The Massachusetts AI questionnaire asks licensed sportsbooks to explain their use of AI and machine learning by 3 November 2026. The Gaming Commission announced the request in a 14 October notice.
The letter, dated 9 October, describes participation as voluntary and says responses will remain confidential for policy development. It follows the earlier Massachusetts sportsbook AI review; the questionnaire is an information request, not a finding of misconduct.
What the questionnaire covers
Its scope includes automated decisions reviewed by humans. Questions cover vendors, promotional targeting, player losses, gambling-risk indicators, safeguards and the records explaining why an individual received an offer. Operators are asked to answer each listed sub-question separately.
The evidence behind a promotional decision
For operators and suppliers, our assessment is that preparing a useful account requires several teams to work from the same description of a system. Marketing may know the campaign settings, a supplier may maintain the model, and a responsible-gambling team may manage exclusions. Their explanations need to fit together.
A practical internal exercise could start with a recently completed campaign. Each team could identify the part it controlled and compare its account with the records available elsewhere. The useful result would be a coherent explanation of what happened, including any gaps the business cannot yet resolve. This could help teams agree which evidence is still missing before assembling a response.
That exercise could expose an important distinction between a documented control and a control that operated for the campaign being examined. A policy might describe a suppression check, for example; the relevant operational evidence would show whether that check ran, when it ran and which result the sending system received.
Supplier discussions could focus on access to those records and responsibility for correcting discrepancies. A general product description may explain what a platform is designed to do. An operator answering detailed questions also needs to understand how that description relates to the configuration its own team actually uses.