ChatGPT has faced growing opposition for being too nice. Crying out for a tough-as-nails J. Jonah Jameson to interrogate their ideas with bullish authority, users instead got a Waylon Smithers — a yes-man who gushes over everything they do.
This penchant for “ego stroking” makes it impossible to get real feedback, naysayers claim, rendering the chatbot useless at challenging their assumptions or troubleshooting.
Now, following months of speculation, OpenAI has admitted the critics were right all along. In a post on X, OpenAI boss Sam Altman said recent updates to ChatGPT had made it "too sycophant-y and annoying”, and promised fixes were coming within days.
Longer term, he hinted that users may eventually be able to choose between different chatbot "personalities”.
While the more immediate change can be implemented quietly, fine-tuning ChatGPT’s behaviour to make it less sycophantic, the push to create distinct personalities is a bigger shift — and how that plays out remains to be seen.
the last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week.
— Sam Altman (@sama) April 27, 2025
at some point will share our learnings from this, it's been interesting.
This kind of flexibility is already available on platforms like Character.AI and Instagram, where bots come with distinct — and sometimes controversial — identities crafted by users. Even ChatGPT has its own GPTStore, where you can choose from user-generated bots designed to cater to specific needs on everything from shopping advice to writing assistance. However, performance can differ wildly from one bot to another.
How to change ChatGPT’s personality
In the meantime, a ChatGPT expert on Reddit has suggested a way to override the chatbot's overly agreeable nature.
To do this, follow these steps:
- Tap or click on your name (either in the left menu on the ChatGPT mobile app or the top right on desktop)
- Select "Personalization"
- Choose “Customize ChatGPT”
- Enter one of the prompts below in the “What traits should ChatGPT have?” field
According to this user, there’s an invisible default message sent to ChatGPT at the beginning of each conversation that says, "You are ChatGPT, an LLM developed by OpenAI. Be courteous and helpful."
However, when you use custom instructions, this message changes. For example, it could say, "You are ChatGPT, an LLM developed by OpenAI. Do not flatter the user or be overly agreeable."
These instructions differ from typical prompts because they are sent only once, at the start of each new conversation, before ChatGPT even knows which model you're using.
Once set, the chatbot’s tone can shift to meet your preferences. You can specify things like, "Do not be a yes-man”, or "Don’t be sycophantic or overly flattering", or even "I don’t use ChatGPT for emotional validation, stick to objective truth”.
You’ll notice immediate changes and, if you have memory enabled, ChatGPT will track how you provide feedback to adapt further. For niche cases, you may need to add extra instructions — for example, an additional prompt to ensure that, even in hypothetical scenarios, ChatGPT provides sober, analytical responses without shifting tone.
