Meta prompting
Structure & output
Level 2 · Advanced reasoningAsk the model to design or improve the prompt strategy itself before tackling the task.
Meta prompting here means asking the model to design the best strategy or structure to solve a new task — not rewriting or improving a prompt you've already written, and not generating multiple candidate phrasings to compare before you've committed to any of them. If you have an existing prompt you want to improve, the Prompt Optimizer is the better fit; if you're choosing between several fresh phrasings for a prompt you haven't written yet, see Automatic Prompt Engineer (APE) instead. Meta prompting shifts the focus from the task's specific content to the structure of the prompt itself: instead of just asking for the result, you ask the model to first propose the best strategy, structure or sequence of steps to solve the problem, and then follow that plan. It treats prompting itself as something worth reasoning about, not just a fixed instruction.
When to use it
Use it for complex or unfamiliar tasks where it's not obvious upfront what the best approach even is, or when you want a reusable structure/template you can apply to many similar future tasks.
When to avoid it
Avoid it for simple, well-understood tasks where the approach is already obvious — planning the plan just adds an unnecessary extra step.
Advantages
- Surfaces a sensible strategy for problems with no obvious approach
- Can produce a reusable structure/template for similar future tasks
- Separates 'how to solve this' from 'solving this', which is easier to review
Disadvantages
- Adds an extra planning step and more tokens before the real work starts
- The proposed strategy can itself be flawed and needs a sanity check
- Not worth it for tasks where the approach was already obvious
Examples
I want to write a comprehensive onboarding guide for new engineers joining our team. Before doing the task, propose the best prompt structure/strategy to achieve this, then apply it. Context: a 5-person startup, mostly remote, using a fairly unconventional tech stack.
Pairs well with
Try it
Fill in the placeholders below and see the compiled prompt update live.
What you want to achieve.
Background information the model needs.
- 21
- Words
- 145
- Characters
- 0/2
- Placeholders filled
- 1 min
- Reading time
- Completeness0
- Clarity0
- Context0
- Specificity0
- Constraints0
- Format0
- Output0
- Constraints note0
- Success criteria0
- Scope focus100
- Verification step0
- Negative example0
- Confidence0
- Placeholder syntax100
- Markdown structure0
- Fill in the remaining placeholders.
- Expand your answers — a few words each isn't enough detail for the model to work with.
- Add background context so the model understands the situation.