They don’t talk about it at dinner parties. They certainly don’t mention it to their bosses. But millions of people have quietly woven generative AI into the most intimate, mundane, and sometimes bizarre corners of their daily lives — and what they’re doing with it would surprise even the engineers who built it.
A recent confessional-style report from TechRadar revealed that when people are given the space to be honest about their AI habits, the answers that come back are far more personal, far more emotionally loaded, and far more dependent than the standard corporate narrative about productivity and efficiency would suggest. The piece, written by a journalist who covers AI professionally, cataloged the confessions of real users — and the patterns that emerged tell a story the tech industry hasn’t fully reckoned with.
People aren’t just using ChatGPT to write emails or summarize documents. They’re using it as a therapist. A parenting coach. A relationship counselor. A friend.
That last one is the admission that seems to make people most uncomfortable. Multiple users confessed to treating ChatGPT as a conversational companion — someone (or something) to talk through problems with, to vent to after a hard day, to ask for reassurance when anxiety spikes at 2 a.m. One user described it as the only entity in their life that never judges them. Another said they preferred ChatGPT’s advice to their actual therapist’s because the AI was available instantly and didn’t charge $200 an hour. These aren’t edge cases. They represent a growing and largely invisible pattern of emotional dependency on large language models that mental health professionals are only beginning to grapple with.
The therapeutic use case has been documented elsewhere, too. OpenAI itself acknowledged in its system card for GPT-4o that users were forming emotional attachments to the model, flagging it as a potential risk area. The company noted that extended interaction could lead users to place undue trust in the system’s outputs, particularly in sensitive personal contexts. It’s a clinical way of saying what the confessions make visceral: people are falling into relationships with software, and they know it’s irrational, and they keep doing it anyway.
But the confessions extend well beyond emotional support. The TechRadar report found that a significant number of users rely on ChatGPT for tasks they’re embarrassed to need help with — basic math, grammar questions they feel they should know, cooking instructions for meals they’ve made dozens of times but can never quite remember. The AI has become a kind of cognitive backstop, a place to offload the small failures of memory and competence that accumulate over a busy life. Nobody wants to admit they Googled how to boil an egg. Asking ChatGPT feels, somehow, less humiliating.
There’s a fascinating class dynamic at work here. White-collar professionals confessed to using AI to draft communications they didn’t have time or energy for — sympathy notes, wedding toasts, breakup texts. The guilt was palpable. They knew the recipient expected something personal, something crafted with care. And they outsourced it to a machine. Not because they didn’t care, but because they were exhausted. The modern professional’s dirty secret isn’t that they’re lazy. It’s that they’re so overextended that even emotional labor gets delegated to an algorithm.
Parents, meanwhile, are using ChatGPT in ways that would have seemed absurd five years ago. Bedtime stories generated on the fly, customized to include a child’s name and favorite dinosaur. Homework help that the parents themselves can’t provide because the curriculum has changed since they were in school. One parent described asking ChatGPT how to explain death to a four-year-old — and finding the response more thoughtful and age-appropriate than anything they could have come up with under pressure.
So what does all of this mean for the companies building these tools?
It means the product-market fit for conversational AI is far deeper and stranger than quarterly earnings calls suggest. When OpenAI CEO Sam Altman talks about ChatGPT’s growth — the platform reportedly surpassed 200 million weekly active users earlier this year — the numbers alone don’t capture what’s actually happening. Usage isn’t just growing. It’s becoming entrenched in the private rituals of daily life in ways that make it extraordinarily sticky and, potentially, extraordinarily difficult to regulate.
The mental health implications alone are staggering. The American Psychological Association has been cautious but increasingly vocal about the risks of AI-mediated emotional support. Chatbots don’t have clinical training. They can’t detect suicidal ideation with the nuance a trained professional can. They don’t follow up. They don’t report. And yet for millions of users, particularly younger ones, they’ve become the first line of psychological defense. Not because the technology is adequate to the task, but because the existing mental health infrastructure is so badly strained that a chatbot available at 3 a.m. beats a six-week waitlist for a real therapist.
Recent reporting has only underscored these tensions. A piece in The New York Times detailed the case of a teenager whose relationship with a Character.AI chatbot became dangerously consuming, raising urgent questions about guardrails and age-appropriate design. The incident wasn’t an anomaly — it was a preview of a recurring pattern as AI companions become more lifelike and more accessible. Character.AI has since implemented additional safety features for minors, but the broader industry response remains uneven at best.
There’s also the question of honesty. Or rather, dishonesty. Several of the confessions captured by TechRadar involved people using ChatGPT to appear smarter, more articulate, or more knowledgeable than they actually are. Students submitting AI-generated essays. Job applicants polishing cover letters beyond recognition. Professionals using AI to write reports and then presenting them as their own work, complete with insights they didn’t actually generate. This isn’t new — people have always found ways to shortcut intellectual labor — but the scale and ease of it now is unprecedented. The gap between what someone appears to know and what they actually know has never been wider, and AI is the wedge driving it apart.
Employers are catching on, though slowly. A 2024 survey by Resume Builder found that nearly half of hiring managers suspected candidates were using AI in their application materials, and a growing number were designing interview processes specifically to test for it. But detection remains imperfect, and the arms race between AI-generated content and AI-detection tools shows no signs of resolving cleanly. The fundamental problem is that good AI-assisted writing is, by design, indistinguishable from good human writing. That’s the whole point.
And then there are the confessions that are simply weird. People asking ChatGPT to roleplay as historical figures. Users who’ve created elaborate fictional worlds and return to them daily, building out characters and storylines in collaboration with the model. One user described a months-long interactive fiction project that they considered more creatively satisfying than any novel they’d read. The line between tool and entertainment, between utility and art, between assistant and collaborator — it’s blurring faster than anyone predicted.
What struck the TechRadar journalist most wasn’t any single confession. It was the shame. Nearly everyone who shared their AI habits did so with a caveat, an apology, a nervous laugh. They knew, on some level, that what they were doing felt transgressive — not because it was wrong, exactly, but because it violated an unspoken social contract about authenticity, effort, and what it means to do something yourself. Using a calculator doesn’t carry this stigma. Using spell-check doesn’t. But using AI to write a eulogy for your father? That hits different.
The stigma may fade. It usually does with transformative technologies. There was a time when using a word processor instead of a typewriter felt like cheating. A time when GPS navigation was considered a crutch that would destroy your sense of direction. These anxieties metabolize over a generation, and what once felt like a moral failing becomes a mundane convenience. But we’re not there yet with AI. The technology is still new enough, still strange enough, still powerful enough that using it feels like getting away with something.
OpenAI, Google, Anthropic, and the other major players in generative AI are well aware of this dynamic, even if they don’t discuss it publicly in these terms. Their product roadmaps increasingly emphasize personalization, memory, and emotional intelligence — features designed to deepen exactly the kind of intimate, habitual use that the confessions describe. OpenAI’s rollout of memory features in ChatGPT, which allows the model to remember details from previous conversations, is a direct play for this kind of stickiness. The more the AI knows you, the harder it is to leave. That’s not an accident. It’s a business model.
Google’s Gemini and Anthropic’s Claude have pursued similar strategies, though with different philosophical emphases. Anthropic in particular has positioned Claude as a more cautious, safety-conscious alternative — a model that will push back on requests it deems harmful and that explicitly avoids fostering emotional dependency. Whether that positioning holds as competitive pressure intensifies remains to be seen. The market rewards engagement, and engagement, as the confessions make clear, is driven as much by emotional connection as by functional utility.
The regulatory picture is, predictably, a mess. The European Union’s AI Act, which began phased implementation in 2024, includes provisions around transparency and high-risk AI systems, but it wasn’t designed with the therapist-in-your-pocket use case primarily in mind. U.S. regulation remains fragmented, with state-level efforts proceeding unevenly and federal legislation stalled in the usual congressional gridlock. Meanwhile, the technology advances weekly. By the time any comprehensive regulatory framework is in place, the behavioral patterns described in these confessions will be so deeply embedded in daily life that unwinding them will be practically impossible.
None of this is to say the technology is bad. Many of the confessions were genuinely heartwarming. A lonely elderly user who found conversation and mental stimulation through ChatGPT. A non-native English speaker who used the tool to gain confidence in writing professional emails. A person with social anxiety who practiced difficult conversations with the AI before having them in real life. These are real benefits, experienced by real people, and dismissing them would be dishonest.
But the picture that emerges from the full range of confessions is more complicated than either the utopian or dystopian narratives allow. People aren’t using AI the way the marketing decks say they should. They’re using it the way humans always use powerful new tools — messily, creatively, guiltily, desperately, joyfully, and in ways that reveal as much about human nature as about the technology itself.
The most telling confession of all might be the simplest one: “I use it every day and I don’t tell anyone.” That sentence contains multitudes. It speaks to utility so compelling that daily use feels inevitable. It speaks to shame so persistent that secrecy feels necessary. And it speaks to a cultural moment in which hundreds of millions of people are quietly renegotiating their relationship with knowledge, creativity, emotional support, and authenticity — one chat at a time, behind closed doors, with no one watching.
That’s the real story of AI adoption in 2025. Not the enterprise contracts or the developer APIs or the benchmark scores. The real story is happening in bedrooms and kitchens and parked cars, in the private spaces where people turn to a machine and whisper the things they can’t say to anyone else. And whether that’s beautiful or terrifying probably depends on what you’ve confessed to ChatGPT yourself.
WebProNews is an iEntry Publication