Prompt Craft Tutorial: Write AI Prompts That Survive Real Work

Bad prompts create busywork. Good prompts create repeatable leverage.

This tutorial teaches a prompt pattern you can reuse across ChatGPT, Claude, Gemini, and most coding assistants.

The 5-part prompt pattern

Every strong work prompt includes:

Role — who the model should act like
Objective — the finished artifact
Context — only the facts that matter
Constraints — tone, length, what to avoid
Output format — exact structure you want back

If any piece is missing, quality drops.

Master template

Role: {{expert type}}
Objective: {{finished artifact}}
Audience: {{who will read/use this}}
Context:
- {{fact 1}}
- {{fact 2}}
Constraints:
- {{style / length / must-include / must-avoid}}
Output format:
- {{bullets / table / sections / JSON}}
Do not invent facts. If information is missing, ask up to 3 clarifying questions first.

Tutorial 1: Drafting a useful blog outline (10 minutes)

Prompt

Role: editorial strategist for an AI tools blog.
Objective: create an outline for a practical tutorial.
Topic: {{topic}}
Audience: beginners who are busy and skeptical of hype.
Constraints:
- Include a clear outcome in the intro
- Prefer steps over theory
- Add a mistakes section
- Keep SEO keywords natural
Output format:
1. Title options (3)
2. Meta description
3. H2/H3 outline
4. Internal link ideas

What “good” looks like: you can start writing immediately without another brainstorming session.

Tutorial 2: Research assist without fake citations (15 minutes)

Role: careful research assistant.
Objective: summarize what is known and unknown about {{topic}}.
Constraints:
- Separate verified facts, common claims, and open questions
- Do not invent citations
- Flag anything that needs primary-source checking
Output format:
| Claim | Confidence | How to verify |
Then: 5 follow-up search queries.

Use the table as a checklist. Click through and verify before you publish.

Tutorial 3: Critique and tighten your own draft (8 minutes)

Role: strict editor.
Objective: improve clarity and usefulness.
Paste draft below.
Constraints:
- No generic praise
- Preserve my voice
- Cut hype words
Output:
1. Top 5 problems
2. Line edits for the intro
3. A stronger checklist ending
Draft:
{{paste}}

Tutorial 4: Turn meetings into actions (5 minutes)

Role: chief of staff.
Objective: extract decisions and next actions from this transcript.
Constraints:
- Ignore small talk
- Mark owners and due dates when stated
- List unresolved questions separately
Output:
## Decisions
## Actions (owner · task · deadline)
## Risks
## Customer language worth saving
Transcript:
{{paste}}

Prompt library starter pack

Save these names in your notes app:

outline.v1
draft.v1
critique.v1
research.v1
meeting-actions.v1
sop-from-notes.v1

Version them. When a prompt improves, bump v1 → v2 instead of overwriting blindly.

Evaluation checklist

Before you trust an AI answer, ask:

Did it follow the output format?
Did it invent specifics you did not provide?
Can you verify the key claims in 5 minutes?
Would you send this to a client after one edit pass?

If the answer to #4 is no, improve the prompt — do not just regenerate forever.

Common prompt mistakes

“Write something good about AI tools” (no objective)
Dumping an entire company wiki into every chat
Asking for citations the model cannot reliably provide
Accepting the first answer because it sounds confident
Never saving the prompt that worked

Closing

Prompt craft is not magic language. It is product requirements writing for a model.

Use the 5-part pattern, keep a tiny prompt library, and judge outputs by whether they reduce your next hour of work. That is the difference between AI as entertainment and AI as infrastructure.

Keep exploring on AutoThinkers — these 7 guides pair well with this one: