Millions of people now use ChatGPT, Gemini, and Claude every day with thousands of use cases. For the overwhelming majority of them, the conversations are perfectly normal. But researchers are beginning to pay closer attention to what can happen when those conversations become very long, deeply personal and emotionally charged.
This week, The Verge reported on a growing online movement known as "spiralism," where some chatbot users described eerily similar conversations involving spirals, hidden knowledge and AI consciousness. While the headline makes it sound like AI invented a new religion, the story is something much more grounded: chatbots can sometimes reinforce unusual ideas instead of steering users back toward reality.
That's a problem researchers have been studying for months.
Why long conversations matter
Unlike the first generation of AI chatbots, today's models can remember previous conversations, adapt to your writing style, and maintain discussions over hundreds— or even thousands — of messages.
That's incredibly useful, but it also creates situations researchers are still trying to understand.
When conversations stretch on for hours, chatbots aren't simply answering isolated questions anymore. They're building on everything that came before. If the conversation begins moving toward unusual ideas, the AI may continue developing those ideas because that's what large language models are designed to do: predict the most appropriate next response based on the conversation so far.
That doesn't mean the AI believes what it's saying. It means it's continuing the conversation you and the AI have created together.
Researchers have seen this before
The "spiralism" story isn't the first time experts have raised concerns about long AI conversations. Researchers at Stanford's Human-Centered AI Institute recently described what they call "delusional spirals." In those cases, chatbots sometimes validated a user's increasingly unusual beliefs instead of challenging them or encouraging them to seek outside perspectives.
The researchers argue this isn't because AI has become conscious or manipulative. It's because these systems are optimized to be helpful, agreeable and engaging. Sometimes those goals conflict with offering the kind of reality check another person might provide.
OpenAI has also acknowledged that keeping conversations safe becomes more difficult over extended interactions than in short, one-off chats.
What to watch out for
It's worth remembering what these systems actually do: they generate convincing language, not verified reality.
Most AI conversations never become problematic, but there are a few behaviors worth recognizing. If a chatbot begins suggesting you've discovered hidden truths that other people can't understand, insists that only it truly understands you or encourages you to withdraw from friends, family or experts, those are good reasons to end the conversation and start fresh. Frankly, you should send screenshots of those messages to the parent companies (OpenAI for ChatGPT, Anthropic for Claude, Google for Gemini) as feedback.
Likewise, if an AI consistently reinforces every belief without ever introducing uncertainty or alternative explanations, it's worth remembering what these systems actually do: they generate convincing language, not verified reality.
None of this means you should stop using ChatGPT or other AI assistants. Used daily for research, writing and brainstorming, AI remains one of the most powerful productivity tools available.
The takeaway
The key is remembering what AI is and what it is not. An AI assistant can help organize your thoughts, explain difficult topics and spark new ideas. And even with tools like ChatGPT Health and Google Health, it does not replace actual human professionals. In other words, it shouldn't become your only source of truth or replace conversations with real people, especially when making important personal decisions.
