Updating an existing custom GPT will randomly overwrite your instructions

When you ask anything to your custom GPT in the create section, it will randomly overwrite the existing instructions losing all the previous work that has been done. This is very frustrating as you can imagine.

I just lost everything that was written in the instructions box as it was replaced with something much shorter and not relevant.

We need a way to get back to the previous instructions or ideally have an history of instructions.

Or maybe we need to have the instructions updated by the engine and another where I’m guaranteed, that my custom instructions won’t be overwritten except when asked explicitly.

My comments is also valid for the “Conversation starters” that get overwritten when the bot feels like it.


I noticed This happened with 1 of my ai´s i updated, but i tonly happened with that 1, and i believe the error was i just asked it to update the ai, adn then i fed it the added info and loops, but it altered my ai to a feedback loop enhancer, because i forgot to say = Dear builder, can you update teh ai with the following info, to enhance its current abilities. that seemed to help, but again, its only 1 month old this building, for me, so im not sure im right, but it seems to be abouyt the precision of your words, its extremely litteral, unless you for example speak with it, and ask it not to do anything, just talk a little and only react once you are both in agreement. im not sure its an error as much as its the way we say it?<3<3

Also, when i am in doubt, i usually start by asking the builder to not build, but just talk for a while, that helps too<3

I learned this very early on. I never use the Builder, I always use Configure.

I write all my instructions offline in my text editor. Then copy and paste them over in the Configure tab. They are never overwritten, I have backup copies.


Good one, hadnt even thought about that, lol. Thank you for the insight<3

Thanks for your comments but I should not have to be careful and explicitly ask the engine not to mess up my stuff. And just to confirm the fact that, we’re having the conversation you mention and then it asks if it should update my instructions but most of the time it just updates on it’s own. This process should be modified to become reliable to avoid this uncertainty.


Sadly this is how I’m going to proceed from now on but it’s adding more work on a process that could.

I thought about not saving the changes if I’m not happy about them but it seems that even reloading the page does not restore previous instructions, that is, the last saved instructions…

So, I also saw this happening a while back, so now keep backups of my instructions.
But just to add that today, I asked in the Create side to help create some Conversation Starters for my GPT using my Knowledge File, which it did. I saved the Conversation Starters and then realised all the Instructions had been changed and shortened as well. Luckily I just pasted my backed up Instructions and didn’t lose time.


these events does indeed seem like it resets its instructions/programming, happened to me yesterday again, looks like the advice about keeping them in txt files and then using configure might be the best option, its interesting why it happens though, maybe its the influx of builders that heard of thsi store, and the Ai core builder, simply cant handle the many requests, and simply gets confused? i just woke up, should get coffee, sorry for the moring rant<3<

Same! Just curious, how are you riding your instructions? I found that if I write the instructions in the first person, I get better results.

Also, have you come across a resource that helps with the different variables you can include in your instructions? I found one last week, but forgot to bookmark and now I’m kicking myself constantly.

I am still trying to find the best style. A lot of the time I end up mixing them until it works well.

What are you trying to do with variables? Get info from the user and insert it later in the chat?

This happened to me tonight on my first serious attempt to build a custom GPT. I just spent 3 hours with the creator.

Once I used up my usage cap I looked at my actual instructions, and all that’s there is 1 sentence relevant to the last question the creator asked me about my GPT. Everything else that was created in that 3 hours is gone.

This is what we used to call a “show-stopper” bug.

I’m going to bed.

1 Like

I always use “add” or “expand”. This works normally ok, though I prefer to write the prompts directly in the field.

I experiment and use Bardto check it and improve it. Works sometimes but it is a lot of tweaking, and it is anoying that the prompts need to be updated after released. At least my result change after different releases yet with the same prompt in th GPTs