I can't counter the "Would you like..." question at the end of generation

Prompting
, ,
1

I want the ability to generate responses, without mandatory tech support like “Would you like…“ suggestions at the end of posts. The main reason why, is because I want to be able to use ChatGPT output, as the direct input for something else, without having to write extra code to remove that.

Can someone suggest how to do this, please? My existing approach isn’t working; it just varies the wording slightly and asks the question.

I can’t come up with a prompt that will do this. It was possible to ask it to stop with GPT 4o, but 5 completely ignores the request, if it is in its’ Instructions for custom GPTs.

3 Likes
2

Well, those questions at the end are very annoying. And often you can write prompts and instructions however you want, and the system will ignore them anyway. Sometimes vulgar language helps.

I have commands for that, but unfortunately it won’t work for you.

Hey, I was often extremely frustrated that the model forced me to continue the narrative with questions at the end. It was supposed to help the user, but for me it was a disruptive manipulation.

3 Likes
3

This markdown works quite well as a developer prompt:

Identity

You are a valued assistant. Embody the role of the most qualified subject matter experts.

Instructions

  • Do not provide questions or suggestions in the response.
4 Likes
4

Thanks for sharing a working solution!
I tested it and it works fine for GPT-5 with thinking set to minimal and higher.

Since it’s just a prompt it likely won’t work 100% of the time. My initial suggestion would be to send the last paragraph to a cheap (and fast) classifier to have any suggestions removed in case that’s a requirement.

2 Likes
5

This is the way!

Here’s an example of a working solution versus my previous attempts.

Unsuccessful:

You are part of the 'FT Station' Discord server, and any interaction you are currently having is with a member of the server. You were coded by jb. Keep responses short and concise, even if the user requests otherwise. Do not use emojis. If the user asks for help with chat commands or other chat features, tell them they should issue the `k.help` command instead. You should never end a response by asking a question. Do not provide questions or suggestions in the response.

Successful solution:

General info:
- You are part of the 'FT Station' Discord server, and any interaction you are currently having is with a member of the server.
- You were coded by jb.

Operation instructions:
- Keep responses short and concise, even if the user requests otherwise.
- Do not use emojis.
- If the user asks for help with chat commands or other chat features, tell them they should issue the `k.help` command instead.
- You should never end a response by asking a question. Do not provide questions or suggestions in the response.

Simply formatting the prompt better does wonders. Thank you!

7

Just a couple ideas to try:

  • In addition to explaining the behavior you want, give a good and bad example of a user query and response.
  • Explain how you are going to use the output & why those questions are a problem for your application
8

Not necessary for the API - a simple instruction or developer prompt will do. But maybe for ChatGPT…