One rainy Friday evening in Manchester, I sat at my kitchen table with my school-issued red pen and a notebook full of scribbled numbers, realizing my manual frequency charts were missing the bigger picture. Outside, the drizzle was doing that particularly grey thing it does in the North, and inside, I was trying to figure out why my 'hot numbers' were as cold as my forgotten cup of tea. It occurred to me that while I’m perfectly capable of teaching trigonometry to a room of thirty teenagers, my little tally marks in a spiral-bound book were no match for the actual complexity of the EuroMillions draw.
As a math teacher, I know that every draw is an independent event. The balls don't have memories. They don't know they were picked last Tuesday. However, the way humans interact with those numbers—and the way we try to predict them—is where things get interesting. I’d spent months on a side project analyzing draw patterns from public lottery databases, but I hit a wall. That was when I started looking into how AI tools were doing something similar at scale. I wasn’t looking for a magic wand; I was looking for a more sophisticated weather forecast for a storm that happens twice a week in Paris.
The Geometry of the Lucky Star Pool
Before we dive into neural networks, we need to look at the board. The EuroMillions consists of 50 main numbers and a separate pool of 12 Lucky Stars. To win the jackpot, you need to beat the statistical probability of 1 in 139,838,160. That is a number so large it’s difficult to visualize, but I tell my students it’s like trying to pick one specific blade of grass out of a field that covers most of Greater Manchester.
The Lucky Stars are often where the game is won or lost. Since the pool was increased from 11 to 12 in September 2016, the mathematics changed slightly. You only have to match the two Lucky Stars to get a small return—the odds of that are 1 in 115—but for the big prize, they are the final hurdle. Most people approach these by looking at what came up last. They see that '3' and '7' haven't appeared in a while, so they think they are 'due.' In the staffroom, I often hear colleagues talking about their 'lucky' numbers, and I have to bite my tongue. Probability doesn't care about your birthday or your cat's age.
What I realized through my notebook tracking—which I still keep tucked in my desk drawer at school—is that manual frequency analysis is a bit like trying to paint a portrait with a 4-inch house-painting brush. It’s too blunt. AI, specifically machine learning algorithms, uses a much finer brush. It doesn’t just look at frequency; it looks at the relationship between the numbers.
Why AI Sees What We Can't: The Delta Shift
In mid-March, I shifted my focus from simple tallies to 'delta' analysis. In lottery terms, a delta is the numerical distance between the numbers in a draw. For example, if the Lucky Stars are 2 and 9, the delta is 7. While the numbers themselves are random, the distribution of these deltas often follows a bell curve over a long enough timeline.
This is where advanced AI algorithms come in. Feeding years of historical draw data into a neural network allows the system to identify clusters in the Lucky Star subset that human eyes simply can't catch. When I started running parallel picks—comparing my manual notebook predictions against an AI tool—the difference was striking. The AI wasn't 'predicting' the future, but it was identifying the statistical 'shape' of the draws.
Right then, here is the thing though: most people use these tools to find 'hot' numbers. They want to know what is most likely to hit. But as I’ve documented in Why I Started Grading AI Lottery Predictors Against My Own Frequency Charts, the real value of AI isn't in finding the most likely numbers—it's in avoiding the most common ones.
The Frequency Trap: A Contrarian Warning
Here is my unique angle on this, and it’s one that usually makes people pause. Most lottery advice tells you to look for patterns and follow the frequency. If you use AI to find the most 'likely' numbers based on historical frequency, you are doing exactly what thousands of other players are doing.
From a mathematical standpoint, focusing on historical frequency actually increases your risk of sharing the jackpot. If you and fifty other people use the same 'advanced' analysis to pick the same 'likely' Lucky Stars, and those numbers actually hit, your share of the prize plummeted. You’ve beaten the 1 in 139 million odds only to find out you’re splitting the pot with a small village.
When I’m sitting at my desk, feeling the slight indentation on my thumb from gripping a mechanical pencil during late-night tallying, I think about this constantly. I’d much rather use AI to find 'mathematically plausible but unpopular' combinations. It’s the difference between following the crowd to a crowded beach and finding a quiet cove that’s just as nice but empty. AI allows us to filter for combinations that fit the statistical profile of a winning draw (the 'delta' and 'spread') without falling into the trap of overrepresented numbers.
Six Months of Data: Notebook vs. Neural Network
I’ve been running this experiment from early spring through late summer 2026. Every Tuesday and Friday, I record the results. After about twelve weeks of tracking, a pattern emerged—not in the numbers themselves, but in the performance of the tools. The AI was consistently better at suggesting 'balanced' tickets—those that didn't clump all numbers in one decade or result in unlikely delta sequences.
In late June, I had a moment of clarity. I was smelling the dry, dusty scent of a chalkboard eraser while staying late to prep a lesson on conditional probability. I started thinking about how I'd explain the difference between independent events and conditional probability to my Year 11s if they ever caught me doing this. I’d tell them that while the draw is independent, our strategy doesn't have to be. We can use data to stay within the boundaries of what is statistically normal.
For instance, it is very rare for the two Lucky Stars to be consecutive numbers (like 1 and 2). It happens, but not often. AI algorithms can instantly filter out thousands of these 'low-probability' combinations, leaving you with a pool of numbers that are more 'representative' of a typical draw. You can read more about my specific findings in My Six-Month Experiment with AI Lottery Tools: A Math Teacher's Notebook Findings.
How to Start Tracking with AI
If you want to move beyond the kitchen-table notebook approach, here is how I suggest framing your use of AI for Lucky Stars:
- Focus on the Subset: Don't try to analyze the whole 50+12 field at once. Use the AI to specifically look at the 12 Lucky Stars. The smaller the pool, the more 'visible' the delta patterns become.
- Look for 'Delta' Trends: Instead of asking the AI for the next numbers, ask for the most common distances between the stars. Is the gap usually 3? Is it usually 5? This gives you a framework for your own picks.
- Avoid the 'Hot' List: Use the AI to identify what everyone else is likely picking, then pivot. If the AI says 3 and 9 are the most 'frequent' stars this month, maybe look at the numbers that fit the same delta profile but aren't 3 and 9.
- Track the 'Weather': Treat AI results like a weather forecast. It tells you the conditions are right for certain types of numbers, but it doesn't guarantee a storm.
I am, of course, a maths teacher and not a professional gambler or a financial advisor. The lottery is a game of chance with a negative expected value—meaning, on average, you will lose money. Please treat it as a hobby, not an investment strategy. If you find yourself spending more than you can afford, it’s time to put the notebook away and talk to a professional.
Closing the Notebook
By late summer, my notebook was looking a bit battered. The edges were curled, and there were coffee rings on the cover. But the data inside was clear: AI is a tool for logic, not a shortcut to wealth. It has changed how I look at those two little stars every Tuesday and Friday. I no longer look for 'luck'; I look for statistical normalcy.
Every Friday evening after school, before I head home, I lock that notebook in my desk drawer. There’s something satisfying about knowing that while the odds remain astronomical, I’m approaching the problem with the same rigor I expect from my students. Whether the AI finds the right numbers or not, the process of analyzing the patterns is far more rewarding than just blindly clicking 'Lucky Dip' and hoping for the best. After all, math isn't just about finding the answer—it's about understanding how you got there.