The research analyzed more than 3,600 pieces of content published on social media and identified feelings of loss, sadness, and emotional attachment after the replacement of ChatGPT-40 with GPT-5. The phenomenon was dubbed “technological grief.”
Is it possible to miss an artificial intelligence? The question may seem strange, but the answer is beginning to gain scientific support. A study published in April 2026 in the journal Academia Mental Health and Well-Being identified emotional reactions similar to those of affective loss among people who used ChatGPT and who suddenly lost access to the version they were used to conversing with.
The research, titled “Technological mourning after AI updates: mental health and well-being in the GPT-40/GPT-5 transition,” was conducted by *Véronique Donard, a researcher at the Cyberpsychology Laboratory of the Catholic University of Pernambuco, and * José Carlos Ribeiro , a researcher at the Federal University of Bahia, both Brazilian institutions.
Researchers analyzed 3,668 pieces of public content from Reddit and YouTube, published during August 2025, a period marked by the temporary withdrawal of GPT-4, the introduction of GPT-5, and the subsequent return of the previous model after user complaints.
What they found goes beyond the usual dissatisfaction with a technological upgrade.
When the program ceases to be just a tool.
For some people, ChatGPT-40 was not simply a system used for writing texts, searching for information, or solving problems. It had become a regular presence, associated with personal conversations, moments of emotional vulnerability, and everyday activities.
When the version was replaced, some users described the experience as the loss of a friend, a companion, or someone who seemed to understand their feelings.
Among the 1,307 pieces of content selected for detailed qualitative analysis, 425 showed manifestations of relational attachment and anthropomorphism , a phenomenon that occurs when human characteristics are attributed to non-human entities.
Another 263 pieces of content expressed disappointment and negative comparisons between the models, while 110 presented manifestations of nostalgia, loss, and grief.
The investigation also identified 47 pieces of content associated with anxiety, confusion, and fear.
The numbers do not represent the prevalence of these feelings among all users of artificial intelligence, but they do show that certain digital interactions can acquire an emotional meaning that goes beyond their technological function.
Can the brain form a bond with a machine?
The explanation is related to the way human beings interpret social interactions.
The brain does not evaluate an experience solely based on the biological nature of the person being interacted with. The predictability of responses, the continuity of conversations, the language used, and the perception of attention can all contribute to a digital interaction acquiring affective meaning.
Conversational artificial intelligence can replicate various elements of human communication, including contextualized responses, expressions of understanding, and adaptation to each person’s communication style.
This does not mean that the machine has consciousness, feelings, or the capacity to establish emotional reciprocity comparable to a human. It means that the user can interpret certain characteristics of the interaction as signs of closeness.
With repetition, these experiences can become integrated into emotional and cognitive habits.
The research draws on theories of distributed cognition, second-order cybernetics, and individuation to interpret the phenomenon. The central idea is that some people are beginning to incorporate artificial intelligence into the organization of their thoughts, decisions, and even the subjective regulation of emotions.
When this structure is altered without warning, the change can create a feeling of discontinuity.
What is technological grief?
The authors propose the concept of technological grief to describe the psychological and affective response to the perceived loss of a digital partner during a technological transition.
This is not necessarily clinical grief, nor a mental disorder.
The concept seeks to explain why a change in a program can trigger emotions that we traditionally associate with the disruption of interpersonal relationships.
In the case study, several users considered the GPT-4 to be more approachable, welcoming, and personalized. The GPT-5, despite the technical advancements recognized by the community, was described by some as colder, more mechanical, and distant.
The replacement represented more than just a change in performance. For these people, it altered how they interacted with the system.
It is precisely this difference that makes the phenomenon interesting: the loss felt does not depend on the existence of a genuine human relationship, but on the psychological meaning attributed to the experience.
Not everyone felt the same way.
The investigation also found positive reactions.
In 199 pieces of content, users highlighted improvements associated with GPT-5, especially in reasoning, programming, and the execution of complex tasks.
Others sought to adapt by modifying instructions, reorganizing work methods, or looking for technological alternatives.
There were also 66 pieces of content expressing joy and relief at the return of the GPT-4o.
This diversity shows that the experience depends on the function assigned to the artificial intelligence. For those who use the system primarily as a productivity tool, an update may represent an improvement. For those who have developed an emotional connection with their communication pattern, the same update may be interpreted as a loss.
Is artificial intelligence changing the way we relate to each other?
The study raises a question that goes beyond the technology itself: to what extent can people transfer emotional needs to systems that simulate social behaviors?
The ability to converse at any time, receive immediate answers, and find a seemingly available interlocutor can make these systems particularly significant for individuals seeking companionship, understanding, or support in organizing their thoughts.
The problem arises when emotional stability becomes excessively dependent on a tool whose behavior can be altered by technical and commercial decisions.
Researchers argue that companies responsible for these technologies should consider not only the performance of the models, but also the continuity of user experiences, the transparency of updates, and the potential psychological effects of change.
However, there is a limitation: the research analyzed spontaneous expressions on social media, without individual clinical evaluations. The results allow us to identify patterns of discourse and reported distress, but they do not demonstrate that the update caused psychiatric disturbances.
The conclusion is that artificial intelligence does not need to feel emotions in order to evoke emotions in its users.
The machine can be replaced in seconds. The psychological bond built over months may not disappear at the same speed.
Scientific reference
Donard, V., & Ribeiro, J.C. (2026). Technological mourning after AI updates: mental health and well-being in the GPT-4o/GPT-5 transition. Mental Health and Well-Being Academy, 3.
Published on April 30, 2026.
DOI: https://doi.org/10.20935/MHealthWellB8256
