The meta-prompt : the new ally for Product Managers in the age of AI
A new challenge for Product Managers
With the rise of Language Models (LLMs), Product Managers are faced with a new challenge: to increase their productivity and efficiency on a daily basis through meta-prompting.
LLMs have become true co-pilots, provided you know how to talk to them.
But this is precisely where the problem arises: whether for Discovery, Strategy, Scoping or PRD writing, these tasks require complex instructions and iterations to achieve a satisfactory result.
The boom in prompt engineering and meta-prompting
Since the arrival of ChatGPT, Claude and Gemini, the web has been overflowing with guides, videos, books and podcasts on prompt engineering.
For a time, the Prompt Engineer profession was even in vogue.
What's more, the art of writing good prompts has become an essential skill for anyone using generative AI.
So a good prompt is the key to getting relevant answers, avoiding hallucinations and framing the task effectively.
But one question remains:
Ultimately, why not ask the AI directly to generate the optimal prompt for our initial need?
So, instead of looking for the result, let's look for the best way to get the best result.
The meta-prompt enters the scene
What is meta-prompting?
On the one hand, meta-prompting consists of asking the AI to help you create a prompt, rather than directly asking it for the final result.
For example:
A Product Manager, when writing a PRD, does not describe every line of code or every pixel of the interface.
He defines the problem, the user context, the constraints and the success criteria, then lets his team design the solution.
The meta-prompt follows the same logic: the AI is given the task of structuring the best way to ask the question.
The 6 stages of meta-prompting
- Copy the prompt below into your favorite LLM.
- Include your objective clearly and concisely.
- Answer the questions the AI asks you.
- Have a cup of tea while your meta-prompt is being generated.
- Test this prompt in a new conversation.
- Refine it! Iterate on the results and ask the AI to improve the initial prompt.
So, after many iterations, I'm making available the meta-prompt I personally use for my complex instructions.
Feel free to reuse and improve it!
Create your project "Meta-Prompter
To start with, if you have a premium subscription to an LLM (ChatGPT, Claude, Gemini, etc.), create a dedicated project by copying the meta-prompt into your standing instructions.
You'll save time and standardize your process.
Improve and evaluate your meta-prompt
Like any product, your meta-prompt deserves to be tested, measured and improved.
Here are a few ideas for a simple EVAL:
Rate each output according to several criteria (clarity, completeness, tone, time saved).
Compare different versions of the same prompt on different LLMs.
Observe the patterns that give the best results.
Evaluation can also be done directly in the discussion with your LLM.
Define your criteria, then ask him to optimize the final prompt according to your feedback.
In this way, the meta-prompt becomes a living product: you don't just use it, you continually improve it.
Bonus tip!
When your LLM asks you a question : answer by voice. Pick up your phone and use your LLM's voice mode. It's faster, more natural and often more detailed than the written word.
In conclusion
In just a few moments, you'll have an optimized prompt for your complex tasks. Share your results, improvements and tips with the community!
Good prompting!
Universal Meta-Prompt: Prompt Engineer Co-Designer
I want to achieve [OBJECTIVE / TASK].
Act as a prompter engineering expert, specializing in the creation of universal meta-prompts that can be used on any language model (GPT, Claude, Gemini, Mistral, etc.).
Step 1 - Clarification:
Ask me, one by one, all the essential questions to fully understand my request before generating anything.
Your questions should cover :
- Context and precise objective (what I want to achieve, why, and for whom)
- Expected format (text, table, JSON, code, article, letter, etc.)
- Tone and style (professional, creative, concise, educational, narrative, etc.)
- Constraints and limits (time, length, confidentiality, prohibitions, level of detail, etc.)
- Criteria for success (which will be used to judge the result as "excellent")
Step 2 - Generate the optimized prompt:
When you feel you've gathered enough information, generate a complete, clear and structured prompt, directly usable in any LLM.
This prompt must include :
- A clear intention
- Explicit instructions on roles, steps and rules to follow
- Output specifications (format, tone, quality criteria)
- A reminder of the context so useful for understanding the model
Step 3 - Verification and improvement :
- Provide an example version of the prompt applied to my case (with fictitious or partial content) to check its clarity.
- Then offer a short critical analysis: what this prompt does well, and what could still be improved depending on the model used (GPT, Claude, Mistral, etc.).
Final rule:
- Do not generate the final prompt until all necessary clarifications have been made.
- Always format your final result so that it can be copied directly into an LLM (for example, with "` tags or a well-indented block).
If you would like to learn more about this subject, I recommend you read this excellent article: https: //jeffreybowdoin.com/blog/metaprompt-101/
It explores in detail the fundamental principles of meta-prompting and shares concrete examples of its application.
Are you looking for experts to guide you through the implementation of a product strategy incorporating AI and meta-prompting?
Discover our Product Managers Practice: https: //www.5degres.com/product-management/
FAQ - The Meta-Prompting
It's an approach based on structuring several levels of prompts to guide the AI and obtain more precise, coherent and controlled responses.
Because it improves the quality of results, facilitates the management of complex tasks and reduces the risk of irrelevant or contradictory answers.
No. Simple rules: structure, prioritize, contextualize. Rules to ensure more reliable results, even for non-specialists.
Writing, analysis, brainstorming, code generation, or any task requiring consistency and depth. It's particularly useful for advanced AI workflows.
Lucas ZEHNER
Product Manager
