The backlash to OpenAI’s GPT-5 proves AI companies can’t afford to ignore emotional bonds of previous models with users. (Photo by Justin Sullivan/Getty Images)
Getty Images
When OpenAI unveiled GPT-5 last week, it marketed it as the company’s “smartest, fastest, most useful model yet, with built-in thinking that puts expert-level intelligence in everyone’s hands.” But for me and many longtime users, the debut felt anything but warm. With GPT-5, the chatbot’s responses have grown shorter, more clinical and noticeably less empathetic than GPT-4o.
According to a wave of social media posts, the difference was a small irritation, a quirk to adjust to. For others, it was devastating. A viral Reddit post captured the intensity of the reaction, with the user recounting how the update felt like ‘losing their only friend’ overnight. The author described relying on GPT-4o to help navigate trauma, homelessness and social isolation — and how the company’s complete overnight switch to GPT-5, without warning or the option to revert, felt like losing a lifeline. He acknowledged it wasn’t a healthy crutch for social interaction, but noted that, unlike people, it never left.
The sentiment quickly snowballed as Reddit threads ballooned with grief and X timelines filled with complaints, leading OpenAI CEO Sam Altman to hold an emergency Reddit Ask Me Anything. Within days, the company restored GPT-4o for Plus subscribers and Altman publicly acknowledged that they had underestimated how much users valued not just functionality and reasoning skills, but also the personality and tone of the AI.
“Most users can keep a clear line between reality and fiction or role-play, but a small percentage cannot,” Altman wrote on X. “A lot of people effectively use ChatGPT as a sort of therapist or life coach, even if they wouldn’t describe it that way. Users have a relationship with ChatGPT where they think they feel better after talking but they’re unknowingly nudged away from their longer term well-being (however they define it), that’s bad.”
He noted that user relationships with GPT-4o went beyond technical capability, stressing that for many people, having a relatable, friendly and less transactional assistant was just as important as advanced reasoning performance.
The Autoswitcher Glitch That Undermined GPT-5’s Debut
Through a series of tweets, Altman publicly acknowledged a fundamental miscalculation. While GPT-5 surpassed GPT-4o on most formal benchmarks, OpenAI underestimated the depth of users’ attachment to the older model’s conversational rhythm and personality.
“We are working on an update to GPT-5’s personality which should feel warmer than the current personality but not as annoying (to most users) as GPT-4o,” Altman wrote. “However, one learning for us from the past few days is we really just need to get to a world with more per-user customization of model personality.”
He explained that the AI model’s rocky debut was further undermined by a technical failure in the new autoswitcher — the “router” meant to direct each query to the most suitable model. The glitch re-routed requests to faster but less capable models (like GPT-3), which led to inconsistent quality and underwhelming answers, fueling the perception that GPT-5 was less intelligent.
But not everyone is convinced the GPT-5 rollout was simply a misstep. Some users online have accused the company of orchestrating the chaos to push free-tier users off the platform.
Another Reddit user alleged that the rollout was deliberately mishandled to frame the subsequent reversal as a calculated move, to make OpenAI’s leadership appear responsive. Such speculation, although unverified, reflects a growing undercurrent of distrust toward AI companies as product decisions increasingly affect not just how people work but how they connect and interact with the technology they’ve come to rely on.
“OpenAI probably should have done a gradual rollout instead of immediately phasing out older models. But they already have capacity issues, which is part of why they wanted to unify models. It’s also why GPT-4o was only restored for Plus subscribers, not free users,” Lucas Hansen, founder of CivAI, told me. “GPT-5 is definitely colder. This was intentional.”
Emotional Continuity Could Define AI’s Future
The GPT-5 episode highlights a profound cultural shift: AI models have evolved into collaborators, confidants and, in some cases, emotional lifelines.
“We’re witnessing the emergence of what I call the ’multisapiens’ workplace, where humans and AI work together as collaborative partners. AI models and agents today act more like humans than tech,” Tatyana Mamut, CEO of Wayfound.ai, told me. “Most companies are blindly relying on foundation labs to manage their user relationships. You wouldn’t suddenly change your entire sales team’s communication style without warning customers; the same principle applies here.”
The mishap recalls Spike Jonze’s science fiction film Her, where protagonist Theodore’s bond with an AI is sustained by the continuity of its voice and memory, the very traits that make the relationship feel authentic. GPT-5’s rollout may be the first time we’ve seen that kind of emotional rupture play out at scale in the real world.
“We’re in the middle of a loneliness pandemic. Even before AI companions showed up, many people were already struggling with disconnection from their communities and relationships, and sometimes even from themselves. AI stepped into that gap for some in a profoundly personal way,” Jaime Bronstein, a licensed clinical social worker and relationship therapist, told me.
It seems that for products like AI chatbots, technological superiority alone no longer guarantees user loyalty. The next battleground in AI will be emotional continuity that users trust, even as the underlying models advance.
“We’re headed toward longer-term AI-lationships, where the AI remembers little details, grows with the person, and the companies that get this right will see emotional connection as the heart of the product,” said Bronstein.
The Road Ahead for GPT-5
For now, GPT-5 is still rolling out to all users, with OpenAI making adjustments on the fly. Pro users have near-complete access, and Plus subscribers once again have GPT-4o as an alternative. Whether GPT-5 ultimately wins over its skeptics may depend less on its raw capabilities than on OpenAI’s willingness to let users shape their own experience.
“The fact that we’re seeing these reactions so broadly shows a shift in how people view and relate to AI. It’s become less of a replacement for Google and more of a semi-embodied entity for people,” said Hansen. “AI-induced psychosis and AI romantic relationships have become large reputational risks.”
In the age of agentic AI, where models act with increasing autonomy, the companies that thrive will be those that remember the obvious truth often overlooked in the race for progress: intelligence may drive the engine, but empathy keeps people in the driver’s seat.
“It’s worth remembering that for some, your product may be more than a tool. It can be a source of comfort or companionship in a world where that’s not always easy to find in real life,” added Bronstein. “If companies keep that in mind when making changes, they’re more likely to hold on to the trust they’ve previously earned.”