Get paid to play video games? How gameplay data work really works

What gameplay data is

Gameplay data for AI is more than a screen recording. What makes it useful is the pairing: what the player saw, and which key, mouse movement or controller button they pressed at that moment, kept in sync. A model can learn from that pairing what a person tends to do next in a given situation. If you're wondering whether this work suits you, our guide for collectors covers pay, payout and how projects run.

It's a kind of AI training data made from recorded actions, and our explainer on what AI training data is shows where it fits among the other kinds. The job is also different from labeling or rating at a computer, and from recording physical tasks with a head camera. Our side-by-side guide to data collection compared with annotation jobs compares the two.

You don't need to stream, coach or win tournaments. You play naturally and let the screen and inputs be recorded the way the project's instructions describe.

Why AI research uses gameplay

Games make a handy practice ground for AI. They have rules, goals and clear feedback, and every action can be logged precisely. Several research projects use recorded play in different ways.

OpenAI's Video PreTraining (VPT) study trained an agent on gameplay video paired with keyboard and mouse actions. The researchers set up recording to "make data collection easier by allowing our human contractors to play the game without modification." The people they hired played as they normally would, and their play was captured as they went.

Google DeepMind's SIMA project, the "Scalable, Instructable, Multiworld Agent", trains agents "to follow free-form instructions across a diverse range of virtual 3D environments", including commercial video games. The GameNGen paper describes "the first game engine powered entirely by a neural model," trained to simulate a game in real time.

Gameplay is one part of wider research into how machines learn to act. For the bigger picture, see our overview of robot learning research.

Play-to-earn apps vs gameplay recording work

Warning signs to check

The FBI's public service announcement on play-to-earn scams warns: "Criminals advertise the apps as play-to-earn games offering financial incentives to players." Walk away if you see any of these signs:

• You're asked to deposit money or crypto to start, to "unlock" levels or to withdraw a balance.
• The app asks to connect to your crypto wallet, or asks for its recovery phrase.
• The balance grows fast, but you can't withdraw it without paying a fee first.
• The offer came out of nowhere, by text or chat, for a job you never applied for.
• Nothing is in writing about what you'll be paid, when and how.

The same tricks show up in the "task scams" the FTC describes: an app shows fake earnings, then asks you "to deposit your own money", usually in crypto, to get them out (FTC task-scam alert). In the first half of 2024 alone, about 20,000 people reported these gamified scams, according to the FTC's Data Spotlight. For more warning signs, see our checklist on how to spot AI training job scams.

Common mistakes

• Don't expect to be paid for winning. Recording work pays for the recording itself, whatever your score.
• Don't leave private things on screen. Close chat windows, notifications and anything personal before you start; what gets footage accepted explains why reviewers care.
• Don't mix it up with physical recording. Gameplay needs no camera on your head. Physical tasks do, and recording first-person video is a separate skill.
• Don't trust an app because it paid you once. Small early payouts are a known scam tactic to win your trust.

How TaskCapture handles gameplay

Gaming is one of TaskCapture's three data types. A gaming recording captures the screen plus every keyboard, mouse or controller input in sync, with no camera, voice, chat, gamer handles or data from the game itself. Applying is free, and you don't need any coding or AI background.

Pay is USD 5–15 per hour of accepted footage. It varies by project and is not guaranteed. Review usually takes about a week, with a contractual limit of 30 days. Once your footage is accepted, you're usually paid within 7 days; the contractual limit is 60 days after acceptance. Payment is in US dollars (USD). The Data Collector Agreement and Privacy Policy set out the terms, and how recording work works explains each step.

Questions