A prompt framework that works the same way twice

Four fields that keep output consistent

After this, you'll be able to design a reusable prompt with four fields: role, audience, format, and success criterion. The structure it produces stays consistent whether you run it this week or hand it to a colleague next month.

Before you start

Complete Working with Claude Code first. This lesson builds on writing precise, structured task descriptions for AI to execute reliably.

The idea

You've run the same prompt twice and gotten two completely different structures back. That's not random. It means the prompt is missing one of four fields that make output repeatable. Role tells the AI who is speaking and what expertise to assume. Audience tells it who the output is for and what they specifically care about. Format specifies structure, length, and tone exactly. Success criterion gives one testable definition of done you can verify in under 10 seconds.

Here is the before and after. A vague prompt like 'Write a summary of this article' returns a different structure every time. Sometimes short. Sometimes a listicle. Sometimes focused on facts. An engineered prompt with all four parts returns a consistent shape across runs, even when the input content changes.

Now try it with a task you do weekly. Fill in all four fields, paste it into Claude with a piece of real content, and run it a second time with different content to confirm the structure stays consistent.

Try it (8 min)

Watch out for

Using 'clear' or 'professional' as the format field. The AI picks its own structure, and it changes with every run.
A success criterion like 'covers all the main points' requires reading everything to verify. You end up re-reading every time without knowing whether the prompt succeeded.
Confusing role and audience. The prompt ends up giving the AI conflicting signals, and the output voice gets muddy.
One run is not a reusability test. The prompt passes when it produces the same structure on a second, unrelated piece of content without any edits.

Paste this into Claude

I need to design a reusable prompt for [one task you do repeatedly, such as a weekly status update, client email, or design brief]. My role is [your job title or context (for example, 'a small business owner' or 'a teacher explaining this to 8th graders'); what expertise should the AI assume you have]. My audience is [describe who reads or uses this output and what they specifically care about; not just their title, but their concerns]. The format I need is [specify length, structure, bullet points vs paragraphs, tone; be as specific as possible]. What done looks like: [one measurable criterion you can verify is true without reading everything]. Write the best version of this prompt using those four inputs. Then tell me which of the four fields had the biggest impact on shaping your output and why.

What good looks like

The prompt contains all four parts: role, audience, format, and a measurable success criterion
You can paste a second piece of content into the prompt and get the same structure without editing the prompt
The success criterion is something you can verify in under 10 seconds
The format field names a specific structure and length rather than vague descriptors like 'clear', 'professional', or 'thorough'

When this breaks

Breaks when the format field is vague because the AI picks a structure based on its training data, not your use case, and the structure changes with each new input
Breaks when the success criterion is unmeasurable because you cannot tell whether the output worked without rereading everything
Breaks when used with new content because the role and audience fields may not fit the new context.

You can now

Paste the completed prompt with a second, unrelated piece of content and confirm it produces the same structure and format as the first run without any prompt edits.

Key takeaways

Four fields make a prompt repeatable: role, audience, format, and success criterion. Any prompt missing one of the four works by luck, not by design.

Role, audience, format, and success criterion are the four parts that make a prompt work the same way twice
A success criterion you can verify in under 10 seconds is the most skipped field. Without it, you cannot tell whether the prompt worked.
Reusable prompts are tested by running them on a second piece of content; consistency is the measure, not quality of the first output
Format means specific structure and length; 'professional' and 'clear' are not formats

Was this helpful?

← Back to lessons

I need to design a reusable prompt for [one task you do repeatedly, such as a weekly status update, client email, or design brief]. My role is [your job title or context (for example, 'a small business owner' or 'a teacher explaining this to 8th graders'); what expertise should the AI assume you have]. My audience is [describe who reads or uses this output and what they specifically care about; not just their title, but their concerns]. The format I need is [specify length, structure, bullet points vs paragraphs, tone; be as specific as possible]. What done looks like: [one measurable criterion you can verify is true without reading everything]. Write the best version of this prompt using those four inputs. Then tell me which of the four fields had the biggest impact on shaping your output and why.