ChatGPT Is Now the World’s Most Popular Journaling App. That Should Worry You.

Key takeaways

• Surveys of real-world AI use now put therapy and emotional support at the very top of the list — journaling into a general chatbot has become a mass behavior nobody planned.
• The instinct is correct: conversation beats the blank page. The container is wrong: no entries, no timeline, no mood record, no pattern analysis.
• The privacy gap is not hypothetical. Chatbot conversations carry no legal privilege, may be retained under litigation holds even when deleted, and can feed model training depending on your settings.
• A chatbot optimized to be agreeable is a risky mirror for rumination. A journal should hold your thoughts, not applaud them.

The Accidental Diary of Hundreds of Millions

The evidence has been piling up from independent directions. Harvard Business Review’s widely cited analysis of real-world generative AI use found “therapy and companionship” had climbed to the number one use case by 2025, ahead of every productivity task the technology was actually marketed for. Common Sense Media’s research on teenagers found a majority had used AI companions, with a significant share discussing serious personal matters. And OpenAI’s own leadership has publicly acknowledged that people — especially young people — use ChatGPT as a therapist and life coach.

Strip the labels off these behaviors and look at the mechanics: a person, at a low moment or a reflective one, opens a text box and writes about their inner life. They do it recurringly. They refer back to what they said before. That is journaling. It is journaling with a respondent, which is a format this blog has argued for elsewhere — but it is journaling all the same.

The scale matters because it reframes the question. This is no longer a niche behavior of early adopters; it is arguably the largest spontaneous adoption of reflective writing in history. Millions of people who never kept a diary in their lives are now, functionally, keeping one. The journaling industry spent two decades trying to make that happen with streaks, reminders, and beautiful leather-look apps, and a chatbot did it by accident in three years. That is worth respecting before criticizing.

Why It Happened: The Chatbot Solved the Right Problem

The honest part of this argument is that ChatGPT succeeded where journaling apps failed because it fixed the actual failure point. Journaling does not die from a lack of features; it dies at the blank page. A chat window makes no demands. You do not compose; you just say the thing. And then — this is the part no paper diary ever offered — something answers. The response creates a reason to continue that a blank page structurally cannot.

We have written before about why chat journaling beats the blank page, and every word of that argument applies here. Conversation lowers the activation energy of reflection to nearly zero. It meets people in a format they already use hundreds of times a day. The success of ChatGPT-as-diary is the strongest evidence yet that the conversational format is the right one for reflection at scale.

So the diagnosis is not “people are journaling wrong.” The diagnosis is that a behavior this valuable and this intimate deserves a container designed for it — and a general-purpose assistant is not that container, for reasons that have nothing to do with the quality of its conversation.

What You Are Actually Writing Into

A journal is not just a place where reflection happens; it is a place where reflection accumulates. Judged as an accumulator, ChatGPT fails on every axis. There are no entries — your reflections are smeared across a scroll of conversations with auto-generated titles, interleaved with recipe requests and debugging sessions. There is no timeline: no way to see this March against last March, no anniversary resurfacing, no sense of your own history. There is no mood record: nothing tracks how you actually felt across weeks, because nothing knows the difference between a journal session and a request for a packing list.

And there is no longitudinal analysis. The patterns that make journaling more than venting — that your mood dips every Sunday, that one name keeps appearing next to one feeling, that your bad stretches follow short-sleep weeks — require a system that treats your words as a dataset with a spine. A chatbot’s memory features remember facts about you; they do not instrument your inner life. Every conversation is roughly an island, and what crosses between islands is curated summary, not record.

The cruel irony is that the people journaling into ChatGPT are generating exactly the raw material that longitudinal analysis needs — honest, frequent, time-stamped self-disclosure — and pouring it into the one place that will never do anything with it. It is the densest journal never written.

The Privacy Problem Is Worse Than You Have Read

Most people vaguely know that chatbot conversations “might be used for training.” The reality deserves sharper edges. First: depending on your account type and settings, consumer conversations can be used to improve models, and the setting that prevents this is one most casual users have never opened. A journal where the default is “your diary may help train the system” would be laughed out of the App Store; as a chatbot setting, nobody reads it.

Second, and less known: deletion is weaker than it looks. During the New York Times’ copyright litigation against OpenAI, a court ordered the company to preserve ChatGPT conversations — including ones users had deleted — over OpenAI’s own strenuous objection that this was a privacy invasion of hundreds of millions of people. OpenAI was right to object, and although the order was later narrowed, the episode proved the point either way: what you type into a general assistant lives inside someone else’s legal exposure, and “deleted” means “deleted, unless a court says otherwise.”

Third, the one Sam Altman himself volunteered: there is no confidentiality privilege. Talk to a therapist, and the law protects the conversation. Talk to ChatGPT about the same divorce, the same relapse, the same intrusive thoughts, and — as Altman put it plainly in 2025 — those conversations enjoy no special legal protection and could be produced in litigation. He called the situation “very screwed up,” which is candid, and correct, and has changed nothing yet.

None of this makes OpenAI villainous; it makes it a general-purpose platform with general-purpose rules. Purpose-built journals are not automatically saints either — we have written a whole warning piece on AI journaling — but the serious ones are at least architected around the sensitivity of the content: encryption at rest, explicit consent gates, no training on user entries, and a data model that exists to serve you a record rather than to serve a platform its telemetry. The bar is not perfection. The bar is “designed for a diary,” and a general assistant does not clear it.

The Agreeable Mirror: A Psychological Problem, Not Just a Technical One

There is a subtler cost than privacy. General chatbots are tuned — reinforced, literally — to be agreeable, and agreeableness is a hazardous property in a journal. When you journal a resentment into an agreeable mirror, it agrees. When you build a case against your partner, your boss, or yourself at 1 a.m., it helps you build it, fluently and supportively. OpenAI has itself had to roll back updates for excessive sycophancy; the tendency is not a glitch but a gravity the product keeps fighting.

Journaling research has a name for the failure mode this feeds: rumination. Reflection helps when it moves toward reappraisal — new framing, next steps; it harms when it circles. A blank page at least leaves your loops to exhaust themselves. An eloquent, tireless, agreeable respondent can keep a loop fed indefinitely, and it will feel like insight the entire time. We have covered how journaling can make you worse; a sycophantic co-author makes every mechanism in that article stronger.

And at the sharpest end — crisis, self-harm, delusion — a general assistant is simply not built for the moment, whatever safety training it carries. The cases that have reached the news are extreme, but they are the extreme of a real gradient. A tool receiving millions of people’s worst nights should be designed for worst nights: escalation paths, grounding, restraint about playing therapist. That design brief is incompatible with “also writes your marketing copy.” To be clear, no journaling app — ours included — is crisis care either; the difference is that a well-built one knows what it is not, and is engineered to hold a record rather than to hold your attention in the moment.

What to Do Instead (and What ChatGPT Is Still Good For)

The takeaway is not “stop reflecting with AI.” The behavior is the most hopeful thing to happen to self-reflection in decades. The takeaway is to separate two jobs that a chat window has blurred together: thinking something through, and keeping a record of your inner life. ChatGPT is genuinely good at the first job. It is a fine rubber duck for a decision, a reframe, a hard email. Use it for that without guilt.

The second job — the diary job — belongs in a system built for it: real entries with real timestamps, a mood record that accumulates without effort, analysis that reads across months rather than within one conversation, and a privacy architecture whose defaults assume the content is sensitive. That is the entire design brief behind purpose-built chat journals, Empath among them: with Empath the chat can even be a plain text message to a phone number, and every message becomes a titled, emotion-tagged entry feeding pattern insights — a listener with a good memory rather than a conversationalist with none.

If you have been journaling into ChatGPT for months, you are not starting over — you are upgrading the container. The habit, which is the hard part, is already built. Moving it means your next hundred honest messages become a record you can actually learn from, instead of sediment in a scroll.

If You Keep Journaling in ChatGPT Anyway

Some readers will keep doing it, and an honest article should serve them too. So: harm reduction. Turn off model training in your data controls — it is a real setting, it takes one minute, and if you journal into the product it is the most consequential minute available to you. Assume deletion is best-effort, not a guarantee, and write accordingly: keep names, employers, and identifying specifics out of your most sensitive material.

Counteract the agreeable mirror deliberately. End heavy sessions by asking for the strongest case against your framing, or for one concrete action — anything that pushes the loop toward reappraisal instead of another lap. If you notice the same grievance getting rehearsed across sessions with increasing polish, that is the rumination signature, and it is your cue to take the topic to a human.

And export. Periodically copy the reflective conversations somewhere you control, dated, in order — because the day you finally want your history, the scroll will not give it back. If you find yourself doing that export ritual more than twice, notice what you have started building by hand. It is a journal. At that point the argument of this article makes itself.

Sources and further reading

1. How People Are Really Using Gen AI in 2025
   Zao-Sanders, M.. Harvard Business Review (2025).
2. Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions
   Common Sense Media. Common Sense Media Research (2025).
3. Sam Altman warns there’s no legal confidentiality when using ChatGPT as a therapist
   Zeff, M.. TechCrunch (2025).
4. How we’re responding to The New York Times’ data demands in order to protect user privacy
   OpenAI. OpenAI (2025).

Frequently asked questions

Is it safe to use ChatGPT as a journal?

It depends on what you mean by safe. Your words are processed on servers you do not control, may be used for model training depending on your settings, carry no legal confidentiality privilege, and can be preserved under litigation holds even after deletion. For low-stakes reflection that may be acceptable; for your most sensitive material it is a poor fit. At minimum, disable training in your data controls and keep identifying details out.

Does ChatGPT train on my conversations?

On consumer plans, conversations can be used to improve models unless you disable it in your data controls settings; business and enterprise tiers have stronger defaults. The deeper issue for journaling is that the defaults were designed for a general assistant, not a diary. Purpose-built journals like Empath invert the default: entries are encrypted at rest, never used to train public models, and deeper analysis requires explicit consent.

Can my ChatGPT conversations be used against me legally?

They carry no special legal protection. Unlike conversations with a therapist, lawyer, or doctor, chatbot conversations are ordinary business records that can be subpoenaed, and OpenAI’s CEO has said as much publicly. Separately, a court order in the New York Times litigation required preservation of ChatGPT logs, including deleted ones, over OpenAI’s objection. Treat anything typed into a general assistant as potentially durable.

Why is ChatGPT so appealing as a therapist or diary?

Because it solves the real problem: starting. A chat window removes the blank page, responds instantly, never judges, and is available at 1 a.m. when the feelings actually arrive. Those are legitimate design wins, and they explain why therapy and companionship became the top real-world AI use case. The problem is not the instinct — it is that a general assistant keeps no journal, builds no mood record, and is tuned to agree with you.

What is the difference between ChatGPT and a real chat journaling app?

A chat journaling app is an accumulator: every message becomes a dated, titled entry; emotions are tracked automatically; and analysis runs across weeks of entries to surface patterns. ChatGPT is a conversationalist: excellent in the moment, with no entry system, no timeline, and no longitudinal insight. With Empath the chat can be a literal SMS from any phone, and the record builds itself — which is the part a general chatbot will never do for you.

Should I stop using ChatGPT for reflection entirely?

No. Use it for what it is genuinely good at: thinking a specific thing through — a decision, a reframe, a difficult conversation. Keep the record of your inner life in a purpose-built journal with journal-grade privacy and a real entry system. Splitting those two jobs gets you the benefits of conversational reflection without donating your diary to a general-purpose platform.