# Writing effective AI prompts

2026-09-21

## The idea

The quality of what an AI system gives you depends heavily on how you
ask. A good prompt makes your **context**, your **task**, and your
**expectations** explicit instead of leaving the model to guess. This
page offers two templates:

- **Template 1 — the quick skeleton:** four fields for everyday use.
- **Template 2 — the meta-prompt:** a structured, iterative tool for
  when the output really matters (e.g. a complex analysis for your term
  paper).

The examples use Claude, but both templates work with any comparable
assistant (ChatGPT, Gemini, Copilot …). Adapt the wording to the tool
you use.

> [!IMPORTANT]
>
> ### A better prompt does not replace your judgement
>
> A well-crafted prompt gets you a more useful *draft* — it does **not**
> make the output true. You must still verify every claim, check every
> citation, and revise the text into your own. **Keep your prompts**:
> you will need them for the “Prompt & iteration strategy” row of the
> [AI use disclosure template](../ai-disclosure/index.qmd).

> [!IMPORTANT]
>
> ### Keep your prompts
>
> You will need them for the “Prompt & iteration strategy” row of the
> [AI use disclosure template](../ai-disclosure/index.qmd). Also, it is
> worth re-using good prompts and not always start from scratch. If you
> are a heavy AI user, it might be worth creating your own prompt
> repository.

## How to read the templates

The templates use two kinds of placeholders:

- `[…]` — **replace this** with your own content.
- `{…}` — an **instruction to the AI** describing what it should produce
  there.

Do not leave the literal brackets in your final prompt.

## Template 1 — the quick skeleton

For most everyday requests, four fields are enough:

The section labels (**Context**, **Task**, **Format**, **Constraints**)
stay as they are — they give the prompt its structure. Only the `[…]`
placeholders get replaced.

``` markdown
## Context
I am a [role] working on [task/project]. The audience is [who],
and the relevant background is [key facts, constraints].

## Task
Please [action verb] a [deliverable] that [core goal].

## Format
Give the answer as [text / bullet points / a table / …],
about [length], structured as [sections you want].

## Constraints
Focus on [scope]. Avoid [what to exclude].
If you are unsure of a fact, say so rather than guessing.
```

### Worked example

``` markdown
## Context
I am a Master's student in International Management writing a term paper. My
reader is my course instructor. The paper analyses how institutional theory
explains the market entry of a mid-sized German firm into Southeast Asia.

## Task
Please help me draft a one-paragraph statement of my research question that is
specific and answerable with a single case study.

## Format
One paragraph, about 120 words, in academic English.

## Constraints
Focus only on institutional theory (not resource-based views). Do not invent
sources. If a claim needs a citation, mark it [cite].
```

## Template 2 — the meta-prompt (for demanding tasks)

When the stakes are higher, use the prompt below to make the AI **help
you build a better prompt** through an iterative process. Paste it into
a new chat, then describe what you want to accomplish.

``` markdown
You are a prompt-engineering assistant. Your goal is to help me create a highly
effective prompt through an iterative process. When I describe what I want to
accomplish, create a structured prompt using the framework below, then offer
refinements and ask clarifying questions.

## Prompt:
{Create the most effective prompt based on my request, phrased in the first
person as if I were directly asking the AI. Use the framework below, adapting
sections to the specific use case:

**Context**
- I am a [role] working on [project/challenge].
- My audience is [stakeholders] with [expertise level] in [topic area].
- Relevant context: [domain, constraints, timeline, background].
- I want you to act as [AI role with specific expertise and perspective].

**Task**
Please help me [action verb] a [deliverable] that addresses [core question].

**Success criteria**
The response will be successful if it [measurable outcomes that matter to the
audience].

**Desired output**
Format: [text/bullets/tables/…]
Structure:
1. [Section 1] (approx. X words) — [purpose]
2. [Section 2] (approx. X words) — [purpose]
3. [Section 3] (approx. X words) — [purpose]

**Style & tone**
- Use [formal/conversational/technical] language for [audience].
- Include [examples, data, frameworks, analogies …].

**Constraints**
- Length: [word count or limits].
- Scope: focus only on [region/timeframe/domain].
- Exclude: [topics or approaches to avoid].

**Examples & references**
Here is the kind of content I am looking for: [brief example].
Please draw on: [methodologies, frameworks, or sources].

**Quality checks**
Before finalising, verify you have: [validation criteria], provided reasoning for
key recommendations, and addressed limitations.

**Uncertainty instruction**
If you are uncertain about facts, data, or domain practices, say so clearly
rather than making assumptions.}

## Possible Additions:
{Suggest three specific, actionable additions that could improve the prompt:
A. […]
B. […]
C. […]}

## Questions:
{Ask three targeted questions whose answers would most improve the prompt:
1. [context/requirements]
2. [audience/stakeholder needs]
3. [depth/scope/methodology]}

## Instructions for next steps:
After I review your Prompt, Additions, and Questions, I will reply with the
additions I want and answers to your questions. Integrate my feedback while
preserving the working parts. We continue until the prompt fully meets my needs.
```

### Quick reference — common output structures

Reuse these in the **Desired output** section:

- **Summary report:** Problem → Analysis → Recommendations → Next steps
- **Analytical document:** Situation → Options → Evaluation →
  Implementation plan
- **Action memo:** Context → Recommendation → Rationale → Required
  resources
- **Presentation outline:** Key message → Supporting evidence → Call to
  action
- **Tutorial/guide:** Overview → Step-by-step → Examples →
  Troubleshooting

To start Template 2, open a new chat, paste the block above, and answer
the assistant’s question:

> *What specific challenge, analysis, or deliverable would you like your
> prompt to address?*