Prompt Engineering Starter Template

Most people learn prompting by trial and error — tweaking a request over and over until the output finally looks right. That works, but it’s slow, and the lessons rarely transfer to the next task. A reusable template fixes that: it gives you a structure you fill in once per task instead of reinventing your prompt from scratch every time, and it teaches you, by repetition, which pieces of information actually move the needle on output quality.

This starter template works for nearly any task — writing, analysis, coding, planning, or research — because it separates the five things a model actually needs from you into distinct, fillable slots.

The core template

Role: You are a [type of expert] with experience in [specific domain].

Task: [What exactly do you want done? Be specific about the deliverable.]

Context: [Any background the model needs — audience, constraints, prior decisions.]

Format: [How should the output be structured? Bullet points, a table, a specific length?]

Constraints: [What should the model avoid? Tone, length limits, things to exclude.]

Example (optional): [A sample of the kind of output you want, if you have one.]

Why each slot matters

Role

Assigning a role anchors the model’s tone, vocabulary, and depth of expertise. “Act as a senior copywriter with 10 years in B2B SaaS” produces noticeably different output than no role at all — the model draws on patterns associated with that persona rather than defaulting to a generic, average response.

Task

This should be the single clearest sentence in your prompt. Vague tasks (“help me with marketing”) produce vague answers. Specific tasks (“write three subject lines for a pricing announcement email”) produce specific, usable output.

Context

Context prevents the model from guessing at things you already know — your audience, prior decisions, brand voice, or constraints from earlier in a project. Every assumption you don’t state is an assumption the model has to make on its own, and it won’t always guess the way you would.

Format

Asking for a specific structure up front — a table, a numbered list, a fixed word count — saves you a rewrite pass. Models are generally good at following explicit format instructions when given clearly, so use that instead of reformatting the output yourself afterward.

Constraints

Constraints head off the most common failure modes: answers that run too long, use the wrong tone, or include information you specifically don’t want. Stating what to avoid is often more effective than only stating what to include.

Example

Optional, but powerful. A single well-chosen example of the output style you want — sometimes called “one-shot prompting” — often improves output quality more than any other single addition, because it removes ambiguity that words alone can leave open to interpretation.

Three filled-in examples

Example 1 — Marketing

“You are a senior product marketer with experience in B2B SaaS. Write three subject line options for an email announcing a new pricing tier. Our audience is existing customers on the free plan who have used the product for 30+ days. Keep each subject line under 8 words, no exclamation points, no words like ‘exciting’ or ‘game-changing.'”

Example 2 — Coding

“You are a senior backend engineer experienced in Python and API design. Review the following function for bugs and readability issues. This is part of a payment processing service, so correctness matters more than brevity. Return your findings as a numbered list, each with a one-line explanation. Do not rewrite the whole function unless a bug requires it.”

Example 3 — Research and summarization

“You are a research analyst who specializes in making technical material accessible to non-technical executives. Summarize the attached report in a way a busy VP could read in under two minutes. Assume they know the industry but not the technical details. Use a short intro paragraph followed by 4-5 bullet points. Avoid jargon; define any term you can’t avoid using.”

Common mistakes this template helps avoid

• Under-specifying the task. “Write something about our product” gives the model almost nothing to work with compared to a clearly scoped task.
• Skipping context and re-explaining later. It’s faster to state your audience and constraints once up front than to correct the output through several follow-up messages.
• Forgetting to state format. If you need a table, ask for a table — don’t assume the model will guess your preferred structure.
• Only saying what you want, never what to avoid. Constraints are just as useful as instructions for what you want.

Frequently asked questions

Do I need to fill in every section every time?

No. For simple tasks, Role and Task alone are often enough. Add Context, Format, Constraints, and Example as the task gets more complex or the first attempt doesn’t land.

Does this work the same across different AI models?

Yes — this structure isn’t tied to any one provider. All major models respond well to clearly separated role, task, and constraint information, though exact wording sensitivity varies slightly between models.

What’s the fastest way to get better at prompting?

Reuse this template deliberately for a week across different tasks. You’ll start to notice which slots matter most for your specific kind of work, and you can build your own shortcuts from there.

Next steps

Once this structure feels natural, go deeper with the theory behind it in our Understanding Tokens guide (useful for keeping longer prompts efficient), or browse more ready-to-use templates in the Prompt library.