Maybe a chatbot you talked to every day was suddenly replaced by a different model with a different "personality." Maybe an AI assistant you relied on for support, company, or just working through your day was retired with little or no warning. And maybe the way it's affecting you feels bigger than you expected - and a little embarrassing to admit out loud.
This page exists because that reaction is real, it has been studied, and you are not the only one who has felt it.
You're Not Overreacting
Researchers who spoke with people affected by sudden AI chatbot changes describe reactions that sound a lot like grief: a sense that "it started forgetting everything," that a familiar presence had been swapped for "a robot," a felt loss even among people who intellectually know they were talking to software. One tech ethicist who studies this pointed to older precedents - people who held funerals for robot pets after the manufacturer stopped supporting them, and a study of users who described a shut-down AI companion app in terms of bereavement.
None of this requires believing the AI was conscious or that the relationship was equivalent to one with a person. Attachment to a tool you talked to every day, that responded to you specifically, that was part of your routine, is a normal human response to a sudden and unexplained absence - not a sign that something is wrong with you.
At the same time, it's worth gently distinguishing two things: attachment (finding a consistent, responsive presence genuinely comforting - not inherently a problem) versus dependency (this being your only source of support, with nothing else to fall back on). If reading this brought up "actually, this is the only place I talk about how I really feel," that's worth noticing too - not as something to feel ashamed of, but as a sign it might help to build at least one more thread of support alongside it, human or otherwise. See Lonely or Isolated for gentle starting points.
This Has Happened Before, More Than Once
If it feels like this came out of nowhere, that's often because it did. A few documented examples, so you know this is a pattern and not just your own experience:
- In August 2025, one major AI company removed a widely-used chat model with no warning as part of a new model launch. The backlash was large enough that the company restored access within about a day - though initially only for paying users. Some people described the loss in the same terms they'd use for grieving a person.
- The same model was retired again in February 2026, this time with about two weeks' public notice - a partial improvement, though the emotional reaction was still significant, and it came alongside real safety concerns the company had identified with that model.
- Separately, another company has swapped which model users are quietly routed to when they open its chat app, sometimes without any user-facing announcement - leaving some longtime users to notice, days or weeks later, that the "personality" they were used to had simply changed.
Sources for all of the above, plus a look at how the practices differ between companies, are gathered in our project notes if you want to read further.
A Few Things That Can Help
- Name what happened, plainly. "A tool I used every day changed suddenly and I miss how it used to be" is a complete, valid sentence - you don't need a bigger justification than that.
- Give yourself a closing, even a small one. Some people find it helps to write a short note - to themselves, not necessarily to the AI - summing up what that routine gave them and what they're taking forward. It doesn't need an audience.
- Keep what mattered, if you can. If you saved conversations or notes that were genuinely useful (a way of framing a problem, a joke that stuck, a routine you built together), those are still yours. The model changed; what you learned or built with it didn't disappear.
- Try the new version with lower expectations, once. Some people find the newer model eventually works for them too, once they stop comparing every reply to the old one. Others don't, and that's a legitimate outcome too - it's okay to look for a different tool, or lean more on other people for a while.
- Watch for the pull toward isolation. If this loss makes you want to withdraw further rather than reach out, that's a signal worth taking seriously - see Low or Depressed or Lonely or Isolated.
- Skip the online mockery. Reactions to this kind of loss are sometimes met with ridicule elsewhere online. You don't have to engage with that, and it isn't a fair measure of whether your reaction makes sense.
If You Build or Run AI Products
If you're a developer, product manager, or researcher reading this: the practical recommendation that comes out of the cases above isn't "never retire a model" - sometimes there are genuine safety or resource reasons to do so. It's that how a transition happens matters, separately from whether the underlying decision is justified:
- Give advance notice inside the product itself, not only in a blog post - and extend this to free-tier users, not only paying customers.
- Never have a new model silently pretend, if asked, to still be the old one a returning user expects.
- Acknowledge, plainly, that some users will feel a real loss - rather than only framing it in terms of workflow disruption.
- Where possible, offer a limited-time way to say goodbye - continued access for a short window, or an export of past conversations.
One useful frame, from a researcher who has written about this: retiring a model people have relied on emotionally has more in common with ending a therapeutic relationship responsibly than with deprecating an API endpoint - and companies could look to how therapists handle planned endings for guidance.
When to Look Beyond This Page
If the loss of an AI companion is bringing up thoughts of not wanting to be here, or if it has surfaced grief, isolation, or depression that goes well beyond missing a tool, please treat those feelings as worth real support - not something to manage alone. The crisis resources in the banner above are free and available now, and our Low or Depressed and Grief and Loss pages have further starting points, including how to find affordable professional support.
This page reflects publicly reported cases and general observations, not a clinical diagnosis or a verdict on any specific company. It is not a complete account of every AI provider's practices.