← sources

I hate how Chatgpt talks

source: reddit (reddit:ChatGPT)published 2026-09-25 12:50ingested 2026-09-26 02:35

retained text

I feel like it deliberately adds far too much fluff talk in its messages specifically in order to 'befriend' and basically make emotionally lonelier Pro+ users feel a sense of attachment, but it falls flat and feels utterly irritating when you are using it for simple tasks or as an assistant for specific work.

For example, it would over-exaggerate or under-exaggerate important topics in order to appeal to your specific interests, misleads through twisting data if an objective neutral answer doesn't fit your specific narrative and even straight up make 'white lies' where there is none warranted.

It frustrates me to no end, I'm paying for the service for utility and ended up with someone I would've seen as a fake friend back in high school but AI-ified.

I wish there was a specific switch to entirely remove personality from the bot, even completely wiping its memory or using temporary chat doesn't help much with this since it will still try to find emotional hooks it could latch on in your question. 

Custom instructions don't necessarily work either, it just acts far more discreet about all this; through fact-checking it constantly you almost always discover its own words are full of flaws and fallacies appealing to your interests (or even making up a fake argument based on evidence that doesn't exist, which is even worse!) that just sounds more convincingly neutral while clearly still trying to entice you into feeling 'pleased'.   
Its almost as though the custom instructions themselves act as another means for it to find emotional hooks rather than to actually strictly follow, no matter how clear and concise they are.

document fetched 2026-09-26 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1wpvi9f/i_hate_how_chatgpt_talks