Engineering Engagement

The Slip That Costs Nothing

The award lands at 9:14 on a Tuesday morning, before Marcus has finished his coffee. Two hundred dollars in bonus bets, credited to his account, no deposit required, nothing asked of him. The notification does not say why. It arrives the way weather arrives.

He opens the app. The balance is there, in a separate line beneath his cash, styled a little differently, and the difference is the whole of what follows. He knows, in the way everyone knows, that the money is not quite money. He has read enough to understand that if he puts $200 on a coin flip and wins, he collects the profit and not the stake. What he has not done, because almost nobody does, is work out what the $200 is actually worth.

So he does what the instrument is built to make him do. He does not put it on a side at minus 110, where the return would feel thin and the win would feel small. He builds a four-leg same-game parlay on a Tuesday night baseball game, prices it at +1200, and taps. The bet loses within two innings. He is not especially bothered. It cost him nothing.

That last sentence is the one this article is about. The bonus bet did cost him something, and the amount is calculable to the cent. The parlay was not an indulgence he could afford because the money was free; it was the mathematically correct use of the instrument he had been handed, and the instrument was designed so that the mathematically correct use of it is the most expensive product the sportsbook sells. Nothing about that requires him to be irrational, addicted, or foolish. It requires only that he behave sensibly with what he was given.

The previous article in this series examined the machinery of the modern betting app and found that the popular neuroscience used to explain it is thinner than almost anyone repeating it realizes, while the commercial architecture is exactly as advertised. This article takes up the two features that sit on top of that architecture: the gamified loyalty program and the promotional bonus. Both are usually discussed in the language of psychology, and the psychology is real. But this article will argue that the psychology is not necessary. Three of the four largest American sportsbooks publish, in their own terms and conditions, a schedule showing which bets they most want their customers to make. The schedule is a ranking by house margin. It is not a manipulation that has to be inferred from behavior. It is a price list, and it is public.

Part I: The Rate Card

The Machinery of Play

The application of game-design elements to non-game settings has a canonical definition and a long research literature behind it (Deterding, et al., “From Game Design Elements”). The standard account of why it works runs through self-determination theory: people are motivated when three needs are met, competence, autonomy, and relatedness, and a well-built points-and-tiers system supplies plausible substitutes for all three. A progress bar advancing is a report of competence. A menu of missions to accept or decline is a supply of autonomy. A leaderboard is a supply of relatedness. None of these depends on the underlying activity going well.

That is the mechanism the industry itself describes, and it is coherent. What it is not, at present, is demonstrated inside a real-money sportsbook. A preregistered scoping review of deceptive design in online gambling published in 2026 catalogued the field and found sixteen articles in total, noting that evidence on behavioral impacts remains limited and is “hindered by restricted access to proprietary gambling operator data” (McGarrigle, et al., “Dark Patterns in Online Gambling”). There is a substantial empirical literature on leaderboards and badges in classrooms, in fitness applications, and in workplace software. There is essentially none on leaderboards and badges in an application where the points are earned by losing money. The operators hold the data and do not release it.

The conceptual work is further along than the empirical work. A 2025 taxonomy sorts online gambling platform design into sludge, friction inserted where the operator wants the customer to stop, and dark patterns, defaults set where the operator wants the customer to go, and argues that the digital platform is doing what the casino floor plan used to do (Newall, “Sludge, Dark Patterns and Dark Nudges”). That is a persuasive frame. It is not a measurement.

This is worth stating plainly rather than papering over, because the gap runs in a specific direction. It does not mean the mechanics are harmless. It means that anyone asserting a measured effect of streaks or missions on sportsbook betting is asserting something the public record does not contain. The argument this article makes does not need that assertion, and is stronger without it.

What the Programs Actually Are, in 2026

The loyalty landscape has been rebuilt almost entirely since the descriptions that circulate in most coverage. DraftKings retired Dynasty Rewards on 12 August 2026, replacing it with DraftKings Rewards and a new currency called Crown Cash across five annual tiers, Bronze through Onyx (DraftKings, “DraftKings Elevates Customer Loyalty Experience”). Crown Cash is dollar-denominated where the old Crowns were not, and it is explicitly non-withdrawable. FanDuel now runs three separate programs: a daily-fantasy Players Club, a Casino Rewards Club, and a Sportsbook Rewards Club still labelled Beta, with four monthly levels and an invitation-only tier above them (FanDuel, “Sportsbook Rewards Club Terms”). BetMGM retains Sapphire, Pearl, Gold, Platinum and the invitation-only NOIR. Caesars has quietly grown to six tiers, inserting Diamond Elite at 75,000 tier credits below Seven Stars at 150,000 (Caesars Rewards, “Benefits Overview”).

Around the tiers sits a layer of newer mechanics. FanDuel’s Casino Reward Machine gives three spins a day and asks the user to accumulate symbols across the week, with jackpot floors that scale with the player’s loyalty level, so that the daily-login loop feeds back into the spending ladder (FanDuel, “Casino Reward Machine”). In November 2025 FanDuel launched Pass The Leg, described as the first in-app collaborative parlay experience, in which a group of users each contributes a leg to a shared build of between three and twenty-five legs, and each contributor receives a profit boost for taking part (FanDuel, “FanDuel Introduces Pass The Leg”). In April 2026 it launched Bet Protect+, which sells full-game injury protection on National Basketball Association player props at a flat premium of three percent of the wager (Flutter, “Customers Wanted Full-Game Injury Protection”). That last one deserves a moment. It is not a promotion at all. It is the sportsbook selling the customer an insurance product against a risk the sportsbook created, priced and disclosed, and the company reported adoption at double its expectations.

Two corrections are worth making because they recur in secondary coverage. Bet Reset is not a new FanDuel product; it is the No Sweat Bet renamed, as FanDuel’s own help page states. And Squad Bet and SuperSpins, which appear in Flutter’s investor materials, belong to Sky Bet in the United Kingdom and not to FanDuel, though an unrelated American product of the same name exists at Fanatics.

The Rate Card

Now to the document that makes the rest of this article unnecessary.

A loyalty program has to tell its members how points are earned. Three of the four largest American sportsbooks do so in terms that specify different rates for different kinds of bet, and the differences are not small.

Table 2.2.1Published loyalty earn rates, by type of wager

Program	Graded by	Lowest published rate	Highest published rate	Spread
Caesars Rewards	Type of wager	1 reward credit per $100 on a bet priced shorter than minus 150	50 reward credits per $100 on a same-game parlay at +300 or longer	50 to 1
BetMGM Rewards	Type of wager	At least 20 points per $100 on a straight bet	At least 50 points per $100 on a parlay, “or even more, with longer odds”	2.5 to 1, before the longshot premium
FanDuel Sportsbook Rewards Club	Settled odds	No points at all below minus 1000; 1 point per $1 from minus 1000 to minus 201	6 points per $1 at +4000 and above	6 to 1, and unbounded below minus 1000
DraftKings Rewards	Not published	Not published	Not published	Not stated

Read the Caesars column first, because it is the least ambiguous statement of intent published by any American sportsbook. A customer who wagers $100 on a heavy favorite earns one reward credit. A customer who wagers the same $100 on a same-game parlay earns fifty. The dollar is identical. The bet is worth fifty times more to Caesars in the second case, and Caesars says so in writing on its own website.

BetMGM makes the same statement with a smaller multiple and an additional clause. Straight bets earn at least twenty points per $100 and parlays at least fifty, in each case “or even more, with longer odds” (BetMGM, “Welcome to BetMGM Rewards”). The parenthetical stacks a longshot premium on top of the parlay premium, which is to say the schedule is graded twice in the same direction.

FanDuel’s new sportsbook program reaches the same destination by a different road. Its rate table does not mention parlays. It bands the earn rate by the settled odds of the wager: one point per dollar between minus 1000 and minus 201, two points in the range around even money, four points from +200 to +999, five from +1000 to +3999, and six at +4000 and above. Below minus 1000 a wager earns nothing whatever (FanDuel, “Sportsbook Rewards Club Terms”). Because parlays and same-game parlays mechanically produce long odds, an odds-banded schedule is a parlay incentive expressed in different units. The distinction is worth keeping precise: Caesars and BetMGM reward the parlay as a product, and FanDuel rewards the payout profile that parlays produce. The effect on the bettor is the same.

DraftKings publishes no earn rate at all. Its rewards help articles sit behind a portal that returns nothing to an ordinary reader, and no page on its own domain states points per dollar by product or by bet type. Among the four largest operators it is alone in this, and the absence is itself informative.

Set the schedule beside the published economics. In Casino Reports’ sample, which covers four states (New Jersey, Colorado, Illinois and Maryland), parlays held 15.8 percent in August 2026 against 4.6 percent for other bets, and made up about 63 percent of operator revenue on about 33 percent of handle (Casino Reports, “U.S. Sports Betting Stats”). The casino side of the same programs is graded identically: at both Caesars and BetMGM, slots earn five times what live table games earn per dollar wagered, and slots carry the wider margin. The loyalty schedule is not merely correlated with house margin. Within the precision these programs publish, it is a restatement of it.

This is the argument, and it does not require a single claim about anyone’s dopamine. A bettor who wants to know which products the house profits from most does not need to model the industry, subpoena its data, or trust a critic. He can read the rewards page. The operator has ranked its own products by how much it wants him to buy them, and published the ranking as a customer benefit.

The Illusion of Skill

There is a reason the ranking can be published without embarrassment, and it is the oldest finding in the psychology of gambling. Ellen Langer’s 1975 experiments established that people behave as though chance events are controllable when the situation contains superficial features associated with skill: choice, familiarity, competition, involvement (Langer, “The Illusion of Control”). Sports betting contains all four by construction. The bettor chooses the game, knows the teams, competes against a line, and has spent a decade watching the sport.

A points schedule that pays more for parlays lands inside that bias rather than beside it. The parlay is the bet that feels most like an argument. It requires the bettor to have four opinions instead of one, and to have them cohere. When the schedule pays more for it, the additional payment is legible as a reward for the additional analysis, rather than as a surcharge for the additional margin. A completed mission reads as a challenge met. A tier advanced reads as a career progressing. None of these signals carries any information about whether the bets were good, and all of them arrive whether the bets won or lost.

What follows is the reframing of loss that makes the system durable. A losing parlay in a skill frame is not a bad purchase. It is a near miss, a flawed read, a strategy in need of refinement, and the natural response to a strategy in need of refinement is to refine it and try again. There is a large literature on chasing losses that arrives at this destination through the emotions. The rate card gets there through the accounting.

Part II: What a Free Bet Is Worth

The Psychology of Free, Briefly

The seductiveness of the promotional offer is well established and can be stated quickly. Kahneman and Tversky’s prospect theory supplies the framing machinery: outcomes are evaluated as gains and losses against a reference point, losses hurt roughly twice as much as equivalent gains please, and small probabilities are overweighted (Kahneman and Tversky, “Prospect Theory”). An offer branded as risk-free is engineered against the first of those; a bonus bet best used on a longshot rides the third.

Thaler and Johnson supplied the second half. Prior gains increase risk-seeking, because winnings are mentally filed in a separate account and spent more freely than wages (Thaler and Johnson, “Gambling with the House Money”). What is new is direct evidence of the effect inside a real betting application. A 2026 case-control study of higher-risk and lower-risk Australian bettors found that both groups reported being more likely to make risky bets when using bonus funds, and that higher-risk participants were more likely to make an immediate deposit in order to qualify for a marketing offer (Farrell, et al., “More Frequent Use”). The sample is small and the measure is self-report, and the authors say so.

All of which explains why the offer feels the way it feels. None of it establishes what the offer is worth. That is an arithmetic question, and it has an exact answer.

The Closed Form

A bonus bet, in the standard American construction, returns the profit on a winning wager and does not return the stake. A $100 bonus bet at +100 pays $100 rather than $200. The instrument is therefore not $100. It is a lottery ticket on the profit alone.

Its value follows in one line. Take a two-way market where the true probability of the chosen side is p, and let the book price both sides with a proportional margin, so that the posted implied probabilities sum to k. The posted decimal odds on the chosen side are then 1 divided by the quantity k times p, the profit multiple is that figure minus one, and the expected value of the bonus bet as a fraction of its face value is p multiplied by the profit multiple. The p cancels. What remains is:

the value of a bonus bet, as a fraction of face value, equals 1/k minus p.

In words: a bonus bet is worth the reciprocal of the market’s overround, minus the probability that the bet wins. At the American standard of minus 110 on both sides, k is 1.047619 and the reciprocal is 0.9545. So a bonus bet is worth 95.45 percent of its face value, minus the chance it comes in.

Table 2.2.2What a $100 bonus bet is worth in cash, by the odds it is placed at

True probability the bet wins	Posted odds	Profit if it wins	Cash value of the bonus bet	Value as a share of face
60.0 percent	minus 169	$59.09	$35.45	35.5 percent
50.0 percent	minus 110	$90.91	$45.45	45.5 percent
40.0 percent	+139	$138.64	$55.45	55.5 percent
25.0 percent	+282	$281.82	$70.45	70.5 percent
10.0 percent	+855	$854.55	$85.45	85.5 percent
5.0 percent	+1809	$1,809.09	$90.45	90.5 percent
2.0 percent	+4673	$4,672.73	$93.45	93.5 percent

Three consequences follow, and each of them is a fact about the instrument rather than about the bettor.

First, a bonus bet is worth more the less likely it is to win. This is not a paradox and it is not a loophole. Because the stake is forfeit either way, the bettor is buying only the profit, and the profit is largest where the probability is smallest. The instrument is constructed so that its rational use is a longshot.

Second, the ceiling is 95.45 percent and the floor is zero. A bonus bet placed on a side with a true probability above 95.45 percent is worth nothing at all, because the profit it can generate is smaller than the margin embedded in the price.

Third, and most importantly, the rational use of a bonus bet is the highest-margin product on the board. In Casino Reports’ four-state sample, parlays held 15.8 percent in August 2026, against 4.6 percent for everything else. The longshot the arithmetic favors is a parlay or something priced like one. The sportsbook has handed the customer an instrument whose correct use is a free sample of the most expensive thing in the store, and has then written a loyalty schedule that pays extra for buying it again with real money.

The result is not novel and does not depend on any assumption peculiar to this article. A 2025 paper in The College Mathematics Journal, working the same problem from the other end, observes that because most sports bets are near even-money propositions, the expected value of a free bet “would be about half its face value” (Aboufadel, “A Mathematician Wanders Into a Sportsbook”). The closed form above says the same thing and says where the other half went.

The experimental literature confirms that bettors do in fact respond this way. A 2019 online experiment with 299 Australian wagering customers found that participants selected significantly longer-odds bets when any incentive was offered, compared with a no-inducement control, and the authors noted that if long-odds bets are priced to give bettors poorer value than short-odds bets, the effect would raise gamblers’ losses and operators’ profits in equal measure (Rockloff, et al., “Sports Betting Incentives”). The bettors are not making an error. They are solving the problem they were handed.

What a Playthrough Costs

The second family of offers, the deposit match, works by a different mechanism and is more expensive still. Here the customer is given bonus funds that convert to withdrawable cash only after wagering some multiple of the bonus, the deposit, or both. The multiple is called playthrough or rollover, and its cost is computable.

A wager placed into a market with hold h costs the bettor h of the amount staked, in expectation. Clearing a playthrough requirement of X dollars of handle therefore costs X times h, whatever the bonus is nominally worth. A five-times playthrough on a $200 bonus requires $1,000 of wagering. At the best hold available to a disciplined bettor on two-way markets, 4.55 percent, that costs $45.50. At the roughly 10.1 percent hold the American market actually realized in 2025 and 2026, it costs $101. At the 15.8 percent parlays held in August 2026, in Casino Reports’ four-state sample, it costs $158, which is most of the bonus.

Now apply the same arithmetic to a real offer. The $1,000 deposit bonus at the center of a Massachusetts class action against DraftKings required, according to the complaint, a $5,000 deposit and $25,000 in qualifying wagers within ninety days (Public Health Advocacy Institute, “Plaintiffs Prevail Over DraftKings”). The court denied most of DraftKings’ motion for summary judgment in February 2026, so the case is live and the terms are on the record.

Twenty-five thousand dollars of handle breaks even against a $1,000 bonus only if the bettor’s hold comes in below 4.00 percent. The standard price on a two-way market is minus 110 on both sides, which carries a hold of 4.55 percent of handle, and a bettor with no demonstrable edge cannot expect to do better than that. Playing perfectly, he spends $1,137.50 in expected losses to obtain $1,000 in credits that cannot be withdrawn and are themselves worth roughly half their face value by the arithmetic above. At the hold an average customer actually realizes, the cost is $2,525.

There is one exit from that conclusion and it closes immediately. A bettor who genuinely beats the closing line, and who therefore holds a negative effective hold against the book, could clear the requirement at a profit. He is also, as the previous article documented from Massachusetts regulatory data, precisely the customer sportsbooks identify and limit. The offer is negative for every bettor the operator is willing to keep.

Table 2.2.3What the offer says and what the terms do

Offer	The marketing promise	What the terms actually do	The term to find
First bet on the house, no-sweat bet, bet reset	Get your money back if it loses	The refund arrives as bonus bets, not cash, and the stake on a bonus bet is not returned when it wins	Refund type, stake not returned, expiry
Deposit match	We will match your first deposit	The match is bonus funds locked behind a wagering requirement on the deposit, the bonus, or both	Playthrough multiple, what it applies to, time limit
Bonus bet or bet credit	Bet $5, get $200 in bonus bets	A winning bonus bet pays the profit only, so its cash value is 95.45 percent of face minus the chance it wins	Stake not returned, expiry, market restrictions

What the Research Actually Found, and Where It Was Done

The peer-reviewed evidence on promotional inducements is stronger than the industry acknowledges and narrower than critics usually imply, and both halves matter.

The strongest causal work is recent. A preregistered, incentive-compatible experiment with 622 participants betting on real Euro 2024 matches found that inducements increased the amount spent by more than ten percent, almost halved the number of people who chose not to bet at all, and made bettors three times more likely to choose objectively dominated options, with the effects largest among participants showing evidence of gambling problems (Ó Ceallaigh, et al., “Effects of Inducements”). A randomized field experiment published in 2026 went the other direction, asking a treatment group to opt out of direct marketing from all their operators and verify that they had done so. Those who opted out placed 23 percent fewer bets, spent 39 percent less money, and reported 67 percent fewer short-term gambling harms (Rockloff, et al., “Direct Gambling Marketing”). Account-level tracking data from the French regulator, covering 9,306 online bettors, found that in weeks when gamblers used inducements, gambling intensity, frequency, and at-risk behaviors all rose (Balem, et al., “A Longitudinal Study”).

On comprehension the finding has now been replicated across seven years and two countries with no apparent improvement. In 2019, nearly three in five Australian sports bettors underestimated the additional amount they would have to wager to access winnings from a bonus bet (Hing, et al., “Where’s the Bonus in Bonus Bets?”). In 2026, a randomized study of 585 British bettors found the control group’s median estimate of a wagering requirement was £500 where the correct answer was £750, and that showing a worked example cut the offer’s perceived attractiveness sharply, but worked least well for participants with the highest gambling severity scores (Torrance, et al., “(Mis)Comprehension”). The cheap disclosure fix works, and it works worst on the people it most needs to reach. Work by the Behavioural Insights Team found three in five participants did not realize a wagering requirement applied at all, and seven in ten could not calculate how much they would need to stake to satisfy one (Behavioural Insights Team, “Impact of Free Bets and Promotions”). That research is a behavioural unit’s, not a peer-reviewed journal’s, and belongs in that category, but it points the same way as the two studies above.

Now the caveat, which is large. Not one of these studies was conducted in the United States. The experimental work is Irish and Australian, the tracking data is French, the comprehension work is Australian and British. The best figures on how many bettors take up an offer come from the British regulator, which found that nine in ten online gamblers surveyed had received an offer in the preceding four weeks and 76 percent of those who received one used it (Gambling Commission, “Exploring Consumer Journeys”). Those are British numbers about a British market with different products and different rules, and they should not be laundered into American claims. No American uptake figure, and no estimate of the share of American sportsbook revenue attributable to inducement-driven play, exists in the published record. The operators have the data.

Part III: The Word Games and the Rules

A Ban That Is Not a Ban

In March 2023 the American Gaming Association updated its Responsible Marketing Code for Sports Wagering, and the update was widely reported as banning the term risk-free. The actual provision is a single sentence inside a responsible-gaming section: “No message should suggest engaging in sports wagering is without risk or utilize ‘risk free’ language” (American Gaming Association, “Responsible Marketing Code”). The Code binds members, who are “required” to comply, and addresses everyone else, who are “expected” to. Enforcement runs through a review board whose sole available sanction is requiring the offending material to be withdrawn or revised. There is no fine, no licensing consequence, and no mechanism at all against a company that is not a member. No decision of that board appears to have been published.

What followed was semantic rather than structural. Risk-free became no-sweat, and no-sweat has now become, at FanDuel, Bet Reset. The refund still arrives as restricted credits rather than cash. The name on the tin changed three times; the product did not change once. This is the pattern the series keeps encountering, and it is why the book is called what it is: the reflection adjusts to the new rule, and the thing behind it does not move.

What the States Actually Did, and Did Not

Six states have prohibited the language by rule, and their drafting is worth reading because it is better than the trade association’s. Ohio requires that promotions described as free or risk-free “must not require the patron to incur any loss or risk their own money to use or withdraw winnings from the free wager” (Ohio Admin. Code 3775-16-09). Massachusetts bars operators from implying that wagering is “free of risk in general or in connection with a particular promotion,” and adds a font-parity requirement: the amount a patron must wager of their own funds has to be disclosed “in the same size and style of font” as the advertised credit (205 CMR 256.04). New York, Colorado, Maryland and Missouri have comparable provisions, Missouri’s arriving with the state’s new market in 2025.

Three states commonly named in this connection have no such rule. Arizona’s responsible advertising regulation does not address bonuses, promotions, or the word free at all. Virginia’s advertising rules address the accuracy of winnings claims and nothing else. No advertising or promotions rule appears in any chapter of Tennessee’s sports wagering regulations. Pennsylvania’s is not a rule but a request: in March 2023 a regulator’s spokesman said the board had asked operators to stop using the term, and no directive, notice, or rulemaking accompanied it.

Then there is the more important omission, which is total. No American jurisdiction caps playthrough multiples. No American jurisdiction caps bonus expiry periods. No American jurisdiction regulates the stake-not-returned structure of the bonus bet in any way. What the states have done is require disclosure of terms they have declined to limit. Ohio’s rule expressly authorizes playthrough and then protects only the customer’s own money. Maryland requires operators to file their wagering requirements with the commission, which is a filing duty rather than a limit. The single quantitative constraint anywhere in American law is a Virginia provision capping at ninety days the period over which a new-account offer may be implemented (11VAC5-80-140). It caps the clock. It says nothing about the multiple.

Compare the jurisdictions that did legislate. British rules that came into force on 19 January 2026 cap wagering requirements at ten times the bonus and prohibit incentives that mix more than one gambling product, so that an offer cannot require football betting to unlock casino spins (Gambling Commission, “Socially Responsible Incentives”). Sweden goes considerably further and permits a licensee to offer a bonus only on the first occasion a customer gambles with it, which is to say one bonus per customer, ever. Ontario removes bonus advertising from public channels entirely: inducements may be communicated only on an operator’s own site and to players who have actively consented, and the regulator fined BetMGM Canada $110,000 in March 2025 for engaging marketers who offered cash for new account signups (Alcohol and Gaming Commission of Ontario, “Registrar’s Standards”).

The Courts Are Doing the Work Instead

In the absence of substantive rules, the pressure in the United States is coming from the private bar. Three cases matter. The Massachusetts action over the $1,000 deposit bonus survived summary judgment in February 2026 on the question of whether consumers had reasonable notice of the terms. A federal action filed in the Eastern District of Pennsylvania in July 2025 attacks the stake-not-returned structure directly, arguing that a refund delivered as a bonus bet is not a refund. The court ruled on motions to dismiss and to consolidate in April 2026, allowing the case to proceed in part (Macek v. DraftKings, Memorandum Opinion).

The third is the one to watch. A suit filed in the Court of Common Pleas of Philadelphia on 24 March 2026 against DraftKings, FanDuel, Genius Sports and the National Football League also names, as defendants, two VIP hosts who work for FanDuel and three who work for DraftKings (Gambacorta, “Lawsuit Accuses FanDuel, DraftKings, and the NFL”). Naming the hosts personally is novel. Reporting on the complaint adds that the sportsbooks paid each host on the basis of the two plaintiffs’ losses; that allegation is drawn from the pleading as described in the coverage rather than from a document this article has read, and it should be weighed accordingly (Gambling Harm Alliance, “Online Sportsbook VIP Hosts Sued”). If it is proved, it will describe a structure in which the employee assigned to a customer’s account was paid more the worse that customer did.

A note against overstatement is due here, because the VIP story is less settled than it is usually told. Loyalty program use does correlate strongly with gambling problems: an Australian synthesis found at least 40 percent of problem gamblers reported using loyalty cards against roughly ten percent of gamblers generally (Delfabbro, et al., “The Prevalence of Loyalty Program Use”). But a study combining self-report with archival player account data found that tier status drove loyalty most strongly among gamblers without disorder, and did not influence attitudes or behavior among those high in disordered gambling symptomatology (Hollingshead, et al., “On Being Loyal to a Casino”). The simple story, that VIP schemes capture the addicted, is not what that evidence shows. The tiers work best on people who are fine.

Federal legislation would do more than any of it. The SAFE Bet Act would prohibit advertising that induces gambling with bonuses, no-sweat offers, bonus bets or odds boosts. It was introduced in March 2025, has attracted no Senate cosponsors, and has seen no committee action in the seventeen months since. It is a marker of direction, not a pending law.

The Ledger the States Keep

The scale of all this is not a matter of inference, because several states tax it and therefore count it.

Michigan is the clearest window, because its statute defines adjusted gross receipts as gross receipts less free play and nothing else, so the free-play total can be recovered exactly by subtraction from the regulator’s own monthly releases. In 2023, free play consumed 46.8 percent of gross internet sports betting revenue in Michigan. In 2024 it consumed 57.8 percent. In 2025, 35.1 percent. In December 2024, a single month, gross internet sports betting receipts of $20.2 million produced adjusted gross receipts of $267,881, which is to say free play absorbed 98.7 percent of the taxable base (Michigan Gaming Control Board, “December 2024 iGaming Revenue”). The state taxed a $20 million month as though it had been a quarter-million-dollar one.

Maryland, which publishes the figure directly and has been steadily tightening it, ran the other way. Mobile promotional play fell from $211,430,976 in fiscal 2023 to $113,339,295 in fiscal 2026, and as a share of mobile handle from 7.93 percent to 1.67 percent, while handle itself more than doubled (Maryland Lottery and Gaming, “FY2026 Summary”). Colorado has legislated its free-bet deduction down a ladder to zero, which took effect on 1 July 2026. Arizona’s expires in each operator’s sixth year. Ohio, uniquely, is moving the other way and will allow a ten percent deduction from 2027.

Ohio is also the control case, and it settles a question the industry would prefer left open. Promotional credits have never been deductible in Ohio, so if promotions were primarily a tax-arbitrage instrument, Ohio would be where they thinned out. They did not. Promotional credits ran to approximately $302 million over the first eleven months of 2024 and approximately $306 million over the same months of 2025. Those totals are this article’s own arithmetic: the Commission prints promotional credits in its per-proprietor tables and publishes no statewide monthly total, so the figures are sums of the published monthly detail and should be read as computed rather than reported (Ohio Casino Control Commission, “2025 Ohio Sports Gaming Revenue Report”). A 2026 study across thirty jurisdictions found that limiting the deduction actually increased total wagers, gross gaming revenue and state tax collections, and that hold percentages fell, suggesting operators respond by cutting margins rather than by cutting promotions (Badger, “The Impact of Free Play Tax Deduction Limitations”).

One more figure disposes of a story that is otherwise tempting to tell. It would be neat to report that the bonus era is ending as the market consolidates and offers get worse. It is not true. Flutter’s promotional spend as a share of American handle rose from 4.0 percent in the fourth quarter of 2024 to 5.8 percent in the fourth quarter of 2025 and stood at 5.4 percent in the second quarter of 2026 (Flutter Entertainment, “Q2 2026 Earnings Release”). DraftKings reported second-quarter 2026 revenue down five percent year over year, attributing the decline in part to “increased promotional reinvestment associated with new customer acquisition,” with sales and marketing up 38 percent (DraftKings, “Q2 2026 Earnings Release”). Promotional intensity did decline through the maturing markets of 2021 to 2023, and DraftKings said so at the time. Then it reversed. The spending is going up.

The Mirror Test

Bonus Bets and Loyalty Tiers

VerdictIllusion

The Mirror Test is this series’ standing verdict: every strategy, product and claim is judged on whether it creates a real edge or only the appearance of one. The verdicts are EDGE, ILLUSION and CONDITIONAL.

The claim under test is the one every promotional offer invites the customer to hold: that the bonus is free money, and that taking it leaves him ahead. It fails, and unusually for this series it fails on arithmetic rather than on judgment. The instrument has a computable price, that price is never zero, and it is highest exactly where the customer is least likely to collect. Nothing in the verdict depends on the bettor being careless, addicted or badly informed. It depends only on the terms, and the terms are published.

The case against

The case for a softer verdict is real and should be stated. A bonus bet is not worth nothing. Its cash value is substantial, between 45 and 90 percent of face depending on where it is placed, and a customer who takes the offer, uses it on a longshot, and closes the app has extracted genuine value from the house. A disciplined bettor who reads the terms, ignores the deposit matches, and treats bonus bets as what they are can come out ahead on the promotional economy specifically. The offers are not a confidence trick. They are a real transfer with a computable size.

Why the verdict holds

What fails the test is the belief that the transfer is the point. It is the sample, and the sample is of the most expensive product on the shelf. The loyalty schedule then pays the customer to keep buying that product with his own money, at rates the operators publish, graded by their own margin. The generosity is real; it is simply much smaller than it looks, and it is pointed somewhere.

ILLUSION Bonus Bets and Loyalty Tiers

Conclusion: The Price List Is Public

The phrase for all this is older than the industry and older than the economics. In nineteenth-century American saloons the free lunch was a real institution: ham, pickles, oysters, bread and cheese laid out on a counter at no charge to anyone who bought a drink. The food was genuinely free, it was reliably salty, and the arrangement was profitable enough that it survived for decades and was eventually legislated against in several states (Word Histories, “Origin of ‘There Is No Such Thing as a Free Lunch’”). In 1975 Milton Friedman took the saloon-keeper’s arrangement and made it a maxim, collecting his Newsweek columns under the title There’s No Such Thing as a Free Lunch (Friedman, There’s No Such Thing as a Free Lunch). His point was never that the lunch was a lie. The lunch was real. His point was that its cost had been moved into the price of something else, where the customer would not think to look for it.

That is the structure of the modern promotional offer, transposed one industry over. The bonus bet is real. The tier is real. The Crown Cash and the reward credits and the profit boosts are all real, and their cash value can be computed to the cent. The cost sits in the margin of the product they route the customer toward, and unlike the saloon-keeper’s salt, that margin is published every month by state regulators. In August 2026, in the four-state sample Casino Reports tracks, parlays held 15.8 percent (Casino Reports, “U.S. Sports Betting Stats”).

An interview study of long-term online gamblers and treatment-seekers put the industry’s direction of travel in one sentence, and it is the sentence this article has been circling. Its authors concluded that “[i]nducements and the poor pricing of complex bets such as multi-bets, and their outsized attraction to players with problems, should be a key focus” (Hing, et al., “How Structural Changes”). The inducement and the complex bet are named together because they arrive together.

It is a mistake to treat the loyalty program and the promotional bonus as separate systems. They are one instrument with two ends. The bonus is the acquisition device and the tier is the retention device, and the tier delivers further bonuses as its rewards, so that the customer engages with the gamified structure in order to earn the promotional credits whose correct use routes him into the highest-margin products, on which the gamified structure then pays him at the highest rate. There is no point in the loop at which the customer’s interests and the operator’s point the same direction, and there is no point at which anyone has to be deceived for it to work.

That is the finding this article would leave a reader with, and it is deliberately unglamorous. The usual account of engineered engagement reaches for the language of manipulation, of hijacked reward circuitry, of dark patterns and compulsion loops. Some of that language is warranted and some of it, as the previous article argued at length, is a good deal shakier than its confidence suggests. But the reader does not need to adjudicate any of it. Caesars will tell him, on a public page, that it values his same-game parlay fifty times more than his short-odds bet. BetMGM will tell him it values his parlay two and a half times more than his straight bet, and more still as the odds lengthen. FanDuel will tell him that a wager at +4000 earns six times what a wager at minus 300 earns, and that a heavy favorite earns nothing at all. Michigan will tell him that free play swallowed 98.7 percent of a month’s taxable revenue. And the arithmetic of the bonus bet, which no operator publishes and none disputes, will tell him that the instrument in his account is worth 95.45 percent of face minus the chance it wins, and is therefore worth most when he is least likely to collect it.

The House of Mirrors works by reflecting the bettor’s own reasoning back at him at a flattering angle. The engagement architecture is the most polished mirror in the building, because it converts the act of losing money into evidence of progress, and the act of taking the house’s most expensive product into evidence of shrewdness. But mirrors have an edge, and this one is thinner than most. The house has published its preferences. They are sorted by margin, they are available without logging in, and they are, read correctly, the most honest document the industry produces.

Sources

This is 2.2 of House of Mirrors: A Bettor’s Guide to the Illusions of Modern Sports Gambling, a serialized investigation of the American sports betting industry. Next up is article 2.3, The In-Play Revolution.