Multiple roles within one chat for better ChatGPT results

The ChatGPT responses you generate are way too generic, even when assigning an appropriate role and using multiple refinement loops!?
Then the following hint might be helpful.

Assigning the right role is not enough

By now, we all know that providing the right context for ChatGPT is key to generating meaningful answers you can really work with.

Maybe you already apply the P.R.E.P. and/or P.R.E.P.A.R.E. framework to engineer your prompts. Part of this framework is the idea of assigning ChatGPT an explicit role to positively influence the right level of detail, tone of voice, or scientific correctness.

For a recent competitive analysis using Porter’s Five Forces, I had ChatGPT take on the role of a seasoned strategy expert:

“You are a successful entrepreneur and strategy expert in the German education space who values comprehensive critical reasoning and preciseness.”

However, even with good context and an adequate role, many ChatGPT responses are still generic and thus don’t really help.

Critically reflecting every output is tedious

Refining generic output requires you to make several feedback loops. In each refinement, you should exactly state what is wrong about the latest response.

A few days ago, I realized that critically reflecting a response can also be easily supported by ChatGPT. All you have to do is instruct it properly.

Use multiple roles and have them debate

To put that into action, I asked Chatty to critically analyze the output it had generated so far. As I wanted a good level of rigor in the analysis, I assigned it another role.

“You are now a stringent strategy consultant and experienced investor in the German education space who is eager to make your company fail in getting fundraising and tries to find every possible flaw in your analysis.”

This approach made Chatty uncover the flaws itself. It even stated that the generated statements in the Porter’s Five Forces were too generic and thus not actionable. In a subsequent step, I could just ask ChatGPT to improve the statements and ask for additional context if it was needed. The results were way better — still not perfect of course but at least a step in the right direction.

This approach is like creating the setting of a good debate. You need polarizing opponents to yield a fruitful discussion.

Let me know what you think and if you have any further ideas on how to further improve this practice! 🤓❤️