Can a game studio sell its data to AI? Tickets, telemetry and design documents

A game studio writes down more of its work than most companies. Every bug is reported, triaged, fixed, tested and mentioned in a release note. Every player problem becomes a ticket. Designers keep long documents explaining why a level or a system works the way it does. That trail is a record of how skilled people find and solve problems, which is what AI teams look for.

Whether the studio can sell or license it depends on three things: what the records are, who made them and whose information they hold.

What a studio keeps that an AI team might want

• Bug reports and fix histories. Each one is a small case: what happened, how to repeat it, who looked, what changed and whether the fix held. Researchers already study this kind of record. A study led by Truelove (posted on arXiv in March 2021) read the update notes of 30 popular PC games and examined 12,122 bug fixes, and found that crashes were the bug type with the highest severity. If researchers find the public version useful, a studio's own history, with its internal discussion, holds more.
• Player support tickets. Players write in their own words, agents answer and the ticket closes with an outcome. That structure suits training models that handle customer requests. Our page on support tickets explains what makes an archive valuable.
• Design and balance documents. Reasons for decisions, tests that were tried and dropped, and revisions over time.
• Source history. The commit log shows how code changed and why. This part has limits, covered below.
• Play logs. Records of what players did. These carry the strongest privacy questions.

Who made it

The employer is generally treated as the author of work employees create as part of their jobs. US copyright law says so in its definition of a "work made for hire" (17 U.S.C. sections 101 and 201(b)). That is why a studio can usually license its own documents, tickets and code written by staff.

Three groups of material do not follow that rule:

• Contractors. Work by freelancers and outside studios depends on the written agreement. If the agreement does not transfer the rights, the studio may not be able to license that work.
• Licensed parts. Music, art, fonts, middleware and engine code are licensed to the studio, not owned by it. A studio cannot pass on rights it does not hold. Open-source parts follow their own licenses, which AI teams read carefully.
• Player-made content. Mods and user-made levels usually follow the terms of service players accepted. Those terms vary, so read yours.

The practical step is a short audit: list what you wrote with employees, what you licensed in and what players made. Only the first group is ready to discuss.

Whose information is inside

Support tickets and logs contain players, and some of them are children. The GDPR (Regulation (EU) 2016/679, recital 30) says online identifiers, such as internet protocol addresses and cookie identifiers, can leave traces that identify people when combined with other information. Gameplay telemetry often holds exactly those. In the United States, the Federal Trade Commission enforces the Children's Online Privacy Protection Rule (16 CFR Part 312), which covers online services that collect personal information from children under 13. The FTC announced two settlements with a game maker in December 2022, including one under that rule, so regulators watch this area.

So the order of work is: take names, handles, emails, addresses, voice or text chat content and identifiers out of the records, then check what is left for combinations that could point to a person. We record what was removed. Our de-identification page explains the steps. For data about children, we will say plainly when a set cannot be offered.

What an AI team looks at

Counts, years and a sample. How many bug reports, from which years, across how many titles, with what proportion closed with a fix. Three complete cases, cleaned, tell a lab more than a long description. Our post on what to have ready before you talk to a data broker lists the facts to prepare.

Source code gets a closer look. Teams check the license of every piece, so engine and library code the studio did not write is not offered.

How a deal usually goes

You tell us what you hold. We sort it into what is yours, what is licensed in and what players made. We plan the removal of personal details, prepare a sample and ask AI labs whether they want it. You keep the final decision on every term, and most deals are licenses, so the studio keeps its games and its rights. Ask your counsel to read any license before you sign it.

The slides on this question walk through it, and the list of industries we work with shows the nearest pages, including software companies. The data value calculator takes about a minute. To talk it through, tell us what the studio keeps.