AI Can Coach You at Video Games — But Players Still Want Humans

Coaching is now a side hustle. Researchers at Carnegie Mellon interviewed 20 experienced freelance game coaches who sell lessons on Fiverr, the gig marketplace where people hire strangers for small jobs. The coaches taught 17 different competitive games, from shooters to strategy titles. None had formal training, and most had never met each other. Yet they all ended up doing the same three things: building a friendly rapport, figuring out exactly what each student does wrong, and adjusting their advice in real time.

The study names two forces that shape this work. The first is "dual precarity" — a fancy way of saying these coaches live on two shaky incomes at once. Gig platforms can cut them off or bury them in search results, and the games themselves keep changing, so a coach who is an expert today can be out of date next season when the developers release a patch. The second is "earned authority": coaches only get taken seriously if they can show they're genuinely good at the same game they're teaching. Rank and reputation are the resume.

Here's where AI comes in, and where it doesn't. Coaches welcomed AI for the boring back-office stuff — scheduling, summarizing a student's match history, spotting statistical patterns. What they rejected was AI in the live session, the actual conversation where a nervous player is being told they're doing it wrong. That moment depends on trust, tone, and reading the room, and the coaches didn't believe software could do it.

The takeaway goes beyond gaming. If AI is going to enter coaching, tutoring, therapy, or personal training, this study suggests the winning setup is a hybrid: let the machine handle analysis and paperwork, and keep the human for the part where someone needs to feel understood. The researchers will present the work at CHI Play 2026, a conference on games and people.