A sportsbook bonus might look like a routine marketing offer, but the technology deciding who receives it is facing new scrutiny. Massachusetts regulators want to understand whether sportsbook AI promotions can respond to customers’ losses, deposits and changing betting habits, potentially encouraging more wagering when restraint would be more appropriate.
The concern reaches beyond one company or one state. Conversations within TheRX betting forum regularly cover sportsbook promotions, account treatment and betting experiences. Such discussions can identify customer concerns, but they cannot establish how an operator’s algorithms selected an individual for a particular offer. That distinction between an observed promotion and its hidden decision process is precisely where regulatory questions begin.
On October 14, 2026, the Massachusetts Gaming Commission publicized a five-page questionnaire asking licensed sportsbook operators to explain how artificial intelligence and automated decision-making influence their marketing. Written responses are requested by November 3, placing the industry under a more specific examination of the relationship between customer data and betting incentives.
Massachusetts Wants to Know Who Controls the Algorithm
The commission’s latest action follows a September investigation by The New York Times alleging that DraftKings used machine learning to help identify customers likely to increase wagering and losses after receiving promotions.
DraftKings disputes that characterization. The company has denied using AI to target customers based on losses or indicators of potential problem gambling. No regulatory finding of wrongdoing has been established through the Massachusetts review.
The regulator’s response is broader than a single operator.
Commission Chair Jordan Maynard directed staff on September 24 to examine the use of AI and machine learning across licensed sportsbooks. Executive Director Dean Serpa subsequently issued the October 9 questionnaire, requesting information about the systems operators use, who develops them and whether automated recommendations require human approval.
The distinction matters because modern promotional systems may involve several connected technologies rather than one identifiable algorithm.
A customer-segmentation tool might categorize users by activity. Another system might estimate whether they will accept an offer. A separate application could determine when that message appears.
Accountability becomes harder when responsibility for those decisions is distributed among internal teams and outside vendors.
How Sportsbook AI Promotions Can Influence Player Behavior
Personalization itself is not evidence of harmful targeting.
A sportsbook might offer basketball promotions to customers who consistently follow basketball. That is different from selecting someone for additional incentives immediately after an unusual increase in deposits or wagering frequency.
The regulator’s questions about automated promotions specifically address possible inputs including betting frequency, wager size, deposits, withdrawals, net losses, session duration, previous promotional responses and customer-value classifications.
These are questions about potential practices, not proof that every operator uses those variables.
The deeper issue is the objective assigned to the model.
An algorithm trained to maximize promotional response might accurately predict who will place another wager. But a successful prediction does not establish that sending the promotion is appropriate.
Consider two customers who receive different bonus offers despite following the same sport. One could be responding to normal seasonal marketing. The other might receive an incentive after sharply increasing their wagering activity.
Without knowing the selection criteria, the difference remains unexplained.
The Difference Between Personalization and Player Targeting
The strongest regulatory test examines not just what an algorithm predicts, but how that prediction affects the customer.
The following comparison illustrates where ordinary personalization can become a consumer-protection concern.
| Marketing activity | Ordinary personalization | Potential targeting risk |
|---|---|---|
| Sports preferences | Offers reflect favorite leagues | Limited concern unless combined with risk signals |
| Deposit activity | Standard promotion available to eligible customers | Larger incentives follow rapidly increasing deposits |
| Betting frequency | Offers reflect normal customer activity | Messaging intensifies during unusually long sessions |
| Previous losses | Loss data informs protective monitoring | Recent losses increase promotional eligibility |
| Responsible-gambling controls | Risk indicators suppress marketing | Marketing continues despite relevant warning signals |
A critical distinction is whether risk signals reduce or increase promotional pressure.
A system that notices unusually frequent deposits and recommends a cooling-off intervention serves a fundamentally different purpose from one that interprets the same pattern as an opportunity to send another incentive.
That creates a measurable responsibility for operators: promotional optimization and player protection should not operate independently when they rely on overlapping customer data.
Why Regulators Need More Than an AI Policy
An operator can maintain a responsible-gambling policy without demonstrating that its algorithms follow that policy in every relevant decision.
Effective oversight requires information about model inputs, automated outputs, internal testing, exclusions, employee authority and decision records.
Massachusetts asks whether operators can reconstruct why an individual customer received a particular promotion on a particular date.
That question goes directly to algorithmic decision transparency.
For example, excluding direct financial losses from a promotional model may not resolve the concern if other variables still reflect related patterns of potentially harmful play.
A regulator therefore needs to examine outcomes alongside documented safeguards.
Research commissioned by the gaming commission also suggests caution. Its AI and player-risk research found that some proposed indicators of gambling harm lack strong supporting evidence, while payment-related indicators have a comparatively stronger evidence base.
Risk detection is consequently not as simple as assigning every customer a definitive vulnerability score.
The challenge resembles problems already encountered in sports betting fraud detection, where automated systems must distinguish meaningful warning patterns from innocent behavior.
Poorly designed controls can miss genuine risks or incorrectly restrict ordinary customers.
What Bettors Can Learn From Their Own Promotion History
Customers cannot independently inspect a sportsbook’s complete marketing model. They can, however, evaluate what reaches their accounts.
Reviewing promotional emails, app notifications and account activity over time may reveal whether offers tend to arrive after deposits, extended sessions or changes in wagering frequency.
A useful record includes the promotion date, offer type, eligibility conditions, recent account activity and whether previously selected marketing preferences were respected.
These observations are not proof of algorithmic targeting. Promotions may also reflect general advertising campaigns, sporting schedules or customer segments unrelated to financial losses.
Bettors should examine bonus conditions before participating and consider disabling optional promotional communications when the messages encourage unwanted activity.
Deposit limits, time-outs and self-exclusion tools provide stronger controls than relying on promotional messages to become less frequent.
A bonus is optional, regardless of how personalized or urgent its presentation may appear.
November 3 Will Test How Much Operators Can Explain
The November 3 response deadline is the next concrete milestone in Massachusetts’ review.
The commission has requested voluntary participation and indicated that responses will be treated confidentially. The questionnaire is not itself an enforcement finding, and there is no established timetable for subsequent regulatory action.
Its questions nevertheless establish a framework for evaluating automated promotions: which data influences an offer, whether risky behavior can increase its likelihood, and whether employees can explain or override a recommendation.
Those standards could help distinguish legitimate personalization from promotional practices that place excessive pressure on vulnerable customers.
For sportsbook AI promotions, the decisive question is not how accurately an algorithm predicts the next wager. It is whether the operator can demonstrate that its marketing decisions respect the same consumer-protection boundaries expected of its employees.
If a system can identify the moment someone is most likely to keep betting, regulators have a legitimate reason to ask whether that information should trigger another offer or a reason not to send one.