The short answer
Most weak AI answers come from weak prompts, and most AI mistakes go unnoticed because nobody checks. ACED is a short routine that covers both. Write a prompt with an Action, some Context and your Expectation, then Double-check what comes back. Four steps, and you can remember them because they spell a word.
The ACED framework
| Letter | Step | The question it answers | Example |
|---|---|---|---|
| A | Action | What exactly do you want done? | “Explain the difference between on-page and off-page SEO.” |
| C | Context | Who is this for, and what is the situation? | “I am a second-year marketing student preparing for a seminar.” |
| E | Expectation | What does a good answer look like? | “Plain English, under 200 words, a two-column table and a one-sentence summary.” |
| D | Double-check | How will you confirm it is right? | “Which claims should I verify?” Then check them yourself. |
You can write A, C and E in any order. Many people start with the context, then the action. The important thing is that all three are there before you press Enter.
The ACED steps
Before and after
Weak prompt
Tell me about SEO.
The tool has to guess your level, your purpose and the length you want. You will get a generic overview.
An ACED prompt
I am a second-year marketing student preparing for a seminar. (Context)
Explain the difference between on-page and off-page SEO. (Action)
Use plain English, keep it under 200 words, give a two-column table,
then a one-sentence summary. (Expectation)
Finally, list the claims I should verify before using this. (Double-check)
The answer is now focused, the right length, in a usable format, and it points you to what needs checking.
Step by step
Action
Start with a clear verb: explain, compare, summarise, draft, critique, list. One main task per prompt works better than several tangled together.
Context
Say who you are, who the answer is for and why you need it. “For a revision sheet” produces different writing from “for a client email”. Include any background the tool cannot know, such as your module, your audience or the data you are using.
Expectation
Describe what a good answer looks like: length, tone, format, level of detail, and anything to include or leave out. Ask for what you want rather than what to avoid. “Write in short sentences” works better than “do not be wordy”.
Double-check
Treat every answer as a draft. Mark the facts, numbers, names and references, then confirm them against a trusted source. Our guide on checking AI answers gives a five-step routine for this.
A reusable template
Copy this, replace the brackets, and delete any lines you do not need:
Context: I am [who you are] working on [situation].
Action: [what you want done].
Expectation: [length, tone, format, things to include or avoid].
Double-check: List the claims I should verify, and how confident you are in each.
Habits that help
- Iterate. Follow up with “make it shorter”, “add a worked example” or “explain step 3 more simply”.
- Ask it to ask you. For bigger tasks, add: “Before you answer, ask me any questions you need.”
- Keep a prompt you like. Save good prompts in a notes file and swap in new details next time.
Common mistakes
- Being vague about the audience. “Explain this” could mean to a child or a specialist. Say which.
- Skipping the Double-check. A confident tone does not mean the answer is correct.
- Sharing private information. Do not paste personal data, student records or confidential material into tools your organisation has not approved.
Using AI in your studies or teaching
Rules on AI use differ between courses, assessments and institutions. Check your module handbook or your employer’s guidance before using AI for assessed work, and be open about how you used it.
Try it yourself
Pick a topic from your current module. Write a prompt with an Action only, then a second prompt using Action, Context and Expectation. Compare the two answers. Finally, do the Double-check: pick three claims in the better answer and confirm them from a trusted source. Note which step made the biggest difference.