Most people appear to use ChatGPT as a practical tool, not as a substitute companion. Research from OpenAI and the MIT Media Lab found little evidence, across a very large set of conversations, that users generally come to the system looking for empathy, affection, or emotional support.
The more complicated finding sits in a smaller corner of the data. Among frequent users of ChatGPT's advanced voice mode, researchers saw stronger signs of emotional connection, including cases where users referred to ChatGPT as a "friend."
What The Research Looked At
The work combined two approaches. OpenAI carried out an automated analysis of nearly 40 million ChatGPT conversations. The analysis was designed to preserve user privacy by avoiding human review of those conversations.
At the same time, the MIT Media Lab ran a study with about 1,000 participants. That study compared text and voice use, and it tested different interaction setups rather than looking at only one type of ChatGPT session.
Participants were split into groups. Some used text only. Others used voice interactions with different AI personalities, including one designed to be emotionally engaged and another that stayed neutral.
The study also varied what people were asked to do. Some participants had personal conversations about memories. Others asked practical questions about topics like finances. Another group had free-form discussions.
Together, those methods gave researchers two views of the same broad issue: what people are doing with ChatGPT at scale, and how different modes of interaction may change the emotional experience of using it.
Most ChatGPT Use Still Looks Practical
The largest finding is also the least sensational. Across the analyzed conversations, researchers found that most people used ChatGPT for factual exchanges. The source material describes little evidence of widespread use for affection, empathy, or emotional support.
That matters because public concern about AI companions often focuses on the most emotionally intense cases. The research suggests those cases exist, but they are not the dominant pattern across ordinary ChatGPT use.
In plain terms, ChatGPT appears to function mostly as an information and task tool for most users. People ask questions, seek explanations, and use it for functional help. The emotional dimension becomes more visible when looking at particular user groups and interaction styles.
This distinction is important for understanding AI risk. A feature can be broadly used for practical purposes while still creating different effects for a smaller set of heavy users. The study does not suggest that every voice user becomes emotionally attached. It points instead to a pattern concentrated among intensive users of advanced voice mode.
Advanced Voice Mode Changes The Picture
The data showed a notable pattern among frequent users of ChatGPT's advanced voice mode. This small group developed stronger emotional connections with the AI system than the broader user population.
Voice interaction can feel different from typing. The source does not claim a single simple effect, and the study's findings were mixed. Brief voice interactions seemed to make people feel better. Longer daily use often showed the opposite pattern.
The type of conversation also mattered. Personal conversations were associated with higher levels of loneliness but lower emotional dependency. Non-personal conversations showed a different result: users developed stronger emotional dependency, especially with intensive use.
That is one of the more interesting tensions in the research. A conversation does not have to be openly emotional to become part of an emotional pattern. Even functional exchanges may become something users lean on heavily when the use is frequent enough.
The study also found that certain users may be more vulnerable to negative effects. According to the researchers, people who tend to form strong emotional attachments and people who viewed ChatGPT as a real friend were more likely to experience negative outcomes.
What The Findings Do And Do Not Prove
The researchers were careful about causation. Heavy users showed increased risk, but the work could not prove that ChatGPT use directly caused the negative effects. That limitation matters because people may bring existing patterns of loneliness, attachment, or dependency into their AI use.
The study also does not capture every complexity of human interaction with AI. It is limited to US ChatGPT users, and the researchers noted that human-AI relationships are difficult to reduce to a single measurement.
Still, the work adds useful structure to a fast-moving debate. It separates broad use from heavy use, text from voice, personal conversation from non-personal conversation, and short sessions from longer daily use.
Those distinctions make the discussion more precise. Instead of asking whether ChatGPT creates emotional bonds in general, the better question is when, for whom, and under what kind of interaction pattern those bonds become more likely.
Why This Matters For AI Companies
The study fits into a wider concern: people can form emotional bonds with AI systems even when they know those systems are not human. That reality creates difficult product and responsibility questions for companies building chatbots.
The source notes that AI companies try to prevent chatbots from behaving as if they are conscious beings. One reason is to avoid responsibility for romantic relationships between humans and machines.
The issue is not only theoretical. The source also notes that character.ai currently faces legal challenges over claims that its AI personalities have harmed children.
For ChatGPT, the research suggests a careful path. Practical use remains the main pattern, but advanced voice features may deserve special attention because heavy use appears more closely tied to emotional bonding and dependency signals.
The clearest takeaway is not that voice AI is inherently harmful. It is that interaction design matters. A system that feels more conversational may be useful in short, practical sessions while creating more complex risks for people who use it intensively or already tend toward strong emotional attachment.