In the meantime, AI agents behaviour is remarkbly similar. I use ChatGPT, Perplexity, Copilot and Gemini hoping that the answers from one will augment the answers from the others.
Unfortunately, in my personal experience, that is not the case. Not only the answers are often identical, even the syntax grammar and writing style are identical .
Much worse is the increasing tendency to not answer the original question, but make up something that the AI assistent thinks is relevant.
It will give either vague answers, or alternatives that I did not ask for, gives long lists of viewpoints that I did not want, and then comes with follow up questions like “would you like me to…”?” or “another possibility is… Shall I do…?”. And always errors.
Thus my 3rd prompt usually is something along the following lines:
“do exactly what i ask. Only answer the question. Do not hallucinate, do not make up answers. Do not give answers to questions I did not ask. Do not give alternatives or ask follow up questions. Give exact answers only, admit if you do not know the exact answer.”
Used in conjunction with the second prompt principle it usually results in more useful material. All four still quit often give nonsense answers or facts that simply do not exist.
See my earlier posts on LinkedIn.
Essential is the formulation of the starting prompt. Just apply the lessons from Requirement Analysis and Problem Deconstruction and you are good to go.
Good luck! Let me know your thoughts.