A useful AI prompt states the task, supplies relevant context, sets limits and describes the output you need. Prompting also includes evaluating the response. Asking an AI to sound confident is easy; establishing whether its answer deserves confidence is your job.
Try a teaching example
A better brief gives you something to check.
An original illustrated exercise. The event and work are fictional.
The request
Write a short caption for the fictional Open Book Saturday reading event. Audience: teenagers and parents. Ask readers to contact the library team. Date and venue are not confirmed. Do not invent them, awards or attendance claims. Keep it under 60 words.
A prepared example answer
Bring a book and a curious mind. Open Book Saturday welcomes teenagers and parents. The date and venue are still to be confirmed. Ask the library team for details.
Review: the audience and next step are clear. Unknown details remain unknown. A real response still needs checking against the source brief.
Both answers were written for this exercise. This comparison does not call an AI service or predict what a particular model will return.
Your turn: list the facts you would need before this fictional event could be announced. Then check each sentence against that list.
Why “write something good” gives you so little to judge
Suppose you ask for a social post about a reading event. Without an audience, event details or a purpose, the system has to guess. You may get polished sentences that leave out the information people need.
You can improve the request by describing the job. You do not need a secret sentence or a wall of impressive-sounding instructions. Start with the facts that affect the answer.
An original prompt you can adapt
This example is for a fictional library event. Replace the bracketed fields with information you are allowed to share. If a detail is unknown, leave it unknown instead of asking the system to invent it.
Example to practise
Task: Draft two short social captions for a library reading event. Audience: local teenagers and parents. Facts: [event name], [confirmed date], [location], [how to enquire]. Purpose: help readers understand who the event is for and how to ask about it. Tone: warm, clear and lightly playful. Limits: do not invent prices, guests, awards or availability. Format: two options, each under 60 words, followed by a list of missing details. Review: point out any statement that needs fact-checking.
Give the response a small, useful test
Compare each option against the same criteria. Is the event clear? Are the details correct? Does the tone suit the audience? Is the next step easy to find?
When a response misses the point, give specific feedback. “Keep the event details, remove the exaggerated claim and make the enquiry step clearer” is more useful than “make it better”.
- Accuracy: check factual statements against their original sources.
- Relevance: remove ideas that do not serve the task.
- Completeness: look for missing constraints or necessary details.
- Usability: check length, format, readability and the action requested.
A source-looking link still needs checking
Open important references and confirm that they support the specific claim. Check calculations separately. For decisions with serious consequences, use the appropriate qualified source rather than treating a generated paragraph as authority.
A request to list uncertainties can help your review, but it does not guarantee that the system identifies every error. Keep the factual check outside the prompt as a separate step.
Use only information you are allowed to share
Avoid placing private client files, student details, passwords or confidential work into an AI service without the required permission. Check the tool’s data settings and the rules that apply to the material.
If you use an AI draft in a portfolio, describe what you asked for, what you checked and what you changed. The useful evidence is your decision-making, not a screenshot of an enormous prompt.
Save the workflow, not just the winning paragraph
Keep the brief, the evaluation checklist and a short note about the revision. Try the workflow on a second fictional example. A process you can explain and repeat is more valuable than one answer that happened to sound good.
Text-based prompting can be practised without programming. Building software integrations or automated systems is a separate technical step; check which activities a course actually teaches.
Official resources for further learning
The exercises and examples above were created for this guide. These references provide additional background; a link does not imply an endorsement or training partnership.
Found an unclear explanation? Send Upnova a correction or question. Read our content and corrections approach.
Put the idea into practice.
Explore the ai prompting learning path, or ask how the current batch teaches and reviews this kind of work.