How players choose an online casino: Behavioural evidence from 56,696 comparison sessions
Almost every gambling regulation written in the last decade rests on the same assumption: a consumer, given disclosed bonus terms, will compare operators and choose between them. Well, we’re in a position to test that assumption directly. This study is what we found.
In short
Regulators assume players compare operators based on bonus terms. But when measured on-site, the median decision takes 26 seconds, and 84.2% of players click through to an online casino without looking at any additional information.
What this study covers
56,696 browsing sessions where a visitor saw at least one online casino presented for comparison. From those sessions, we observed that 27,608 went on to click through to an operator. From there, we observed what brands entered view, how long each held a visitor’s attention, what information the visitor opened, how many operators they tried, and when.
The existing systematic review of objective online gambling behavioural data covers 56 peer-reviewed studies, with each one drawing on operator data recorded after a player has already registered
[9]
. None of those studies observed the moment the choice itself was made. For example, a 2025 commentary in Addiction names a lack of access to behavioural data as the single most critical barrier to research on gambling platform design
[10]
. This study rectifies this by covering the step before registration, which is the step nobody has been able to see.
Every figure is a median, a distribution or a rate. We don’t publish traffic volumes, website-level breakdowns, named operators, or information about where brands sit in our listings. We set out the methodology and limitations in full at the end, including three findings that didn’t hold up.
Decision speed
26s
median time from arriving to choosing an operator
Bounded at both ends, so no assumption about when a session ends. Half of all decisions are made faster than this.
Informed choice
84.2%
of players click through to a casino having opened nothing about it
No review, no expanded detail, no list re-sorting. 87.5% never open a review.
Consideration
7
casino brands were seen in the median session
Around 4.8 seconds of screen time each. This is more than shoppers give most consumer categories.
Comparison
69.8%
of those who click through try exactly one operator
27.8% try more than one, but almost always by going back and clicking again, not by comparing first.
Finding 1: The decision takes 26 seconds
The cleanest measure we have is the time between a visitor arriving on one of our sites and clicking through to a casino operator. It is bounded at both ends (a page load and a deliberate act), so there’s no assumption about when a session ends. The median time is 26 seconds.
Session duration tells the same story from a different angle. Sessions where a visitor saw at least one casino ran to a median of 36 seconds, while those that ended in a click ran to a median of 50 seconds. We checked the obvious objection (that short sessions are just people leaving, and the real deliberation is hidden in sessions we cannot see to the end), but it runs the other way. Sessions ending in a click have a clear endpoint, yet they are actually longer than sessions that simply stop. The 26-second figure isn’t hiding a longer decision-making process. If anything, it makes the decision time look shorter than it actually is.
- Sessions that clicked through (n=27,608)
- Sessions that saw a casino brand (n=56,696)
Percentage of sessions still active at each elapsed time
Show data table Hide data table
| Elapsed time | Sessions that clicked through | Sessions that saw a casino brand |
|---|---|---|
| 5s | 97.2% | 90.1% |
| 10s | 90.7% | 79.8% |
| 30s | 70.7% | 55.1% |
| 60s | 53.5% | 38.1% |
| 2min | 37.5% | 24.7% |
| 5min | 20.2% | 12.7% |
| 10min | 10.8% | 6.8% |
Source: Kasinohai on-site event data, 2026-05-01 to 2026-07-31.
For context rather than contrast, research on web behaviour generally finds most page visits are abandoned within 10 to 20 seconds, with the average visit running just under a minute [7]. Choosing where to gamble, in other words, takes about as long as deciding whether to keep reading a web page. It’s not a fast decision by internet standards. Instead, it’s an ordinary one, which is a problem because regulation assumes it’s not.
Finding 2: Seven brands is a lot, but more of them makes things worse
In a median session, visitors see 7 casino brands, with a quarter of sessions seeing three or fewer and one in eight seeing more than 25. The instinct is to read the number seven as small. However, compared with other consumer categories, seven is definitely not a small number.
- Online casinos (this study)
- Published measures, other categories
Number of brands actively considered Scale runs to 10.
Show data table Hide data table
| Category | Brands in the consideration set |
|---|---|
| Online casinos (this study) | 7 |
| Coffee | 4 |
| Deodorant | 3.9 |
| Frozen dinners | 3.3 |
| LED televisions | 3.19 |
| Financial services | 1.4 |
Source: This study, with comparison categories from published consideration-set measurement.
Measured consideration sets in ordinary consumer categories show that visitors see about four brands for coffee, 3.9 for deodorant, 3.19 for televisions and 1.4 in financial services [3]. Against that, gambling consumers look unusually thorough. However, the next result suggests they aren’t being rewarded for comparing more options.
Pooled across all sessions, click-through rises with the number of brands seen, from 41.4% at one to three brands to 62.8% at 26 or more. At first glance, that looks like a straightforward case for casino affiliates to show people more options. It is a Simpson’s paradox: sessions that see more brands are simply longer, and longer sessions convert better. When restricting the comparison to sessions of similar length, however, the relationship inverts.
- Clicked through
- Did not
Show data table Hide data table
| Brands seen | Pooled click-through | Duration-controlled | Controlled sample |
|---|---|---|---|
| 1–3 | 41.4% | 61.9% | 2,767 |
| 4–6 | 50.7% | 70% | 4,450 |
| 7–12 | 47.8% | 59% | 5,879 |
| 13–25 | 48.6% | 51% | 7,394 |
| 26+ | 62.8% | 52.7% | 3,606 |
Source: Duration control is restricted to sessions lasting 30–300 seconds.
Held constant for time, the click-through rate peaks at 70% among sessions seeing four to six brands and falls away steadily after that. This is a live commercial replication of one of the most contested results in consumer psychology: a larger assortment attracts more attention but produces fewer decisions [1]. That result has had a difficult history (the definitive meta-analysis puts the average choice-overload effect at close to zero across 50 experiments [2]). Therefore, we report ours as one observational data point in a contested literature, not as a settled matter. But the shape is definitely there, in the field and at scale.
Figure 4 shows the distribution of consideration-set sizes and what each additional brand costs in attention.
- Smaller sets
- Best-converting band
- Larger sets
Share of sessions seeing each number of casino brands Scale runs to 50%.
Show data table Hide data table
| Brands seen | Share of sessions |
|---|---|
| 1–3 brands | 26.5% |
| 4–6 brands | 21.4% |
| 7–12 brands | 20% |
| 13–25 brands | 19.9% |
| 26+ brands | 12.2% |
Source: Sessions that saw at least one casino brand (n=56,696).
Finding 3: Almost no one opens something before deciding
This finding matters for policy, and it is the one we would most like to be wrong about.
Of the sessions that clicked through to an operator, 84.2% did so without a single information-seeking act beforehand. By this, we mean that users didn’t open a review, expand brand details or re-sort/filter the list. In fact, 77% performed no such act at any point in the session, before or after. Expanding brand details in place is effectively extinct, observed in only 0.45% of sessions.
We should be careful how far we push that, because the comparison literature cuts both ways. Only one or two retail software shoppers per thousand ever open a licence agreement [4], and three-quarters of people skip a privacy policy outright when given the option [5]. Against those benchmarks, the 12.5% of gambling consumers who open a review behave with unusual diligence, at roughly 60 times the rate at which people read software terms. The problem is not that gambling consumers are careless by the standards of online consumers generally. The standard is extraordinarily low, and gambling regulation is written as though it were high.
The minority who do read additional details behave completely differently. Review-readers click through to an operator at 63.7% against 45.8% for non-review-readers, and their sessions run to a median of 147 seconds against 27 for the other group. Reading reviews isn’t a proxy for engagement. Instead, the conversion advantage holds even after controlling for session depth.
But even review readers barely read in depth. 33.8% of sessions containing both a review and a click opened the review after already clicking through to the operator; i.e., they conducted research retrospectively to confirm a decision rather than make one in the first place. Among those who did read a review first, the median gap between opening the review and clicking out was 45 seconds.
Finding 4: Many people do compare, but only afterwards
The honest counterweight to Finding 3 is that a substantial minority genuinely do shop around. 27.8% of sessions that click through usually do so for more than one distinct operator.
A note on that number, because we got it wrong ourselves at first. Counting raw clicks gives 34.7%, but that includes people clicking the same operator twice. Deduplicated to distinct brands, the number drops to 27.8%. We also found that accidental double-taps (defined as the same brand being re-clicked within two seconds) account for only 0.34% of clicks, so this is a genuine correction rather than mere noise removal.
Comparers do not do what the word implies. Up to the first click, they are indistinguishable from everyone else: median deliberation of 22 seconds versus 25 for single-operator sessions, which is marginally faster, not slower. The entire divergence happens afterwards. They click out, return a median of 43 seconds later, and try another option. Interestingly, 67.2% of the time, the second operator they tried was not even on screen when they made the first choice.
That seems more like trial and error, rather than deliberation. Comparison in this market is serial and retrospective: consumers sample operators one at a time and reject them one by one, rather than weighing disclosed terms against one another before choosing. It is a real form of consumer discipline, but it is not the form that disclosure rules are designed to enable.
Finding 5: The late-night gambler is not in the data
We expected to find that gambling research clusters late at night, and that late-night sessions would look more impulsive. Neither is true, so we report the null because the assumption is widespread enough to warrant correction.
In Finland (our largest market, with a single time zone and no daylight-saving transition within the window, so local time can be assigned exactly), sessions between 23:00 and 05:00 account for 16.9%. Those six hours are a quarter of the clock, so late-night activity is running at an index of 67 against an even distribution. Therefore, it is under-represented, not over-represented.
Session starts by local hour and weekday, indexed so the weekly average = 100
Show data table Hide data table
| Day | Peak hour index | Late-night average (23:00–04:59) |
|---|---|---|
| Mon | 177 | 60 |
| Tue | 177 | 63 |
| Wed | 251 | 73 |
| Thu | 173 | 74 |
| Fri | 196 | 74 |
| Sat | 152 | 72 |
| Sun | 140 | 56 |
Source: Finland only. Markets cannot be pooled (see methodology below).
The behavioural comparison is a flat null. Sessions starting late at night convert at 54.5% against 52.9% in the day, open reviews at 21.6% against 21.8%, run to a median of 42 seconds against 41, and see the same median number of brands. On every measure available to us, the small hours look exactly like the afternoon for all intents and purposes.
The weekly shape also contradicts the leisure assumption. Wednesday is the busiest day of the week, and Sunday is the quietest, with the week’s hottest cell falling on Wednesday 20:00, at an index of 251. Choosing a casino, therefore, seems like a weekday-evening activity, closer in rhythm to online shopping than to going out at the weekend.
Finding 6: The screen changes what you see, not what you do
Desktop sessions appear to consider far more operators than mobile ones, giving a median of 16 against 6. That gap is mostly an artefact of rendering: wide screens lay brand cards out side by side, so many more of them cross the viewport without anyone choosing to look at them. In vertical list layouts, where both devices show one card at a time, the gap narrows to roughly 1.6x.
Underneath the rendering, the decision is the same. Time from arrival to choice is 26 seconds on mobile and 25 seconds on desktop. Information-seeking is set by market rather than device. And the device that shows fewer options converts better. Mobile users click through at 50.2% versus desktop’s 38.7%, consistent with Finding 2 and with the possibility that seeing more is not the same as knowing more.
There’s one detail worth stating because it constrains everything above: on every device, the median number of casino brands visible without scrolling is 0. No one chooses from the first screen. Every decision in this dataset required at least some form of deliberate scrolling.
What this means for disclosure
Taken together, the findings point to a consistent consumer. They arrive with intent, scan roughly seven operators for a few seconds each, choose one in under half a minute without opening anything about it, and (in a substantial minority of cases) come back later and try another. They are not irrational. Instead, they are running a fast, cheap sampling strategy in a market where trying an operator costs almost nothing and reading about one costs time.
That behaviour is a poor fit for the regulatory model built around it. Mandatory disclosure of bonus terms and wagering requirements assumes consumers read the information at the point of choice. However, on this evidence, roughly only one in eight consumers read anything about the terms and conditions, and a third of those read it after choosing. Research on comprehension suggests that even those who do read frequently misunderstand what they have read. For example, around three in five sports bettors underestimate the amount they must stake to release a bonus [6].
None of this should surprise regulators, because their own evidence says the same thing. The UK Gambling Commission’s participation research found that only around a fifth of gamblers had ever read an operator’s terms and conditions [11]. Behavioural Insights Team work found that three in five people did not realise a wagering requirement applied at all, and seven in 10 could not calculate one [12]. Our figure (around one consumer in eight opening any information before choosing) sits alongside those, measured behaviourally rather than by survey, and at the moment of decision rather than in recall.
The legal standard those rules are written against is the “average consumer” of European consumer law: reasonably well-informed, reasonably observant and circumspect [13]. That is a normative construct rather than an empirical claim, and it’s not our place to say the courts have got it wrong. But it’s worth stating plainly that the consumer in our data is observant for about 26 seconds and well-informed only if the information reached them inside that window.
The more constructive reading is that regulators have already begun acting on this, whether or not they framed it this way. The UK capped wagering requirements at 10 times from January 2026; Denmark caps at the same level and additionally requires a worked cash example of what must actually be staked. Both are structural rules that bind regardless of whether anyone reads anything, and both implicitly concede that disclosure alone was not enough. Finland’s 2027 regime will go further still, capping wagering requirements on all bonuses at five times.
Here’s the practical conclusion we would draw: If consumers don’t come to the information, either the information has to move to where the decision happens (into the few seconds of attention a brand gets while it is on screen, not behind a click) or the product has to be constrained so that the decision matters less. Disclosure and structural limits are not equivalent instruments, and, on this evidence, the industry’s consistent preference for the former deserves more scepticism than it usually receives.
We also need to note what this study doesn’t show. It cannot tell you whether these consumers chose poorly. Fast decisions are not automatically bad ones, and a market where switching is cheap may not require carefully considered first choices. What this study shows is narrower and firmer: the deliberation that disclosure regimes are built to inform is not happening when those regimes assume it is.
Methodology
The dataset is our own on-site event stream, covering 2026-05-01 to 2026-07-31. The analysis population is the 56,696 sessions in which at least one casino brand card entered the viewport, of which 27,608 produced an outbound click to an operator. Recorded events include brand card entry and exit from the viewport, outbound clicks, review opens, in-place detail expansion, list re-sorting, page loads, and scroll depth.
Population discipline. 39.3% of all sessions contain exactly one event, which drags the median duration across all sessions to two seconds. That figure is an artefact, and we don’t use it. Every duration in this study states its population, and the headline decision measure (time from arrival to first outbound click) is bounded at both ends and therefore immune to the problem.
Confounds we tested. The relationship between consideration-set size and click-through is a Simpson’s paradox in its raw form, and we report it after controlling for duration. The apparent device gap in consideration is largely a rendering artefact, and we report it with a layout control. We present time-of-day analysis for a single market because pooling markets across different time zones produces a meaningless curve.
What we don’t publish. This is our own site data, and the study is a consumer-behaviour statistic rather than a disclosure about our commercial business. Every figure is a median, distribution or rate. We don’t publish traffic volumes, breakdowns by website or named operators. Plus, we deliberately don’t analyse where operators appear in our listings or how placement relates to clicks. That last exclusion is a commercial consideration, and, for full transparency, readers of this study should know it.
Three findings we could not stand behind
- Regulated versus grey markets. Information-seeking looks dramatically higher in regulated markets than grey ones, which would have been a striking result about whether licensing regimes change consumer behaviour. However, it doesn’t survive scrutiny: the split is confounded with which of our properties serves each market, too few properties carry volume in both regimes to separate the two and the spread within regimes rivals the gap between them. Therefore, we report it as not established.
- One market’s clock. A second large market shows a nearly flat 24-hour curve, which would have been an interesting contrast with Finland. A neighbouring country in the identical time zone shows a sharp curve, which points to geolocation contamination rather than behaviour, so we excluded it.
- Cross-session journeys. We hoped to measure how many visits precede a decision. The persistent identifier is populated on around 4% of events, so we can’t answer the question with this data, and we make no claim about it either.
Limitations
- This is a population with intent. These are people who arrived at a comparison site, mostly via search and are already looking for an operator. They are not the general public, and they are plausibly at the most deliberate end of the market, which makes the shallowness of the decision a conservative finding, not an exaggerated one.
- We are one route to market. Players also arrive at operators through advertising, other affiliates, app stores and word of mouth. Nothing here describes those journeys, and there is no reason to assume they involve more deliberation.
- The absence of an event is not the absence of behaviour. A visitor may read the bonus terms printed on the brand card itself without triggering an event we can record. Our information-seeking measure captures deliberate acts (such as opening a review, expanding detail or list re-sorting) and will understate passive reading.
- Dwell time is quantised. Per-card dwell is derived from cards entering and leaving the viewport, which the instrumentation samples rather than measures continuously. We report it directionally and don’t draw conclusions from it alone.
- Scroll is partially instrumented. Scroll depth is recorded on roughly half of sessions, so we have kept it out of the published figures and use it only as supporting context.
- Choice overload is contested. Finding 2 is consistent with a classic experimental result, but its meta-analytic support is weak. We present it as one large observational replication rather than a confirmation.
Reuse
Journalists and researchers are welcome to reproduce any figure or chart from this study with attribution and a link to this page. Given that the existing evidence base on gambling behaviour is drawn almost entirely from post-registration operator data, we would particularly welcome academic replication of these measures against other pre-registration datasets. For the underlying aggregate tables, or for questions about method, contact the Kasinohai editorial team.
References
- [1] Iyengar & Lepper, Journal of Personality and Social Psychology. When choice is demotivating: can one desire too much of a good thing? (accessed 2026-07-31).
- [2] Journal of Consumer Research. Can there ever be too many options? A meta-analytic review of choice overload (accessed 2026-07-31).
- [3] WARC / Ehrenberg-Bass Institute. A new measure of consideration set size: the average number of salient brands (accessed 2026-07-31).
- [4] Bakos, Marotta-Wurgler & Trossen, The Journal of Legal Studies. Does anyone read the fine print? Consumer attention to standard-form contracts (accessed 2026-07-31).
- [5] Obar & Oeldorf-Hirsch, Information, Communication & Society. The biggest lie on the internet: ignoring the privacy policies and terms of service policies of social networking services (accessed 2026-07-31).
- [6] Journal of Gambling Studies. Bonus bets and the misunderstanding of wagering requirements among sports bettors (accessed 2026-07-31).
- [7] Nielsen Norman Group. How long do users stay on web pages? (accessed 2026-07-31).
- [8] Nielsen Norman Group. Scrolling and attention (accessed 2026-07-31).
- [9] Journal of Gambling Studies. A systematic review of research employing objective online gambling behavioural data (accessed 2026-07-31).
- [10] Addiction (Society for the Study of Addiction). Access to behavioural data as the critical barrier to online gambling research (accessed 2026-07-31).
- [11] Gambling Commission (Great Britain). Gambling participation and attitudes survey — consumer awareness of operator terms and conditions (accessed 2026-07-31).
- [12] Behavioural Insights Team. Should wagering requirements on gambling bonus offers be capped? (accessed 2026-07-31).
- [13] Official Journal of the European Union. Directive 2005/29/EC on unfair commercial practices — the “average consumer” standard, recital 18 (accessed 2026-07-31).
- [14] Gambling Commission (Great Britain). Exploring consumer journeys: playing online casino games (accessed 2026-07-31).