A hand-built frequency spreadsheet and three separate AI lottery tools were given the exact same job for twelve weeks, and only one of those approaches ever tried to convince me it was smarter than it actually was.
Quick disclosure before anything else: a few of the links below are affiliate links, and if you use one, I earn a small commission at no extra cost to you. Everything mentioned here has actually been run through my own testing as part of an ongoing lottery strategy habit, logged in the same paper notebook every week. I'm a maths teacher, not a financial adviser, and nothing here should be read as advice to spend beyond what you're comfortable losing.
Every EuroMillions draw from that stretch sits logged in a physical notebook I keep at school, mostly because doing statistics for fun is not something I want my Year 11s finding out about. Before getting into the results of this maths teacher's experiment, it's worth answering the question I get asked most: does any of this actually work, or is it just spreadsheets dressed up as insight?
None of this started from a place of confidence, either. A few months before the twelve-week test, I'd been taking tips from a YouTube channel that recommended picks based on numerology, life-path numbers calculated from birthdays and cross-referenced against draw dates, delivered with the kind of certainty that makes you forget the presenter has no maths background whatsoever. It was good entertainment. Once I actually tracked those picks against real draws, they performed exactly as well as chance, which is to say no better than closing your eyes and jabbing at a play slip. That's what pushed me toward a proper controlled comparison instead of taking anyone's word for it, including my own.
What This AI Lottery Tools Experiment Actually Compared
Right then, here's the actual design. Every week for twelve weeks, I generated four sets of numbers for the same EuroMillions draws: a random quick pick as a control, a line from my own manual frequency spreadsheet, and a pick each from three separate AI tools. Every result got tracked against a negative-expected-value baseline rather than any fantasy of winning big — the maths behind why that baseline is negative is laid out properly in calculating the real odds of a EuroMillions Lucky Star, and for this piece the short version will do: however the numbers get chosen, every ticket starts from the same lopsided odds. The three AI lottery tools were LottoChamp, Lottery Defeated, and Lotto Master Key, and each earned very different marks by the end of the test.
Whether Twelve Weeks Is Long Enough to Mean Anything
Here is the thing though: twenty-four draws is a small sample by any measure. I tell my Year 11s something similar when one good mock exam result convinces them a revision method 'worked' — one strong outcome from a small sample tells you almost nothing about what happens over the next fifty attempts. The same logic applies to a EuroMillions line matching two numbers three times in a row. It feels meaningful. Mathematically, it barely moves the needle. None of this is the Gambler's Fallacy — the belief that a number becomes 'due' after a dry spell, which is a different mistake, and one that gets unpacked properly elsewhere alongside a few other patterns that look meaningful and mathematically aren't. Whether the AI tools were quietly leaning on old-fashioned hot-and-cold number logic under the hood is impossible to say for certain from outside the software, and nothing in the results over the twelve weeks pointed clearly either way. What twenty-four draws actually can tell you is whether a tool is systematically doing something different from chance, even before you know whether that something is useful.
Where My Own Frequency Chart Led Me Astray
My own spreadsheet is the one that embarrasses me most. Partway through building it up from historical draws, three numbers kept showing up together far more often than chance should allow, and I started weighting my own picks toward them, convinced I'd found a genuine cluster. The mechanics of building that kind of chart are covered properly in building their own EuroMillions frequency chart; what happened once I actually leaned on mine is the less flattering follow-up. Over the draws that followed, the cluster dissolved back into the same flat distribution every number eventually settles into. I hadn't found a pattern. I'd fitted my own model to noise and then admired my own cleverness for spotting it, which is a very particular kind of foolish. Chalk that up as one thing in this whole experiment that flatly did not work: trusting a short-term cluster in my own hand-built data.
Weighing the Three AI Tools After Twelve Weeks
Lottery Defeated turned out to be the pleasant surprise of the three. Its number-frequency dashboard does something my Excel sheet does badly: it shows how often specific pairs of numbers land together across the whole draw history, not just whether one number looks 'due'. During one stretch of testing it produced a genuinely balanced ticket, a mix of even and odd numbers, highs and lows, and matched enough small prizes to more than cover the cost of tickets for a while afterwards. The dashboard has a steeper learning curve than the other two, and it costs more, but the frequency tool on its own made it worth persisting with.
Lotto Master Key sits at the opposite end of that spectrum. It skips the charts and the historical depth entirely and just produces a set of numbers quickly, based on its own internal weighting. For someone who wants something better than a random quick pick without wanting to learn a dashboard, that simplicity is the whole appeal, and it's noticeably cheaper than the other two as well. Its historical database is thinner than LottoChamp's and it updates less often, which shows if you're the sort of person who checks assumptions against fresh data regularly.
Not every AI tool I tried during this stretch made it into that main three. One separate app I downloaded on a whim advertised something close to a guaranteed improvement in your odds of matching more numbers, strong language, suspiciously vague on method. I ran it alongside the others for several draws out of curiosity, and its hit rate sat right in line with the random control group, nowhere near what the marketing implied. That's the one claim from this whole experiment I'd flag as actively misleading rather than just optimistic: a tool that can't beat expected value has no business being advertised as though it can.
When I Got My Own Numbers Wrong
There's a specific quiet to an empty school corridor on a Tuesday evening, once the last after-school club has finished and even the photocopier down the hall has gone still. That's usually when the notebook comes out properly, well away from Year 11 and away from Fenella's running commentary. And yes, in exactly that quiet, I still managed to get my own numbers wrong once, which is a bit mortifying for someone who marks other people's arithmetic for a living. Partway through the test I misread my own shorthand. A slash meant 'matched two main numbers'; a tick meant 'plus a Lucky Star'. One week I copied a tick as a slash while transferring results off a till receipt. That left an entire week's tally sitting wrong until the totals stopped adding up and forced a recheck. I remember the walk back from the newsagent near Levenshulme Market around that same stretch. Running a finger down the LottoChamp slip, I realised both Lucky Stars matched the ticket in my pocket — nowhere near a big prize, but enough to make me stop on the pavement and check it twice before it went in the notebook properly. Small mix-ups like the slash-versus-tick error are exactly why I don't trust anyone's lottery 'data', including my own, without a way to double-check it.
When a Syndicate Can't Agree on Numbers
A small syndicate of four of us splits a line most Fridays, and we hit exactly this problem a while back. Two of the group wanted to keep playing birthday-and-anniversary combinations because that's what they'd always played, while I wanted to switch the syndicate's line over to whatever the AI tools were generating that week. Fenella, a neighbour of mine who has an ongoing bet with me over whether any of this pattern-hunting means anything at all, suggested we just alternate: birthdays one week, tool-generated numbers the next, and let the results settle the argument over time. It didn't settle anything. That's exactly what basic probability would predict from a sample that small. One of the birthday weeks matched an extra number purely by chance, though, so for a while afterwards I had to listen to it being treated as proof of concept. We still split the line every week. We've just stopped pretending either method has been vindicated.
The Reader Question About Different Lottery Formats
One reader, Dariusz Wojcik, emailed after reading the frequency chart piece to ask whether the same approach holds up if you play both EuroMillions and Poland's Multi Lotek, since the two games draw from pools of different sizes. He asks these things with a dry, matter-of-fact directness that cuts through wishful thinking rather quickly. The honest answer is that the underlying maths scales to any pool size, but the specific numbers absolutely do not transfer: a EuroMillions frequency chart tells you nothing useful about a Multi Lotek draw, because they're separate systems generating separate distributions. Whether what these AI tools call 'pattern detection' is anything more than maths with better branding is a genuinely open question, and one I'd rather name plainly than pretend to settle in a single article.
Whether Any of This Is Actually Worth Paying For
If you want a longer, more sceptical look at whether the budget option holds up under real scrutiny, I did exactly that to see if it passes the math teacher test, worth reading before committing to any of the three. As for the overall verdict after twelve weeks: none of these tools turned twenty-four draws into a fortune, and nobody selling one honestly claims they can. What they did do, roughly, is push the small-prize hit rate above what the random control and my own spreadsheet managed on their own, not dramatically, and not consistently enough to call it proof, but consistently enough that I kept using one after the test officially wrapped up. Of the three, LottoChamp is the one I'd point a fellow sceptic toward first: the interface hasn't aged well, but its filtering logic is transparent about what it's doing and why, and a money-back window means you're not stuck with it if you end up disagreeing with me.
Playing on regardless is what most of us end up doing anyway, myself very much included, so you might as well strip out the obviously bad picks before spending on a line at all. LottoChamp is where I'd start, not because it promises anything the maths doesn't allow, but because it's the version of 'better than guessing' I trust most after tracking all three side by side over the same twelve weeks. And if any of this ever stops feeling like a bit of fun and starts feeling like something you can't stop, please talk to a professional or a gambling support service rather than a maths teacher's blog.