Write a good request — Pazi Documentation

Write a good request

You do not need a special prompt formula. Give the agent the same briefing you would give a capable colleague: the outcome, useful context, constraints, and the form of a good result.

Start with the outcome

Lead with what should be true when the work is finished. “Research competitors” is a topic; “recommend the three competitors we should study before setting pricing” is an outcome.

Prompt

Recommend the three project-management products we should study before pricing our tool for agencies with 5–25 employees. Compare their entry price, collaboration model, strongest differentiator, and the customer they appear to serve best. Deliver a cited HTML report and end with your recommendation.

Add the context that changes the answer

Useful context includes your audience, business model, geography, deadline, existing assets, previous decisions, and what you have already tried. Link or upload source material instead of asking the agent to guess.

Tell the agent which facts are fixed and which are open to challenge. If an existing website, brand guide, spreadsheet, or codebase matters, provide it at the start.

State constraints and permissions

Be explicit about boundaries such as budget, time, allowed sources, tone, data handling, and whether the agent may contact anyone or change an external system.

Prompt

Draft the campaign, but do not publish it or contact customers. Use only claims supported by the attached research. Keep the paid-media budget below $2,000 and flag anything that needs legal review.

Define the deliverable

Ask for the format you will actually use: a short decision memo, an HTML report, spreadsheet, slide deck, source code, image set, or a message ready to send. Mention required sections, columns, or evidence when they matter.

Avoid over-specifying layout when the content matters more. The agent can choose tables, charts, and sections based on the request, and you can refine the presentation afterward.

Break up high-risk work

For irreversible or externally visible work, split planning from execution.

  1. Ask the agent to research and propose a plan.

  2. Review facts, recipients, costs, and permissions.

  3. Approve the exact action or ask for changes.

  4. Verify the result after execution.

This is especially important for purchases, account creation, production changes, outbound messages, legal or financial material, and deletion.

Guide the agent while it works

You can correct assumptions, narrow scope, or add sources in a follow-up. Ordinary messages queue so the current turn can finish in order. Use Send now only for a genuinely urgent correction that should interrupt the active run.

If the agent asks a structured question, answer it in the form above the composer. The form can stay open while you continue normal chat, and your submitted answers become part of the conversation.

Turn successful prompts into systems

Do not keep pasting a long instruction when the same standards recur.

  1. Put stable facts and preferences in memory.

  2. Put a repeatable method and quality bar in a skill.

  3. Put time-based repetition in Tasks & Calendar.

  4. Keep the current project's changing details in its conversation or files.

A practical template

Prompt