Audience, constraints and response format
Specify audience, constraints and a checkable response format for a task.
Defining the audience
"Explain photosynthesis" produces very different answers for a 9-year-old, a biology undergraduate and a farmer deciding on greenhouse lighting. Tell the model:
- who the reader is ("small-business owners with no accounting training");
- what they already know ("they understand revenue and costs but not depreciation");
- what they need to do next ("decide whether to buy or lease a van").
The last point is often the most useful: it makes the model focus on what matters for the reader's decision.
Constraints the model can follow
Constraints are rules the output must obey. Good constraints are specific and checkable:
| Weak | Checkable |
|---|---|
| "Keep it short." | "Maximum 120 words." |
| "Don't be too technical." | "Avoid jargon; define any technical term in brackets the first time it appears." |
| "Use reliable information." | "Use only the information in the article below. If something is not covered, say 'not covered in the source'." |
| "Be balanced." | "Give two arguments for and two against, with equal length." |
Common constraint types: length, vocabulary/reading level, sources allowed, content to include or exclude, tone, language or spelling convention, and what to do when unsure.
A note on length: models count words and characters only approximately. If an exact limit matters (a form field, an SMS), check the output yourself.
Specifying the response format
The format is the easiest part to specify and the easiest to check. Options include:
- Lists — numbered steps, bullet points, a fixed number of items.
- Tables — name the columns: "Columns: Option | Cost | Main risk".
- Templates — give the exact skeleton: "Subject: … / Greeting: … / Body (3 sentences): … / Sign-off: …".
- Sections with headings — "Use the headings Summary, Evidence, Open questions."
- Structured data such as JSON — useful when another program or spreadsheet will read the output (Module 9).
Showing the format is often clearer than describing it. A one-line example row of a table can be enough.
Putting it together
## Task
Explain the difference between a fixed-rate and a variable-rate loan.
## Audience
Adults taking out their first loan; no finance background; they need to decide which type to ask their bank about.
## Constraints
- Under 200 words.
- No jargon without a plain-English explanation.
- Do not recommend a specific product or give personal financial advice; suggest questions to ask the lender instead.
## Output format
A two-column table (Fixed rate | Variable rate) with rows: How the rate works, Monthly payments, Main advantage, Main risk. Then three questions to ask the lender, as a numbered list.
Stage 1 assessment
Today's completion rule is the Stage 1 assessment, covering Modules 1–3. It unlocks when you have completed days 1–5. Objective questions are scored automatically; the written question is assessed against a published rubric with AI assistance, and you can request a human review. You can retake it.
Practice exercise
Before the assessment, practise on one task:
- Write the same request for two different audiences (e.g. a teenager and a manager). Change only the audience section.
- Add three checkable constraints and a named-column table format.
- Run both versions if you can and record in your log how the audience changed vocabulary, examples and focus.