PREPARED
Propose · Role · Explicit · Presentation · Ask · Rate · Emotions · Diversity — Eight levers rather than eight steps, and the only common framework with a bias check and a self-review built in.
Two versions circulate
This framework has been published three times under three names. It began as PREP, was extended to PREPARE, and its author’s 2024 article presents the eight-lever PREPARED shown here, with Propose replacing Prompt and Presentation replacing Parameters. Institutional guides still teach the seven-letter PREPARE. If you share a prompt built on it, say which version you mean.
Use it for
Producing assessable artefacts where quality criteria matter and you cannot easily judge them yourself — rubrics, training materials, internal guidance, policy drafts.
Avoid it for
High-volume automated pipelines. Ask needs a human to answer the questions, and Rate produces self-scores that are not independently trustworthy — both break unattended.
The slots
Each letter is a question the prompt has to answer.Propose
What is the command, precisely?
One unambiguous instruction that sets the scene. Avoid vague verbs like "help with".
Example: Create an assessment rubric for our new-starter compliance training module.
Role
What hat should it wear?
Name the profession and the specialism it should reason from.
Example: Act as a learning and development manager who designs competency assessments for financial-services onboarding.
Explicit instructions
What are you assuming it already knows?
Spell out the domain specifics, the audience and the method you want used.
Example: Cover anti-money-laundering basics, data protection and the whistleblowing route. The audience is non-finance graduate hires in their first fortnight.
Presentation
What shape, tone and reading level?
Format, length, register and reading age — the controllable output parameters.
Example: A table with four performance levels per criterion, written at a reading age of 14, no more than 600 words total.
Ask Optional
What does it still need from you?
Instruct it to put clarifying questions to you, in bullets, alongside or before the draft.
Example: After the draft, ask me five bullet-point questions about our escalation process that would sharpen it.
Rate Optional
How good does it think this is, and on what grounds?
Ask for a score plus the criteria used and the evidence for each. The criteria are the useful part; the number is not a quality gate.
Example: Rate the rubric out of ten and set out, in a table, the criteria you scored against and your evidence for each.
Emotions Optional
Does stakes-framing help here?
A short line on why it matters. Treat this one as optional and unproven — the research behind it is weak, and on current models the effect is largely gone.
Example: This rubric will be used for every hire this year, so accuracy matters.
Diversity Optional
Whose perspective is missing?
Ask it to name underrepresented viewpoints or blind spots in its own output.
Example: Identify any groups whose circumstances this rubric would disadvantage — for example, staff with English as a second language.
Strengths
- Rate forces the model to surface the criteria it applied, which often exposes a misread brief faster than reading the output does.
- Ask turns a vague request into a specified one without the user needing to know what was missing — the highest-leverage slot here for non-expert users.
- Diversity is, as far as this research found, the only bias check promoted to a first-class slot in any widely-taught framework.
Weaknesses
- Emotions is the weak link. It rests on research whose effects are small, task-dependent and inconsistently replicated, and on current models the gain is largely gone. Treat it as optional and unproven.
- Self-rating is not evaluation. Models are poorly calibrated about their own output and tend to score generously, so Rate is a prompt for reflection rather than a quality gate — and readers will over-trust it.
- Unstable naming: three published variants exist with two letters redefined between them, which makes it genuinely hard to cite cleanly.
Sources and grounding
- 1The perfect ChatGPT prompt doesn’t existDan Fitzpatrick, Forbes (10 August 2024)The author’s own article setting out all eight levers with a worked example.
- 2AI literacy: promptingRadboud University LibraryTeaches the earlier seven-letter PREPARE version and confirms who developed it.