NYT: DraftKings AI Targeted Bettors Likely to Lose More

DraftKings AI targeting steered promotions toward customers expected to gamble and lose more, according to a September 19 New York Times investigation. The report also describes shelved gambling-risk projects; DraftKings denies loss-based targeting.

What the DraftKings NYT investigation describes

The Times describes a 2023 casino model called an elasticity score, estimating customers’ response to incentives.

The Times also reports that predictive projects designed to flag gambling harm were shelved. Their effectiveness is not publicly established. DraftKings denies targeting customers based on losses and defends its existing monitoring.

The incentive behind personalized promotions is documented in DraftKings’ own financial disclosures. Its 2025 Form 10-K says return-on-investment models, customer segmentation and retention help determine promotional levels. It also explains that customer incentives reduce reported revenue. The business question is whether the additional activity an offer encourages justifies its cost.

That is why the objective of a model matters. A system optimized for commercial returns does not, by that objective alone, measure a customer’s well-being. The same information can support different decisions depending on what the operator asks the system to predict and what action follows its output.

Dyutam’s background on AI in gambling explains the broader use of behavioral data. Here, the practical distinction is between deciding who receives an incentive and deciding who needs an intervention.

Promotional scoring and problem-gambling risk prediction

These functions should not be treated as interchangeable. A voluntary assessment depends on a customer choosing to participate. Account monitoring examines recorded behavior. Predictive screening attempts to anticipate an outcome. Evidence that one of these functions exists does not establish the performance of another.

Nor does expected loss identify the result of an individual session. The distinction between house edge and an individual gambling outcome matters when interpreting claims about profitable customers. A marketing prediction is not evidence that a particular game has been manipulated.

DraftKings’ existing tools are part of the record

On September 14, five days before the investigation, DraftKings announced Gamalyze American Football, a responsible-engagement campaign and expanded cool-off options. The release dates Gamalyze Casino’s integration into My Budget and Controls to January 2026 and names My Budget Builder, My Stat Sheet and player-set limits among its tools. These initiatives predate the article; they are not a response to it.

According to that announcement, customers can select custom cool-off periods between three and 364 days. The company also described a campaign featuring Kevin Hart and Nick Jonas. These are statements about available controls and education, not measurements showing how much gambling harm those initiatives prevent.

Mindway’s separate GameScanner product illustrates the distinction. Its product description concerns operator-facing analysis of betting patterns and risk profiles. Gamalyze is a voluntary self-assessment. A partnership involving one product should not be described as deployment of the other.

That distinction also matters when evaluating DraftKings’ public responsible-gaming messaging. Education, controls and operator intervention serve different purposes. Announcing them does not settle which approach reaches customers at the right time.

What the research can establish

A January 2026 study by Michael Auer and Mark D. Griffiths examined whether betting records could identify self-reported gambling problems. It supports the possibility of identifying risk from behavior, but it did not test DraftKings or demonstrate that an intervention prevented harm. Its reliance on self-reported outcomes, the sample’s limits and disclosed industry relationships are relevant qualifications.

A Massachusetts Gaming Commission research summary also identifies uneven evidence for behavioral indicators, alongside privacy, consent and cross-operator data barriers. The useful distinction is between recognizing a warning sign, predicting a future problem and successfully preventing harm. Each requires its own evidence.

For a reader assessing a protective tool, the central question is what happens after a flag. A score or a message is an intermediate step. A claim of effectiveness needs evidence about the outcome the intervention was intended to change.

Existing safeguards and proposed AI gambling restrictions

New Jersey’s Division of Gaming Enforcement launched a behavioral-monitoring initiative on January 1, 2023. Its announcement requires operators to examine recorded activity for warning signs and use escalating interventions. Examples include increasing gambling time, repeated cool-offs and visits to self-exclusion pages without completing exclusion.

That framework documents a regulatory obligation to use information for protection. It does not independently establish how effectively any particular operator applies it. The distinction is between a requirement on paper and evidence of its implementation and results.

New York’s March 2026 draft goes further on promotional AI: it proposes restrictions on personalized promotions and wager suggestions, alongside identification and support of at-risk patrons. As checked on September 20, the commission lists it as pre-proposal material. The SAFE Bet Act, introduced in Congress on March 13, 2025, also proposes restrictions on AI-created individual sports-wagering promotions. Neither proposal should be presented as an enacted response to this investigation.

The accountability question behind the reaction

In the X threads reviewed for this article, some users treated the findings as ordinary casino economics; others focused on the obligation to intervene. That small selection is not a measure of public opinion. The commercial incentive to retain profitable customers does not answer how a company should identify and support customers experiencing harm.

The public documents support a more specific test: identify what each system is designed to do, establish whether it is used, and examine the evidence for its results. Customers seeking information about warning signs and exclusion options can consult Dyutam’s responsible-gambling resources.

FAQs

---

Sources

• September 19 DraftKings investigation — The New York Times
• 2025 Form 10-K — DraftKings
• September 14 responsible-engagement announcement — DraftKings
• GameScanner product description — Mindway AI
• Using machine-learning algorithms to predict self-reported problem gambling — Michael Auer and Mark D. Griffiths
• AI and player risk identification and response: research snapshot — Massachusetts Gaming Commission
• Online betting behavior and at-risk player initiative — New Jersey Division of Gaming Enforcement
• Potential responsible-gambling rulemaking, Part 5404 — New York State Gaming Commission
• Proposed rules and pre-proposal status — New York State Gaming Commission
• SAFE Bet Act, S.1033, introduced text — United States Congress