AFG AI Field Guide Prompt to workflow

AI Field Guide - Classroom bridge

Turn the prompt that worked into a workflow you can reuse.

A good prompt is a win. A workflow is the part you can sell: same job, same inputs, same rules, same output shape, and the same proof check before you use the answer.

Build the card Back to first prompt Join Classroom

Starter video path

Step 1

A prompt is what you typed. A workflow is how you do the job again.

Do not stop at "that answer was pretty good." If the task will come back, save the recipe. The recipe is what makes AI feel useful instead of random.

Prompt only

"Write a reply to this customer." It worked once, but next time you still have to explain the tone, format, and rules.

Workflow

"Use my customer reply workflow." It already knows the job, the inputs, the tone, the format, and what to check.

Step 2

Fill out this workflow card.

This is the paid-product bridge. It is simple enough for a beginner, but strong enough to become a Classroom habit. 

AI workflow card

Workflow name:
[name the repeat job]

Use this when:
[when this workflow should be used]

Inputs I need:
[notes, message, transcript, photos, customer question, rough idea, etc.]

Rules:
- do not invent facts, prices, dates, names, or promises
- keep the tone [plain / warm / direct / calm / professional]
- ask one question if a missing detail matters

Output format:
[email / checklist / plan / table / script / summary]

Proof check before using:
[what a human must review]

Save location:
[where this workflow card lives]
Job

What task does this repeat?

Rules

What should AI follow every time?

Proof

What must you check before use?

Step 3

Example: customer reply cleanup.

Keep the first workflow boring. Boring is good if it saves time and you can check the result.

Name

Customer reply cleanup.

Use when

A customer asks a normal question and you need a clear reply.

Inputs

Customer message, what you know, what you still need.

Rules

No prices, dates, guarantees, or fake certainty.

Output

Short reply plus one missing-detail question if needed.

Proof

Check names, numbers, promise, tone, and next step.

Use my Customer Reply Cleanup workflow.

Here is the customer message:
[paste message without private details]

What I know:
[facts I can verify]

Rules:
- sound plain and helpful
- do not invent prices, dates, proof, or guarantees
- ask one question if a missing detail matters

Return:
1. the reply I can send
2. what you assumed
3. what I should verify before sending

Step 4

Save rules that protect you from bad AI work.

A workflow is not just a better prompt. It is a guardrail. The rules keep the answer from sounding fake, promising too much, or making up details.

Tone rule

Plain, useful, and close to how Chance would say it.

Fact rule

If it is not in the input, mark it as unknown.

Promise rule

Do not promise timing, price, guarantee, result, or approval.

Question rule

Ask one useful question when missing details would change the answer.

Step 5

Test it on a second task.

You do not know if it is a workflow until it works twice. Run the saved card on a similar job and see what breaks.

Run

Use the saved workflow on a similar task.

Fix

Update the rules where the answer drifted.

Save

Keep the cleaner version as the real workflow.

I am testing this workflow on a second task.

Tell me:
1. did the workflow fit this task?
2. what rule should I add or change?
3. what part of the answer still needs human review?
4. what should I save for next time?

Step 6

Use this explanation on YouTube.

This is the plain spoken version Chance can use without making it sound like a tech lecture. 

If one AI prompt helps you, do not just walk away.

Turn it into a workflow.

That means you save the job, the details it needs, the rules it should follow, the answer shape you want, and what you need to check before using it.

That is how you go from playing with ChatGPT to actually saving time.

One good prompt is a win. One saved workflow is where the money is.