone.

For work involving sensitive information, use your organization's approved environment and rules. The same text may be acceptable in one setting and inappropriate in another. Removing names alone is not a complete confidentiality check.

Find uses you have not yet turned into templates

Prompt Ready · English UI
Prompt Ready · English UI — Product-interface reference, not an execution of this article's fictional case. Enlarge image

Prompt Ready groups work by purpose. The home screen provides categories; the ideas view makes specific jobs easier to find. They help with two different problems: discovering what to ask an AI to do, and reconstructing the conditions for today's task.

Choose a use, add the source text or topic and conditions, inspect the request, then copy it into your usual AI. Prompt Ready constructs the request locally; it does not write the external AI's answer or send a customer email. Once you paste information into another service, that service's data terms apply.

Start with one task you actually repeat. Keep its completed result and the checks that made it usable. Expand to another task when you understand the first workflow, not merely when you have collected more prompt templates.

Toybird Labs / Masaki Iino. We develop and provide the products discussed here.