There is not yet evidence that ordinary AI chatbot use is making people generally less able to empathize with other people. Studies do find that people often rate human-attributed responses as more empathic than AI-attributed ones, and some patterns of chatbot use are associated with loneliness, emotional dependence, or problematic use. Those findings concern perceived empathy and users’ reported experiences—not a demonstrated decline in users’ empathy skills over time.
What “deskilling empathy” would mean—and what studies have measured
To show that chatbots are deskilling human empathy, researchers would need to establish that people’s ability to understand or respond to another person’s feelings declines after using them. That is different from asking whether a chatbot’s reply sounds caring, whether a user feels lonely, or whether someone becomes emotionally attached to a chatbot.
As an Amazon Associate I earn from qualifying purchases.
The studies discussed here measure those latter questions: judgments of responses, reported well-being and social experience, and associations with patterns of use. They do not directly demonstrate that users’ capacity to empathize with people has deteriorated. Keeping those outcomes separate is essential to interpreting the headline.
Free tools Windows power users keep installed
One-click scans. No signup required.
What the major studies found
| Study and design | Finding | What it does not establish |
|---|---|---|
| OpenAI and MIT Media Lab collaboration, March 21, 2025: automated analysis of nearly 40 million ChatGPT interactions, alongside targeted surveys and a four-week randomized study | Affective cues—such as empathy, affection, or support—were absent from the vast majority of assessed conversations. A small group of heavy Advanced Voice Mode users had a high proportion of emotionally expressive use. | The interaction analysis was automated, not a human-coded census of emotional relationships. These ChatGPT-specific findings do not establish how often people use all chatbots emotionally or whether users’ empathy skills change. |
| MIT Media Lab four-week randomized controlled experiment, March 21, 2025: 981 participants and more than 300,000 messages | Researchers varied text, neutral-voice, or engaging-voice interaction and open-ended, non-personal, or personal conversation. They detected no significant effects from the assigned conditions on measured outcomes. | The experiment measured loneliness, social interaction with people, emotional dependence on AI, and problematic use—not empathy skill. Its assigned-condition result is distinct from associations involving voluntary use. |
| Nature Human Behaviour, published June 30, 2025: nine studies with 6,282 participants | Human-attributed empathic responses were rated as more empathic and supportive than AI-attributed responses. Participants also preferred human interaction when seeking emotional engagement. | These were judgments about responses and their attribution, not evidence that participants became less empathic after using AI. |
| AAAI proceedings, “The Illusion of Empathy,” published April 11, 2025: 155 conversations across two datasets | GPT-based chatbots received higher ratings for conversational quality but lower ratings for perceived empathy than human partners. | The work examines perceptions of conversations; its comparatively small dataset does not measure long-term changes in users’ social skills. |
| JMIR Medical Education, 2026: quasi-experimental study of a role-prompted GPT-4o chatbot in medical-student communication practice | The study examines a chatbot as a possible tool for practicing empathic communication in a professional educational setting. | Medical-student practice does not establish the effects of consumer companion use, nor does it settle whether that use can have harms. |
Do AI chatbots make people less empathetic?
The evidence summarized here does not show that they do. The MIT Media Lab’s randomized study found no significant effects from its assigned interaction modes and conversation types on the psychosocial outcomes it measured during four weeks. Those outcomes included loneliness, social interaction, emotional dependence, and problematic use; they did not include a direct measure of participants’ empathy toward other people over time.
#1 Best Overall
The same study reported that participants who voluntarily used the chatbot more showed consistently worse outcomes, and that higher trust or social attraction was associated with more dependence and problematic use. Those are associations, not proof that increased use caused the outcomes. People who use a chatbot more may differ in important ways from people who use it less, and the study does not turn that relationship into evidence of empathy loss.
The OpenAI–MIT collaboration also reported nuanced associations: brief voice use was associated with better well-being, while prolonged daily use was associated with worse outcomes. Personal and non-personal conversations showed different associations with loneliness, emotional dependence, and problematic use. The authors cautioned against broad generalization. The findings do not establish that a particular interaction pattern changes a user’s empathy skill.
Rank #2
Why human responses can feel more empathic
In the nine Nature Human Behaviour studies, the same kind of supportive response could be judged differently depending on whether it was attributed to a person or to AI. Human-attributed responses received higher empathy and support ratings; participants’ belief that AI had helped produce a response presented as human also reduced perceived empathy and support. This is evidence that identity and attribution shape how empathy is perceived—not that AI use erodes the reader’s ability to empathize.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesThe AAAI analysis points to a related distinction: a conversation can score well on conversational quality and still be perceived as less empathic than a human exchange. Empathic-sounding wording alone may not make users feel that a chatbot understands them. A chatbot can generate language that appears supportive; these studies do not establish that it experiences empathy or provides reciprocal human care.
How far the findings can be generalized
The OpenAI–MIT work concerns ChatGPT, not every chatbot. Its controlled study involved adults in the United States who used English, and the study period may have been too short to detect lasting behavioral change. The collaboration identified self-report and classifier error as limitations; its findings had not been peer-reviewed when the report was published. Its interaction analysis was observational, and its emotional-cue classification should not be mistaken for a direct measure of people’s relationships.
Likewise, the sample sizes and interaction counts in these studies describe particular research projects, not the share of the public whose empathy has been affected. The findings do not support claims about children, other languages, other chatbot products, or long-term population-wide effects. The medical-education study is a different context from consumer companionship, so it cannot be used as a general verdict on companion chatbots.
Rank #4
What readers can reasonably take from this
It is reasonable to distinguish using a chatbot to produce supportive language from receiving care from another person. OpenAI and MIT Media Lab put the possibility of emotionally oriented use this way: “ChatGPT isn’t designed to replace or mimic human relationships, but people may choose to use it that way given its conversational style and expanding capabilities.” That is the collaboration’s statement about ChatGPT, not proof that chatbot conversations replace human relationships.
If you use a chatbot for emotional conversation, treat its fluent, empathic-sounding replies as generated responses rather than evidence of a reciprocal relationship. If your goal is to practice communication, a chatbot may offer a setting to rehearse; the medical-student study investigates that possibility, but does not establish that practice transfers to every real-world interaction. For claims that chatbot use is deskilling empathy, the missing evidence is direct, sustained measurement of users’ empathic abilities—not another rating of how caring a chatbot sounds.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
