Ask an AI tool to "make a rubric for my design class" and you'll get a rubric. It will have four levels, some criteria about creativity and presentation, and nothing you could hand to your actual students without rewriting most of it.

The tool is guessing. A prompt is just the message you write to the AI, and when that message is vague, the model fills the gaps with the most average answer it can find. The fix isn't learning a secret syntax. It's writing the kind of brief you'd give a capable colleague who has never met your students.

The four parts of a useful prompt

Almost every prompt that produces something usable has four ingredients:

• Role: What it answers: Who should the AI act as?; Example: "Act as a graphic design instructor with experience teaching first-year university students."
• Task: What it answers: What exactly should it produce?; Example: "Create a 90-minute lesson plan."
• Context: What it answers: For whom, and under what conditions?; Example: "Second-semester students, 25 per group, one projector, no prior branding experience."
• Format: What it answers: What should the result look like?; Example: "A table with columns for objective, teacher activity, student activity, materials, time and assessment."

Role sets the vocabulary and level of expertise. Task keeps it from wandering. Context is where most of the quality comes from, because it's the part only you know. Format saves you from reformatting paragraphs into tables by hand.

Leave any of them out and the model guesses. Include all four and you're mostly editing, not rewriting.

A complete example

Here's what that looks like put together:

Act as a graphic designer with experience teaching at university level.

I need a 90-minute lesson plan on "color theory applied to branding".

My students are in their second semester. They know the basics of

design history but have never applied color theory to a real project.

The group has 25 students in a classroom with a projector.

By the end of the session, each student should have a short practical

exercise ready to hand in.

Give me the plan as a table with these columns: learning objective,

teacher activity, student activity, materials, estimated time,

assessment criterion.

Nothing in there is clever. It's just specific. That's the whole trick.

Four mistakes that produce generic output

Being vague. "Write something about design" gives the model nothing to work with. Name the topic, the angle and the purpose.

Not asking for a format. "Make a rubric" might return a paragraph, a bullet list or a table. If you want a table with four levels, say so.

Stopping at the first try. The first response is a draft. Treat it like one. Follow up with what's missing: "Make the criteria more specific to typography", "Cut this to one page", "Add an example of excellent work for each level".

Skipping the student level. "First-semester students" and "graduate students" produce very different results, even for the same topic. The model adjusts vocabulary, depth and assumptions based on what you tell it.

Three techniques worth keeping

Start from what you don't want

If you're struggling to describe the result you want, ask for the opposite first. "Write a deliberately bad set of instructions for this exercise: vague, too long, full of jargon." Then: "Now rewrite them to fix every one of those problems." Seeing the failure spelled out often makes the target clearer, for you and for the model.

Iterate in small steps

Instead of one giant prompt, build in rounds. Ask for a draft. Then ask for one improvement at a time: more concrete materials, a 15-minute warm-up at the start, a version shortened for a slide. Each round is easy to check, and you can back up if a change makes things worse.

Give it two roles at once

"Act as both a working graphic designer and a design educator" pushes the model to balance two concerns: whether an exercise reflects real practice and whether it teaches something. You'll often get activities that are more grounded than when you ask for either role alone.

One more that isn't in most prompting guides for teachers but works very well: show an example. If you have a rubric or exercise you like, paste it in and say "Follow this structure and level of detail." Models are very good at matching an example.

Five reusable templates

Copy these, replace the parts in square brackets, and delete anything that doesn't apply. They're written in plain language on purpose, so they work with cloud tools and with local models.

Rubric

Act as a design instructor with experience in competency-based assessment.

Create a rubric to evaluate [assignment or topic].

Students are [level, e.g. third-semester graphic design undergraduates].

Include [number] criteria, each with four levels: insufficient, basic,

satisfactory, excellent. Each level needs a clear description of 2 to 3

lines that a student could use to check their own work.

Return it as a table.

Practical exercise

Act as a senior graphic designer who also teaches.

Create a [duration]-minute practical exercise on [topic].

Students already know [prior knowledge].

Include: clear instructions written to the student, required materials,

a description of an example outcome, and how it will be assessed.

Format it as a numbered list.

Reading or handout

Act as a design writer.

Write a [length]-word text on [topic] for [level] design students.

Tone: [informative / conversational / formal].

Include at least two concrete, practical examples.

End with one question students can discuss in class, not a summary.

Slide content

Act as a presentation designer.

Write the content for [number] slides on [topic] for [audience].

For each slide give: a title, 3 to 5 short key points, and a suggestion

for an image or diagram.

Number the slides and separate them clearly.

Feedback on student work (anonymized)

Act as a supportive but honest design instructor.

Below are my rough notes on a student project. Rewrite them as feedback

addressed to the student: start with what works, then give no more than

three specific improvements, each with a concrete next step.

Keep it under 200 words. Do not invent details that are not in my notes.

My notes:

[paste notes, with names and identifying details removed]

That last template brings up something worth a separate note.

What not to paste

Prompts are a lot of copying and pasting, and it's easy to paste more than you meant to. Before you send anything to a cloud tool, remove student names, grades, emails and anything else that identifies a person. "Student A" works just as well for the AI. If you regularly need to work with real student data, that's a good reason to set up a local model on your own computer, where the text never leaves your machine.

Check before you use it

AI output sounds confident whether it's right or not. Read everything before it reaches students. Check facts, dates and names of designers or works, because models will occasionally invent them. Check that time estimates in lesson plans add up. And check that the result sounds like you, since students notice when the voice of a handout suddenly changes.

Try this

Pick one thing you'll need next week: a rubric, an exercise, a short reading. Write the prompt once using the four parts. Then write it again with the "show an example" technique, pasting in something you've made before that you liked. Compare the two results. If the second one needs less editing, that's the habit worth keeping.