Writing good prompts: The 7 elements of effective AI communication

← Back to resources

Effective AI communication starts with a clear task.

A good prompt describes the goal, material, desired result and review criteria. Universities, research support services, grants offices, foundations, associations and non-profit organisations also need to consider confidentiality and institutional rules.

The following recommendations help you formulate these instructions. Professional review and responsibility remain with the people in charge.

The 7 elements of effective AI communication

The following seven elements help structure a prompt:

Role · Task · Context · Material · Format · Boundaries · Review criteria

Adapt the model to the task. A simple request often needs only a few of these details.

Less helpful:

Make this text better.

More helpful:

Revise the following text for an internal announcement to a university administration. Aim for a clear, concise and factual version. Do not add new facts. Flag statements that need subject-matter review.

The second prompt specifies the audience, task, tone and review criteria. This makes the result easier to assess.

Why prompting is part of AI literacy

Here, AI literacy means the ability to use AI meaningfully, critically and responsibly in your own work. ^[1]

This involves six questions:

• Which task should AI support?
• Which information may be entered?
• Which tool is appropriate for the situation?
• What should the result look like?
• What risks or uncertainties are there?
• Who reviews and takes responsibility for the output?

Prompting turns these considerations into a specific set of instructions.

The seven elements in detail

1. Role: Which perspective should the AI take?

Use role instructions to describe the perspective and tone you want. Check subject-matter claims independently.

Act as a communications adviser for a Swiss university.

Specify the perspective you need, for example ‘communications adviser for a university’.

2. Task: What exactly needs to be done?

The task is the core of the prompt. Start with a clear verb: analyse, simplify, structure, review, compare, draft or write.

Analyse the following draft for clarity, repetition and missing review points.

A specific task makes it easier to assess whether the result meets the goal.

3. Context: Who is the result for, and what is its purpose?

Describe the relevant organisational context: audience, purpose, tone, length, prior knowledge and institutional constraints.

Include only information needed for the task.

4. Material: Which sources should the AI use?

Clearly separate texts, notes or data from the instructions:

Use only the material between the dividing lines. Do not add external facts.

For example, use dividing lines:

---
[Insert material]
---

This separation defines the intended source material. It guarantees neither that the AI will use only those sources nor that they will remain confidential. Establish beforehand whether the material may be entered into the tool.

5. Format: What should the output look like?

Specify the format you need, such as a table, an email, a summary, review questions or a comparison of options.

Present the result as a table with the columns Benefit, Risk, Points to clarify and Next step.

The format should suit the next step in the work.

6. Boundaries: What should the AI avoid?

State boundaries, for example:

• Do not add new facts.
• Do not invent sources.
• Flag uncertainties.
• Do not provide a legal assessment.
• Identify missing information.
• Avoid exaggerated promotional language.

These instructions are not technical safeguards and do not reliably prevent errors. An instruction such as ‘Do not use confidential information’ cannot replace reviewing the material or obtaining approval for the tool.

The absence of uncertainty flags is no proof of accuracy either. Check important information independently against reliable sources. ^[2]

7. Review criteria: How will quality be assessed?

Define how you will assess the output: subject-matter accuracy, clarity, appropriate length, verifiable statements and clearly identified uncertainty.

Distinguish clearly between facts, assumptions and recommendations.

Then check for yourself whether these criteria have been met.

Prompting involves two roles: architect and editor

Two working roles help with quality assurance:

As the architect, you design the instructions using the details relevant to the task.

As the editor, you review terminology, statements, limitations and suitability for your organisation. Good prompts support this review; they do not replace it.

Good prompts often develop in stages

For demanding tasks, it is worth handling individual steps separately:

• Clarify the task
• Structure the material
• Produce a first draft
• Critically review the draft
• Revise
• Shorten or format the final version

Review intermediate results and adjust the instructions as needed. ^[2]

Spoken prompts

Dictating a prompt can help capture context and examples. Afterwards, check whether the task and priorities are clear.

Spoken input is subject to the same requirements for tool suitability and confidentiality as typed input.

Three example prompts for professional use

Adapt these examples to your own work. Enter only material approved for the tool you are using.

Example 1: Clarify the proposed use of AI

I want to assess whether AI could usefully support a work task. First, ask me up to five questions about the goal, audience, material, confidentiality, desired output and necessary human review. Only then recommend whether AI could help, what risks are involved and what a suitable initial prompt might look like.

Example 2: Revise a text for a university administration

Act as a communications adviser for a Swiss university. Revise the following text for an internal announcement to administrative staff. Aim for a clear, factual, concise and actionable text. Do not add new facts. Flag statements that need subject-matter or institutional review. First provide a short assessment of weaknesses, followed by a revised version.

Example 3: Prioritise AI use cases in a non-profit organisation

Act as an adviser on responsible AI use in a non-profit organisation. Prepare an initial prioritisation of potential AI use cases. Consider benefits, risks, data sensitivity, feasibility and the review effort required. Present the result as a table: Use case, Benefits, Risks, Points to clarify, Next step. Explicitly identify assumptions. Final prioritisation and approval remain with the responsible people in the organisation.

Common prompting mistakes

Unclear instructions are one possible cause of weak AI results. Common mistakes include:

• The task is too vague.
• The audience and purpose are missing.
• The desired format is not specified.
• The AI is asked to do too many things at once.
• Boundaries and review criteria are missing.
• Uncertainties are not identified.
• Data protection and confidentiality are considered too late.
• The wording of the output is accepted without reviewing its substance.

What good prompts can improve

Good prompts can help simplify texts, prepare meetings, compare options, develop review questions or produce initial drafts.

In universities and non-profit organisations, possible applications include internal announcements, draft FAQs, workshop preparation, grant administration and initial use-case analysis.

What good prompts cannot resolve

Even a very good prompt does not automatically make AI output true, current, complete or legally correct.

AI-generated texts can sound convincing while containing false or fabricated information. Such fabricated details are often called hallucinations.

Subject-matter, legal and data protection checks, as well as institutional approvals, remain necessary whenever the task requires them, regardless of the prompt.

Responsible use starts before the prompt

Before entering information, establish:

May this material be entered into this tool?

Consider the following:

• What type of material is being used?
• Does it contain personal data or confidential information?
• Which tool will be used?
• What task should the AI perform?
• What output will be produced?
• Who will review the result?
• Which rules apply in the organisation?

The four-question check ‘Material × Tool × Task × Output’ structures this assessment: Can I use AI for this? The four-question check for responsible AI use.

The AI Traffic Light provides initial guidance for specific situations.

Use the AI Traffic Light

Shared use within organisations also requires suitable tools, rules, training and clear responsibilities. ^[3]

For an explanation of how good prompts fit into a broader AI literacy workflow, see AI in research support: From individual prompting to shared AI workflows.

When a prompt is no longer enough

For recurring tasks or the use of curated organisational knowledge, a specialised AI assistant, chatbot or voicebot may be useful.

One possible technical foundation is retrieval-augmented generation (RAG), which retrieves information from selected sources to inform the answer. ^[4]

The quality of those sources and review of the result remain essential.

An internal prompt mentor could, for example, help staff turn vague requests into clear, reviewable instructions.

The choice of solution depends on the task, data and review effort required.

Conclusion

Good prompts structure tasks. The seven elements provide an adaptable guide.

Whether a result can be used depends on a review against subject-matter and institutional requirements.