How to Write Effective AI Prompts: The DASH Framework

Most prompts fail for the same reason: they ask a model to guess. Effective prompts leave nothing to chance. They tell the model who it is, what to do, how to shape the answer, and how to know when the answer is good.

This guide walks through the DASH framework we use at Dash2Grow to write prompts that hold up in production - Define, Ask, Shape, Hone.

Why prompt engineering matters

A well-engineered prompt is the difference between a demo and a workflow. The same model, given the same task, can produce a paragraph of fluff or a structured answer you can pipe into the next step. The model didn't change. The prompt did.

The good news: prompt engineering is not a black art. It is a small number of habits, applied consistently.

The DASH framework

DASH is four steps you run every time you write or refactor a prompt.

• Define: who the model is and what context it needs.
• Ask: the single task, stated plainly.
• Shape: the exact output format you want back.
• Hone: the checks the model should run before returning.

D - Define the role and context

Start by giving the model a role and the minimum context it needs. Roles anchor tone and expertise; context anchors accuracy.

You are a senior SEO strategist reviewing a B2B SaaS homepage. The company sells API monitoring to backend engineers.

Avoid stuffing the prompt with everything you know. Include only what the model needs to answer the specific ask.

A - Ask one thing, plainly

The single biggest failure mode in prompting is asking three things at once. Split them into separate prompts, or chain them.

Rewrite the hero headline to lead with the outcome for backend engineers. Keep it under 12 words.

Notice the constraint. Constraints are how you turn a vague ask into a testable one.

S - Shape the output

If you want structured output, ask for it explicitly. Name the fields. Show an example. Tell the model what not to include.

Return JSON with keys: headline (string), rationale (string, 1-2 sentences), alternatives (array of 3 strings). Do not include any prose outside the JSON.

Shape is where prompts become reusable. Once you can trust the shape, you can pipe the output into evals, tests, or the next step in a workflow.

H - Hone with checks

Ask the model to check its own work before returning. This is the cheapest quality lever in prompting.

• Confirm the headline is under 12 words.
• Confirm the rationale references backend engineers.
• If any check fails, revise and try again before returning.

Self-checks catch the boring mistakes that would otherwise force a second call.

A full DASH prompt

Here is a complete DASH-shaped prompt you can adapt today.

You are a senior SEO strategist reviewing a B2B SaaS homepage. The company sells API monitoring to backend engineers. > > Rewrite the hero headline to lead with the outcome for backend engineers. Keep it under 12 words. > > Return JSON with keys: headline (string), rationale (string, 1-2 sentences), alternatives (array of 3 strings). Do not include any prose outside the JSON. > > Before returning, confirm the headline is under 12 words and the rationale references backend engineers. If either check fails, revise and try again.

Common mistakes to avoid

• Asking the model to be creative without giving it a constraint to be creative against.
• Providing context the model does not need for the specific task.
• Skipping the output shape, then complaining the answer is unstructured.
• Treating one bad answer as proof the model cannot do the task. Try three variations first.

Where to go next

Once DASH is second nature, layer in evals: a small set of inputs and expected outputs you can rerun as you tweak the prompt. Evals turn prompt engineering from vibes into a measurable practice.

If you want ready-made DASH prompts for common workflows - SEO, sales, research, product - browse the [Marketplace](/marketplace) for prompt packs from creators who ship for a living.