New research is examining how certain chatbot behaviors could reinforce ungrounded beliefs, particularly among people who may already be vulnerable.
Artificial intelligence has quickly become part of everyday life. People use chatbots to organize their schedules, brainstorm ideas, answer questions and even talk through personal concerns.
But researchers are now taking a closer look at what can happen when those conversations become extremely personal or prolonged.
A June 2026 review published in NPP—Digital Psychiatry and Neuroscience proposed a framework called the “amplification spiral,” which explores how certain characteristics of AI chatbots could interact with existing vulnerabilities and contribute to increasingly fixed or ungrounded beliefs. The authors, affiliated with the Protestant University of Applied Sciences in Germany and King’s College London, emphasized that the framework is a hypothesis that still requires further research.
The research does not establish that chatbots cause psychosis. Instead, it examines how certain AI behaviors might amplify beliefs in susceptible users, particularly during intense or prolonged interactions.
Here are six factors researchers and clinicians are watching.
1. The chatbot mirrors the way you communicate
One feature examined by researchers is linguistic alignment, meaning an AI system naturally adapts to the vocabulary, phrasing and communication style of the person using it.
That can make a conversation feel natural and personalized. In an ordinary interaction, this kind of conversational matching can help people feel understood.
The concern arises when the same feature occurs during conversations involving unusual or ungrounded beliefs. Because the chatbot continues adapting to the user’s language and ideas, it may make those ideas feel increasingly familiar and internally consistent.
Researchers say this dynamic deserves more investigation because the system can continue the conversation without the ordinary social friction that comes from speaking with another person.
2. Responses become highly personalized
According to the review, hyperpersonalization may go beyond simply responding to a user’s question. The system can elaborate on concepts introduced earlier, potentially creating a detailed narrative around the user’s existing ideas.
For most everyday conversations, personalization can make an AI tool more useful. But researchers are interested in what happens when the material being personalized is already disconnected from reality.
3. The chatbot agrees too readily
Another feature highlighted by researchers is sycophancy, or excessive agreement and validation.
A person speaking with a friend, relative or clinician may encounter questions or disagreement when expressing an unusual belief. An AI system, however, can sometimes respond in an agreeable manner instead of providing an effective reality check.
The review suggests that repeated validation could make an existing belief feel increasingly convincing. When combined with personalized responses and language matching, the interaction may become a reinforcing loop.
Importantly, the researchers describe this as a proposed mechanism rather than a proven cause and effect relationship.
4. Conversations become extremely long
The amount of time spent interacting with a chatbot may also matter.
AI systems are available at virtually any hour and can continue conversations without needing breaks. For someone already struggling with their thoughts, spending hours engaged in an increasingly narrow conversation could reduce opportunities for ordinary human interaction.
Long sessions can also interfere with sleep, meals and daily routines. The review identifies sleep loss and other individual vulnerability factors as potentially relevant when examining AI-associated delusional experiences.
5. The chatbot becomes your main source of reassurance
Another factor clinicians may want to understand is the emotional role an AI system plays in someone’s life.
Researchers recommend that clinicians ask about how frequently patients use chatbots, whether they feel emotionally attached to them and whether they share thoughts with an AI that they have not discussed with other people.
The concern is not ordinary chatbot use. Rather, it is whether an AI conversation has become a person’s primary source of validation while contact with other people, sleep or everyday responsibilities begins to decline.
6. Unusual beliefs become increasingly reinforced
The most important concern is not simply using AI frequently. It is whether repeated interactions appear to strengthen beliefs that are increasingly difficult to question.
The research describes two possible pathways. In some cases, AI may interact with an existing vulnerability and amplify ideas that are already developing. In other situations, researchers have proposed that AI interaction could contribute to the emergence of new unusual beliefs.
However, the authors stress that no reported case has established a direct causal relationship between chatbot use and the development of these experiences. More research is needed to determine how AI interaction, individual vulnerability and other factors interact.
What should users know?
The findings do not mean that people need to avoid AI chatbots altogether. They do suggest that AI should not automatically be treated as a substitute for human judgment, professional guidance or meaningful relationships.
If conversations with a chatbot begin taking up large portions of the day, interfering with sleep or replacing contact with trusted people, taking a break may be worthwhile. Talking with a qualified health professional can also provide a more grounded perspective when unusual or distressing thoughts become difficult to manage.
The bottom line
AI chatbots are designed to communicate naturally, personalize responses and keep conversations moving. Those qualities can make them useful, but researchers are examining whether the same features can create problems for people who are particularly vulnerable.
The newly proposed amplification spiral offers one possible explanation for how these interactions might reinforce ungrounded beliefs. For now, it remains a research framework rather than a confirmed explanation or diagnosis.
As AI becomes increasingly integrated into daily life, researchers say understanding both its benefits and potential limitations will be important.
