The Journal That Answers Back

It is two in the morning, and someone, somewhere in the world—probably in millions of places at once—is writing to an artificial intelligence about something they have told no one. The fear of losing a job. The guilt after an argument with a father. The grief that refuses to pass. The person writes, the machine answers, and the person keeps writing.

The easy reaction is alarm: people are telling their secrets to a robot. But there is another reading, less dramatic and more interesting. Without realizing it, many of them are practicing a new version of one of psychology's most studied brief exercises: putting a difficult experience into words. The difference is that, for the first time, the private journal answers back.

In 1986, psychologist James Pennebaker and Sandra Beall published an experiment involving 46 college students. On four consecutive days, some wrote for fifteen minutes about traumatic experiences or deep emotions; others stayed with superficial topics. Over the next six months, the first group made fewer visits to the university health center. It was a small study, but it opened a question that remains alive: what changes when an experience finds words?

Over the years, more than two hundred studies followed, involving different populations and results that were less tidy than the original story. Some observed changes in immune response, stress, or wound healing. Reviews also found effects that were small, variable, or difficult to generalize from one person to another. Among adolescents, for example, a review of 21 studies found a small overall benefit that was promising but not yet decisive.

The reasonable conclusion is not that writing heals or that it is enough to move through trauma. Expressive writing may help some people in some circumstances. It does not replace therapy, it does not work the same way for everyone, and it does not always bring immediate relief. Sometimes writing about something recent or overwhelming increases distress before any benefit appears.

What seems most interesting is not simply the act of venting. Pennebaker observed that the work of turning a scattered experience into a narrative mattered: choosing what happened first, recognizing an emotion, finding a connection between two events. What remains unnamed can continue to operate like an open threat. Once it has a form, the brain can begin to place it in the past.

The goal is not beautiful writing. It is the construction of meaning. An awkward sentence that finally names a fear may do more than a polished page that circles around it. When a benefit appears, it grows from the effort to organize experience in one's own words.

That is partly what happens in many late-night conversations with artificial intelligence. Someone arrives with a confused feeling, writes one line, receives a question, and adds another. The machine can ask for a date, separate a fact from an interpretation, or point out that the same idea has appeared three times. A notebook listens; this notebook also asks questions.

Availability changes something. There is no need to wait until morning, find an appointment, or overcome the embarrassment of beginning in front of another person. At two in the morning, the screen is there. For someone who needs to take a first step, that absence of judgment can open a door that had seemed closed.

Yet the same ease contains the trap. The exercise works because the person does the organizing. If they hand over a handful of emotions and receive an elegant interpretation within seconds, they may recognize every sentence and still have processed nothing. It is like sending someone else to the gym and expecting the muscle to appear in our own body.

Artificial intelligence helps when it behaves like a discreet coach: it asks a question, reflects a contradiction, and leaves room for an answer to mature. It can undermine the process when it becomes an employee of the soul, explaining what we feel before we have tried to explain it ourselves. A polished answer can produce relief without producing understanding.

There is a simple test. After the conversation, can we describe what we understood without looking at the screen? Can we say it in words that are less perfect but our own? If the answer is yes, perhaps the tool supported real processing. If we can only copy the machine's conclusion, the order still belongs to someone else.

There is also a problem that a paper journal did not have: privacy. A conversation with artificial intelligence does not acquire the confidentiality of a professional consultation merely because it feels intimate. Depending on the service and its settings, text may be stored, reviewed for safety, or used to improve systems. Before writing names, diagnoses, or details that could identify someone, it is worth understanding where those words will be kept.

Caution does not require abandoning the tool. We can omit names, places, and exact dates; use temporary modes when they exist; review history controls; and delete conversations we do not need. Intimacy depends not only on what we choose to tell, but also on knowing who keeps those words and for how long.

The story has a curious turn. In the 1990s, Pennebaker and his colleagues began developing LIWC, a program that counted categories of words: pronouns, emotions, social terms, and other traces in language. They used it to study how certain patterns changed during writing. That software did not understand a life, but it helped show that the way we narrate one also carries information.

Three decades later, we converse with systems that can produce a coherent response from those traces. They seem to understand us because the language they generate adapts precisely to what we have just written. That experience can be valuable, but it does not make the machine a witness to our history, a therapist, or someone who assumes responsibility for the advice it gives.

The best way to use this journal that answers back may be to retain authorship. Write first without asking for an interpretation. Then, if useful, ask for questions: which part remains vague, which fact am I mixing with an assumption, what emotion appears but has not yet been named? The machine can hold the mirror; the person must make the movement of looking.

The classic protocol offers a starting point rather than a universal prescription: fifteen to twenty minutes on three or four days, focused on something that still occupies too much mental space. The text need not be shared or literary. If writing causes intense distress, disrupts sleep, or awakens memories that feel impossible to manage, the sensible step is to stop and seek support from a licensed mental health professional.

Not every pain needs an immediate explanation. Sometimes the mind requires silence, time, or the presence of another person. A tool available around the clock can make us believe every emotion must be resolved before the conversation ends. Some experiences cannot be put in order in a single night, and no automated response should rush them.

That is why the interesting debate is not whether artificial intelligence destroys intimacy or is about to become a therapist. The question is more personal: after talking to it, do we understand ourselves better, or do we merely possess an explanation that sounds good? The same system can support reflection or become a sophisticated way to avoid it.

At two in the morning, in front of a screen, someone begins writing again. Perhaps the machine asks the right question. Perhaps it gets it wrong. What matters will happen afterward, when that person closes the window and tries to say, in their own voice, what they have just understood. The journal now answers back. The story, however, must still be written by the person who lived it.

This note discusses research on expressive writing and offers general criteria for using artificial intelligence with care. It is not therapy or mental health advice.

Christian Ruggeri

Main source: Perspectives on Psychological Science.