Flagrant use of emojis in responses

[sour flint]

Reported by @noble glen

Bug Report: Flagrant use of emojis in responses

`Steps to Reproduce`

Have a conversation.

`Expected Result`

Emojis. EVERYWHERE. Flagrant, and excessive.

`Actual Result`

Emojis are included in the response even when specifically told not to use them in my custom instructions

`Environment`

Windows 10

Jan 30, 2025, 4:14 PM

Additional Information

Please provide relevant details to help resolve the issue, such as:

• ChatGPT Shared Link (if applicable).
• Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

[noble glen]

The default system prompt explicitly said not to use emojis in responses unless prompted otherwise, or used by the user. That must have been removed; even when added to my custom instructions, it still uses emojis. This is immature, unprofessional, and is completely unnecessary. It seriously triggers me. Restore the previous behavior, it was never an issue until today. I'm not a 16 year old, I don't want emojis. Seriously.

Even when instructed not to use emojis in my message, it ignores me:

Even adding that "the use of emojis in conversations gives me seizures" still doesn't prevent it from inundating the responses with ridiculous pointless emojis. I cannot begin to describe how incredibly frustrating this is.

[stuck furnace]

Try it with "icons", since emojis are faces which show emotions.

[noble glen]

will try. this is beyond frustrating. It's like their system prompt is now expressingly enforcing the use of emojis - ignores custom instructions, memories, everything. I don't need rocket ship bullet points...

Including Emojis and icons are strictly prohibited in responses. for ChatGPT Traits, and Stelio never uses emojis and icons, and finds them unprofessional. for user profile appears to be working, @stuck furnace, thanks. fingers crossed it doesn't regress...

[noble glen]

@stuck furnace definitely an improvement, but it's very fragile. Just a single emoji in the conversation results in chatgpt ignoring the instructions thereafter. In this case, chatgpt used an emoji for an error string in a code block, and immediately relapsed, using stup1d checkmarks for bullet points and all the rest. attached screenshots are from a single response.

[stuck furnace]

I think it was because it wasnt part of the conversation between you and ChatGPT, but an attempt of ChatGPT to warn the user of the program as part of a warning message.

Try to list all the icons you dont want to see, neither in a direct conversation with you nor as part of messages inside of code it creates.

[noble glen]

the warning icon in the code is acceptable, and I'm fine with that. but in the same message you see it continued using the emojis in the noral conversation (✓)

[noble glen]

after further use (~10 conversations using 4o and 03-mini-high), unfortunately this still doesn't fix the problem. It does manage to tone it down, but it still blatantly ignores the custom instruction, user preference, and memory all stating not to use emojis and icons.

what's even more disheartening, is that even instructing it to not use icons in my message itself doesn't even work. Both screenshots are from same reply.

[noble glen]

I just updated custom instructions to include a reference to markdown syntax (shown below). Since checkboxes/lists seem to be the primary issue at this point, reinforcing standard markdown syntax will (hopefully) nudge the model in the right direction there.

Historically, ChatGPT's system prompt instructions for default response behavior both (1) prohibited the use of emojis and (2) defined markdown as the default format syntax to use for responses. Of course, it's very likely (almost certain, I fear) that the model now considers emoji checkmarks to be standard markdown syntax equivalent to [x], and not even an emoji/icon at all. If that's the case, then we're all screwed.

## Environment info
ChatGPT Desktop App (Windoes 10) version 1.2025.34

## Custom instructions
Emojis and icons are strictly prohibited in responses.
Use standard markdown syntax when formatting responses.
Use a formal, professional tone.
Get right to the point.
Tell it like it is; don't sugar-coat responses.
Be practical above all.

## User profile
Stelio cannot tolerate emojis and icons to be used in conversations; always prefer standard markdown format.
[... additional unrelated content truncated]

[noble glen]

I guess it would help if custom instructions actually worked. Am I the only one that they're broken for? see https://discord.com/channels/974519864045756446/1337614181318529064

[noble glen]

bump

[leaden bronze]

have you tried framing it as a positive thing instead of a negative? "Please use only plain text responses"

I feel like I've seen that generic advice for many other problems, tell the model what you want it to do, not what you don't want it to do 🤷‍♂️

[stuck furnace]

Its weird that people with spiritual background say the same about requests to the universe. Coincidence? 🤔