The 6-Part Prompt Structure That Stops the Rewrite

There’s a specific moment when people give up on AI for real work.

It’s not when the output is bad. Bad output is expected. It’s when the output is almost right — correct in substance, wrong in every detail — and fixing it takes longer than writing the thing yourself would have.

That’s a prompt problem, and it has one cause: the prompt described a task but not the shape of the answer. So the model produced a reasonable shape, which was not yours.

Six parts fix it. Not six paragraphs — six pieces of information, most of which are one line each.

The six parts

1. Who is speaking

One line. Not “you are a helpful assistant” — that phrase does nothing. Something with a stake in the outcome:

The second clause is what changes the output. It gives the model a bias — in this case, against cleverness — and biases are what produce a voice instead of a average.

2. What I actually want

One sentence, stated as an outcome rather than an activity. “Write a landing page” is an activity. “Get a visitor who has never heard of us to understand what we do in five seconds” is an outcome, and it constrains every decision that follows.

3. The material

Everything the model needs that it can’t infer: your product, your buyer, what you’ve tried, what failed, the actual numbers. This is the part people skip, and it’s the part that determines whether the output is generic.

Rule of thumb: if you’d have to explain it to a new freelancer on their first day, it goes in.

4. The constraints

Length, tone, things to avoid, things that are non-negotiable. Be specific about what to avoid — negative constraints work better than positive ones because they’re unambiguous:

5. The output shape

This is the part that eliminates the rewrite. Show the structure you want:

When you specify the shape, the model fills slots instead of composing prose. Slot-filling is far more predictable than composition, and it’s trivially editable.

6. What to do when it’s uncertain

Without this, the model invents. With it, you get a visible gap instead of a confident error:

This single instruction is the difference between a draft you can check and a draft you have to fact-check line by line.

Before and after

Before:

What comes back: three paragraphs about “unlocking productivity,” a bulleted list of features phrased as benefits, and a call to action. It’s grammatical, it’s plausible, and it’s unusable, because it could describe any product in the category.

After:

What comes back is shorter, flatter, and usable. Not because the model got smarter, but because there was nothing left to guess.

Why part 5 matters most

If you only adopt one of the six, adopt the output shape.

The reason is mechanical. Free-form prose requires the model to make dozens of small structural decisions — where to put the claim, how to open, whether to add a transition. Each decision is a chance to diverge from what you wanted. A specified shape removes those decisions.

It also makes iteration cheap. When the output has named slots, you can say “the PROOF line is weak, give me three alternatives” instead of “rewrite this.” Targeted edits beat full rewrites every time.

The mistake underneath all the mistakes

People write prompts as if they’re talking to a person who shares their context. The model doesn’t. It has no idea what your product is, who buys it, or what you tried last month that didn’t work.

Every one of the six parts is a way of transferring context that you have and it doesn’t. Skip one, and it fills the gap with the average of everything it’s read — which is exactly what generic output is.

The test for a prompt, before you send it: could a competent stranger produce the right output from this? If not, the prompt is missing context, not the model.

Making it a habit

Writing six-part prompts from scratch every time is too slow, so people stop.

The workable version: build the context block once — your product, your buyer, your constraints, your banned words — and save it. Then each new prompt is the context block plus two lines: what you want, and the output shape.

That’s the whole system. One saved block, reused across every task, with the task-specific part kept short.

If you want the context blocks already written for the jobs that come up most — marketing, sales, content, support, operations — that’s what the AI Prompt Vault is: 80 prompts, each with the context block and output shape filled in, so you’re editing rather than composing.

Keep reading

• 7 Best AI Automation Tools for Small Business Workflows (Plus Setup Blueprints)
• The Best AI Tool Stack for Solopreneurs and Freelancers in 2025 (Plus Plug-and-Play Automations)
• The Best AI Automation Tools for Solopreneurs and Freelancers in 2025 (With Plug-and-Play Blueprints)