← All articles

The AI Companion Boom That Mostly Isn't: What Usage Data Says People Actually Want

October 6, 2026 · 2 min read

If you only read headlines, you'd think millions of people are quietly falling in love with chatbots. The usage data tells a more interesting story. When researchers actually count what people do with general-purpose AI assistants, emotional companionship turns out to be a sliver of the picture, and the real action is in far more ordinary places. That gap between the narrative and the numbers matters if you're choosing, building, or just trying to understand these tools.

Most use isn't about work, and isn't about feelings either

The largest study of ChatGPT usage to date, a National Bureau of Economic Research working paper from OpenAI and Harvard economists, found that non-work messages grew from 53% to more than 70% of all usage as of mid-2025. That sounds like a personal-life takeover, but look at what those messages are. "Practical guidance," "seeking information," and "writing" together account for nearly 80% of conversations. People are asking how to fix things, what something means, and for help with an email. The assistant is acting like a patient, always-available generalist, not a confidant.

Companionship is rare, even where you'd expect it

Anthropic ran a similar analysis on Claude conversations, using a privacy-preserving method that aggregates usage without exposing individuals. Emotional support and personal advice made up about 2.9% of conversations. Companionship and roleplay combined came to less than 0.5%, and romantic or sexual roleplay was under 0.1%. Most of Claude's usage was work-related, with content creation leading.

Two caveats are worth keeping in mind. These are single-vendor snapshots of general assistants, so they say little about dedicated companion apps, whose users chose a product built for relationships. And a small percentage of a very large number of conversations is still a lot of people. Rare does not mean unimportant.

What the quiet cases reveal

The more useful finding is about where the emotional conversations come from. When people do seek personal help, the topics skew practical: improving mental well-being, career transitions, and communication skills. And Anthropic observed that some long conversations drift from a task into something closer to companionship, even though the person started with a practical question. Those very long sessions were uncommon, but they point to a design reality: nobody signs up for a "companion." They start with a task and, occasionally, the relationship forms around it.

What this means for you

First, be skeptical of market-size headlines. Many "AI companion statistics" roundups circulate numbers that trace back to a single aggregator, and the careful research tells a narrower story. Second, if you're picking an assistant, judge it on the work it does well, because that's what you'll use it for most. Third, if you do use AI for emotional support, treat it like any tool with limits: it can help you think through a hard conversation, but it isn't a substitute for the people in your life.

The takeaway is simple. The most common AI relationship is not a romance. It's a useful, slightly overqualified helper that you consult a dozen times a week about ordinary things. Design and expectations should start there.

Sources: How People Use ChatGPT (NBER Working Paper 34255); TechCrunch on Anthropic's companionship research; Axios on Claude emotional support usage