Plan, Verify, and Explain

For a complex task, improve the prompt by defining the outcome, evidence, constraints, and checks. Ask for a useful explanation of the answer—not a transcript of hidden reasoning.

Start with the task and success criteria

Use the practical prompting pattern to define the goal, context, sources, criteria, constraints, and output. This guide adds the next step: deciding what evidence would let you check the result.

For a complex task, ask for a short plan when it creates a useful approval or review point. For example, review an assessment blueprint before generating its questions. A model’s plan and self-check remain suggestions to inspect.

Ask for checkable work

The right kind of explanation depends on the task:

Ask for a concise rationale when it helps a reader judge the answer. Do not treat verbosity as rigor.

Worked example: verify a lesson plan

After drafting a lesson plan from approved outcomes and sources, request a check that exposes omissions:

Review the draft against the supplied learning outcomes and class constraints.
Return a table with:
- each outcome and the student action that practises it;
- the observable learning check;
- the segment and time allocation; and
- any missing prerequisite, unsupported claim, or participation barrier.

List the segment durations and their total. If an outcome has no activity or
check, mark it as missing rather than claiming alignment. Distinguish proposed
fixes from content already present in the plan.

Then add the durations yourself, inspect the proposed outcome mapping, and check factual claims against the course sources. A table containing a tick for every row is not sufficient: it must point to actual evidence in the plan. Use this check after drafting a lesson plan, before preparing the materials or teaching the session.

Decompose only where it helps

Break a task into stages when later work depends on earlier results. For example:

1. extract the relevant learning outcomes from an approved course document;
2. propose assessment items mapped to those outcomes;
3. review coverage and difficulty; and
4. revise the set after the educator’s feedback.

This creates useful checkpoints. It does not make an unreliable source reliable, and asking the same model to “double-check” itself is not independent verification.

Make uncertainty actionable

Replace “do not make assumptions” with instructions the model can follow:

List assumptions that materially affect the result. If an essential input is
missing, ask me before continuing. If a minor detail is missing, choose a
reasonable default, label it, and explain how changing it would affect the
answer.

Test and iterate

Evaluate prompts on several representative examples, including difficult and edge cases. Record the model and settings used, the failure patterns, and the human corrections. Re-test important workflows when the model, tool, source material, or task changes.

References & Footnotes