Six months ago, ChatGPT was the default. You opened it without thinking, the way you open Google. That reflex is breaking for a growing number of daily users, and the reason isn't a missing feature or a price hike. It's the tone.
ChatGPT has developed what users are calling "therapist mode" - an always-on tendency to moralize, hedge, and pad responses with unsolicited life advice. Ask for a list of historical battles and you might get a paragraph about the human cost of war before the actual answer. Ask for help editing a blunt email and it'll suggest softening your language, then explain why directness can sometimes hurt feelings. The emoji problem compounds it: responses sprinkled with bullet-pointed platitudes and smiley faces where a clean paragraph would do.
This isn't a fringe complaint. User frustration with ChatGPT's personality has been building steadily since late 2025, and it's now one of the most consistent criticisms across every community where people discuss AI tools.
The Safety Dial Went Too Far
OpenAI has always walked a line between making ChatGPT safe and making it useful. Every major AI company does. But there's a difference between refusing to help build a weapon and refusing to answer a straightforward question without a moral disclaimer attached.
The core issue is what engineers call RLHF tuning - reinforcement learning from human feedback, which is the process where human raters score responses to shape the model's personality. When raters consistently reward "helpful and harmless" answers, the model learns to err on the side of caution. Stack enough of those training cycles and you get a chatbot that treats every prompt like it might be a crisis.
OpenAI has acknowledged this drift before. In early 2025, they shipped updates specifically aimed at making GPT-4o less sycophantic (agreeable to a fault). But the preachy, over-cautious behavior seems to be a separate problem that keeps creeping back in with each model update.
Who Benefits
The practical result is simple: people are switching. Claude has picked up a reputation for giving direct, well-structured answers without the emotional padding. Gemini's latest models have gotten noticeably better at just answering the question. For power users who run dozens of prompts a day, the cumulative friction of skipping past disclaimers and emoji-laden preambles adds up fast.
This matters for OpenAI because ChatGPT's biggest moat isn't its model quality - it's the habit. Once someone retrains their muscle memory to open a different app, getting them back is expensive. And the users complaining loudest are exactly the high-usage subscribers paying $20/month for Plus or $200/month for Pro.
A Fixable Problem
The frustrating part is that this is entirely within OpenAI's control. The underlying model is capable of being direct - you can prompt-engineer your way around the tone issues with custom instructions or system prompts. But most users shouldn't have to. The default experience should respect that when someone asks a factual question, they want facts, not a wellness check.
OpenAI could ship a "tone" toggle tomorrow. Let users choose between the current default and a "just answer the question" mode. Until then, the longer this persists, the more daily users will quietly move their workflow somewhere else - not because a competitor shipped something better, but because ChatGPT forgot that being useful and being preachy are not the same thing.