DraftKings Tells Congress It Does Not Use AI to Target Bettors Likely to Lose

DraftKings has answered the three Democrats questioning its VIP programme with a flat denial that it targets problem gamblers or likely losers. It confirms that machine learning picks who receives promotions, and it declines to give the loss and revenue figures the lawmakers asked for.

• DraftKings told Senator Richard Blumenthal and Representatives Paul Tonko and Valerie Foushee in a six-page letter dated 5 November that it does not target anyone based on indicators of potential problem gambling
• The company says it does not use AI to find the customers most likely to lose, but confirms it uses "data science and machine learning, not generative or agentic AI" to choose who receives promotions
• It declined to give the average losses of VIP customers, calling exact figures "competitively sensitive", and said it does not report revenue by rewards tier
• Fewer than 1% of customers are in its top two tiers, Diamond and Onyx, and the company says it knows of only one celebrity video sent to a customer
• The letter answers the 22 October requests to six operators, which cited the New York Times report on DraftKings' promotional models

DraftKings Answers the VIP Questions on the Deadline Day

DraftKings Inc. replied to the lawmakers in a letter dated 5 November, the deadline they had set, signed by the company rather than a named executive. InGame, a trade publication, obtained and published it. It restates and answers each of their questions in turn.

On targeting, the company is categorical: "DraftKings does not target anyone based on indicators of potential problem gaming, nor does it seek to identify the customers most likely to lose and get them to bet more." It says customers under self-exclusion, cool-off periods, suspensions or bans are on an ineligibility list that is refreshed almost hourly and applied before any email, push notification or in-app message is sent through its messaging platform, Braze. Customers returning from a break get no promotions for a further 10 to 30 days.

The company describes its loyalty scheme, renamed DraftKings Rewards from Dynasty Rewards in August 2026, as covering every customer from the day the account opens, across five tiers from Bronze to Onyx. Fewer than 1% sit in Diamond and Onyx, the tiers that can include a personal host. Because enrolment is automatic, DraftKings argues, there is no relevant figure for losses before joining. On losses after joining, it points to the My Stat Sheet tool that lets customers see their own results, and says exact figures are competitively sensitive. It does not report revenue by tier.

Machine Learning Picks the Audience, Hosts Pick the Offer

DraftKings says it does not use AI to target likely losers or customers showing signs of problem gambling, but that it uses machine learning "to target promotions based on customer preferences, a basic marketing practice". It estimates that 80 to 90% of its sportsbook promotions are site-wide offers, though the amount, which it calls generosity, can vary by customer. The rest go to cohorts or come from hosts. Its models choose who receives a promotion and which standard offer fits, and are designed so that a customer whose usual maximum stake is $10 to $15 is not prompted into a much larger bet, the letter says.

Hosts tailor offers individually using the Salesforce customer relationship management system, a process the company says is not automated. They must escalate signs of problem play to the Responsible Engagement Team and each month contact 10 randomly assigned customers to ask whether they are playing within their means; hosts who miss any of those checks in a quarter lose incentive pay for it. Asked whether hosts are coached to build relationships that encourage gambling, DraftKings says their primary goals are "an engaging experience that encourages long term retention" and winning customers from rivals. It knows of one Cameo video, of New York Knicks player Miles McBride, sent to a customer.

The AI Denial Is Narrower Than It First Reads

The New York Times reported in September on a 2023 model that scored online casino customers by how much they were expected to lose for each promotion. DraftKings' letter denies seeking out the customers "most likely to lose", but it does not say which inputs its audience-selection models use, whether past losses are among them, or whether the 2023 casino model is still running. Its 80 to 90% estimate is given for sportsbook promotions only, and the casino product at the centre of the report is not addressed. The line it draws between machine learning and "generative or agentic AI" is narrower than the question, which asked about AI used to target gamblers with perks.

The Numbers That Would Settle It Were Withheld

The lawmakers' first question asked for the average loss of VIP customers before and after enrolment. DraftKings says it read the questions as aimed at its top tiers, then answered the enrolment question for its whole customer base, where automatic sign-up makes it meaningless. It could have given figures for customers moved into Diamond or Onyx. The same three lawmakers called FanDuel's September answers misleading and incomplete.

The Letter Will Be Read in Baton Rouge as Well as Washington

The lawmakers cannot compel answers, but Louisiana's regulator can. Its board extended DraftKings' permit only to February and required formal responses on its use of AI, which will now be read against the account given to Congress.

DraftKings has given Congress a detailed account of its safeguards and a firm denial on intent. It has not yet given anyone the loss data that would show whether that account holds.