Hello. I spent a few months working with o1 to write an advanced document transformation program. o1 handles all the prompting like a champ. However, when I put those exact same prompts into GPT-5 it fails miserably, even on high reasoning effort. What’s even worse, GPT-5 completely ignores my instructions. I am trying to modify my o1 prompt to work with GPT-5, but the changes that I make just get ignored.
What is going on here, and how do I fix it?
4 Likes
Welcome to the forum.
Yes, gpt-5 prompting is a bit different from previous models.
Have a look at these articles, they might help you a bit:
1 Like
I’ve found the GPT-5 prompt optimizer helpful for a few things.
https://platform.openai.com/chat/edit?prompt=
You will likely need to adapt the prompts, though.
Were the old ones reasoning too?
Depending on your prompt, the verbosity parameter (default medium) may be problematic.
For example, we translate entire documents and have to set verbosity to low.
1 Like
Basically, your old o1 prompt isn’t clicking with GPT-5 because the two models think in different ways.
-
o1 is great at reading between the lines — it reasons deeply and fills in gaps.
-
GPT-5 is more literal — it wants clear, structured, no-nonsense instructions.
So when you reuse your o1 prompt, GPT-5 either ignores parts of it or gets confused.
Here’s how to fix it:
-
Put your must-follow rules in the system message, not the user prompt.
-
Use structured outputs (like JSON mode or function calling) instead of hoping it formats things right.
-
Set temperature = 0 so it stays consistent.
-
Give it a tiny example of what you want.
-
Don’t rely on hidden “reasoning” — spell out the steps you expect.
In short: make your prompt more explicit, structured, and strict, and GPT-5 will follow along just fine.
1 Like