The four Ds of AI fluency

The four Ds at a glance

Delegation

Decide what you will do, what AI will do, and how you will work together.

You do not have to make one decision for an entire task. You might complete some parts yourself, ask AI to support others, or work without AI. A useful delegation decision considers the result you need, the abilities you want to develop, and the help a tool can actually provide.

Know your goal

Identify what a good result looks like and what you need to learn or practise. When AI carries out part of a task, you may miss the practice that part would have given you. Before handing it over, ask what practice you would lose and whether you need it.

For example, develop your own argument before asking AI to raise objections. Attempt a problem before requesting a hint. Read and summarize a source yourself before asking AI to suggest points to revisit. These choices keep the intended thinking with you while allowing useful support.

Know your AI tool

Check whether the tool can handle the particular job and which features you can access. Results can vary across tasks, and limits on messages or file uploads may affect your plan. Consider what context the tool needs and what you will have to check, change, or finish yourself. Try a small example if its suitability is uncertain.

Divide the work

Break the task into parts and assign a clear role to yourself and to AI.

Description

Communicate what you want AI to do and how you want to work together.

A prompt is a request or instruction you give an AI tool. Description includes that opening request, the relevant material you provide, and the corrections you make as the work continues. You do not need special phrases or a long message. Supply the details that matter for the task. Clear instructions help communicate expectations, but they do not guarantee a correct response.

Describe the result

Explain your goal, who the result is for, and what a useful response would look like. Provide the relevant context: the topic, the intended level, what you already understand, and where you need help. The framework calls this product description.

Specify the output you want and the limits that matter. For example, ask for three rehearsal questions based on your speaking notes and explain that AI must not rewrite your explanation. If format matters, provide an example. Make clear which sources it should use; do not assume it can see other files, pages, or earlier conversations. Supply only material you are allowed to share.

Guide the work and interaction

Process description explains how the work should proceed. You might ask AI to propose a slide layout before creating slides, or to ask one question and wait for your answer before giving feedback. For a larger task, identify the points at which you want to review or make a decision.

Performance description explains how AI should interact with you. Instead of only asking it to act as a tutor, specify the behaviour you need: use plain language, explain unfamiliar terms, and point out gaps in your reasoning. If a tool can act independently, define stopping points and check its permissions. Written instructions alone do not guarantee that it will stay within your limits.

Adjust as you go

Read the response and name what works and what should change. "Keep the everyday example, but explain the technical terms" gives more useful direction than "Try again." If you are unsure what information is needed, ask AI to pose clarifying questions. If the conversation becomes confused, start a fresh one with the revised request and necessary context.

I'm preparing a five-minute presentation for classmates. Use my speaking notes to ask three questions about how the concept applies to my example. Ask one at a time and wait for my answer. Don't write answers or rewrite my notes. After each answer, identify anything unclear and ask me to explain it further.

Ask only the first question and wait. Don't show an example answer unless I request it after trying.

Discernment

Evaluate AI's work and the way you are working with it.

A clear, confident response can still be false, unsupported, incomplete, or directed at a slightly different question. Even accurate information may not serve your purpose. Discernment involves checking the result, examining the reasoning and assumptions behind it, and considering whether the interaction is helping you.

Check the result and evidence

Compare the response with your actual goal and instructions. Are the facts and examples accurate? Do the conclusions follow from the evidence? Verify important parts against something outside the AI response. Find an original source and check that it supports the specific claim. Recalculate a result and check its units. For a slide deck or webpage, inspect the content and try the links, layouts, and other features people will use.

Asking "Are you sure?" is not verification. Agreement from another AI tool is not sufficient either. Give more attention to errors that could change a conclusion or affect someone's decisions. If you cannot judge an important point, seek evidence or help before relying on it. "I can't tell yet" is a useful conclusion.

Check the assumptions

Look for missing conditions, exceptions, alternative explanations, and perspectives. Examine your own request as well as the response. "Why are evening meetings best for our group?" assumes the answer and may overlook members with jobs or caring responsibilities. "What should we consider when choosing a meeting time?" leaves room to investigate. You still need to ask the people affected about their availability.

Ask whether you are seeking evidence or agreement. AI can suggest other perspectives, but it cannot speak for the people affected. Familiarity with a topic helps you spot problems, yet your own understanding can also contain gaps.

Check how you are working together

Is AI providing useful feedback or simply praising your ideas? Are you doing the thinking you intended to practise? Consider whether you have explained the task clearly and whether AI is appropriate for this part of it. You may need to revise the request, consult a source, attempt the work yourself, or stop using AI for the task.

Diligence

Take responsibility for your AI use from start to finish.

Diligence includes what you put into a tool, how you explain its contribution, and what happens when you share or act on the result. Check the rules for the particular task.

Take care before sharing

Creation diligence concerns the tool, the material you provide, and how the work is made. Check whether the service and account are approved for the information involved. An available account does not mean that every kind of data may be uploaded. Share only relevant material you are permitted to provide.

Personal notes may contain someone else's information or work. Group messages, interviews, and workplace or research material may have additional restrictions. Removing names may not protect people if other details identify them, and the material may remain confidential. Use a fictional example when permission is uncertain. Follow requirements for other people's writing, images, and data; AI involvement does not remove them.

Explain AI's role

Transparency diligence means explaining AI's contribution honestly. Name the tool, the specific help it provided, and what you checked or changed. "AI helped" says little. "I used AI to suggest headings, selected three, and wrote the paragraphs myself" explains the contribution. Use that wording only if it accurately describes your process.

Keep relevant drafts and checking notes, stored appropriately. Distinguish completed checks from planned ones. Follow any required disclosure and citation conventions: a disclosure describes the contribution, while a citation follows a reference format. You may need both.

Take responsibility before sharing the result

Deployment diligence means deciding whether work is ready to submit, publish, or use. Apply Discernment to important claims and sources, and make sure you understand what you are presenting. Consider who might be affected, whether private information is exposed, and whether assumptions or recommendations unfairly exclude people. Give collaborators a chance to review work associated with their names.

Resolve important problems before sharing or seek help. Use AI purposefully, recognizing that it also requires electricity and equipment. Stop revising when the result meets the need.

Source and licence