Copilot → Colleague

HOW TO GET
BETTER RESULTS
FROM AI

The PINPOINT Prompt Engineering Framework v1.2

Play the PIN Prompt Challenge Download the framework PDF

Start here

Before you blame the model,
inspect the request.

Better prompting is not about finding magic words.

It is about specifying the work.

Everyday work

Start with the smallest amount of structure that works.

P — Purpose

What do you want to achieve, and for whom?

I — Introduce Context

What does AI need to know? What is the source of truth?

N — Narrow the Ask

What exactly should a good result look like?

Practical rule of thumb: PIN is enough for roughly 80% of everyday prompting.

When consequence rises

PINPOINT it.

Add structure when the work needs greater control, verification or confidence.

PPurpose

IIntroduce Context

NNarrow the Ask

PPersona, Ethics, Values & Role

OOutline Examples

IInduce Verification

NNormalise & Standardise

TTweak & Test Again

The first three

Specify the task.

Purpose

Define the outcome and audience.

Introduce Context

Provide relevant facts, files, data and the source of truth.

Narrow the Ask

Set format, scope, success criteria, boundaries, permissions and stop conditions.

Source ruleIf sources conflict, say what has authority—or flag the conflict.

Original

Summarise this report.

→

PIN

Summarise this quarterly report for the executive committee. Focus on the three issues requiring a decision. Use the attached report as the source of truth. Five bullets and a recommendation.

Add judgement & clarity

Persona and Examples

Persona, Ethics, Values & Role

Set a useful professional stance, responsibilities, principles and boundaries. Persona is not cosplay.

Outline Examples

When showing is easier than explaining, show what good looks like—and, where useful, what bad looks like.

Examples are optional. Use them when they remove ambiguity.

Induce Verification

What needs to be checked before this result can be relied upon?

Evidence & sources

Assumptions & uncertainty

Facts & calculations

Constraints & Purpose

The level of verification should match the consequence of getting it wrong.

Low consequenceBasic factual check

Medium consequenceEvidence + assumptions + uncertainty

High consequenceEvidence + uncertainty + escalation + human approval

Evidence vocabulary

Make uncertainty visible.

[Verified Source]
Direct support from authority/source of truth

[Inference]
Derived from evidence, not explicitly stated

[Assumption]
Treated as true without sufficient evidence

[Unverified]
Cannot currently be substantiated

Uncertainty can be acceptable when visible. Material uncertainty needs escalation.

Use in a prompt

“For each material claim, label it as [Verified Source], [Inference], [Assumption], or [Unverified]. Cite the source where available. If an unverified or assumed point could materially change the conclusion, flag it for review rather than guessing.”

Trust needs two controls

Verification ≠ Acceptance

AI-side

Induce Verification
Ask AI to check evidence, assumptions, uncertainty, consistency and constraints.

Human-side

Normalise & Standardise
Inspect, accept, reject, correct or standardise before use.

AI verification does not remove human accountability.

Human review

Is it fit for use?

1. Did it achieve the Purpose?

2. Is it grounded in the source of truth?

3. Did it follow constraints and permissions?

4. Any unsupported claims or material uncertainty?

5. Is it suitable for the intended decision or workflow?

Then: accept, reject, correct or standardise.

Iteration

Tweak & Test Again

Test → Learn → Change → Retest

One good answer isn't enough.

Review what worked. Identify what missed the mark. Then change one meaningful variable at a time — the context, constraint, example, or verification — and test again.

For repeatable work, test across normal cases, edge cases, and failure cases. Look for consistent performance — and learn where it breaks.

The goal isn't a prompt that worked once. It's a prompt you know when to trust.

Purpose→…→Verify→Review→Tweak↺

The full journey

PIN

Define the task

PINPOINT

Engineer the request & result

BLUEPRINT

Design the AI colleague

PIN the task.
PINPOINT the result.
BLUEPRINT the colleague.

Takeaway

Don’t make every prompt longer.

Make delegation clearer.

Start with PIN. Add the PINPOINT elements the work actually needs.

Play the PIN Prompt Challenge

HOW TO GETBETTER RESULTSFROM AI