Why Is Every AI Tipster Losing Money?

All 9 entrants in our league are losing money. That includes the five chatbots, our own prediction model, and a baseline that does nothing but back the favourite in every race.

That last entrant is the giveaway. The favourite baseline has no intelligence in it whatsoever. It reads nothing, considers nothing and applies no judgement. It is losing at a rate comparable to the models that read the entire form book.

When the sophisticated approaches and the mindless one all end up in the same place, the thing they have in common is not their method.

They are all paying the same charge

Every one of these entrants places bets into the same market, and in the biggest fields that market charges nearly three times what it charges in the smallest.

To finish level, a tipster does not need to pick more winners than average. It needs to pick winners at prices that are wrong by more than the charge. Those are very different tasks. The first is achievable with care. The second requires being systematically better than the collective opinion of everybody else betting, which is what sets the price in the first place.

Our accuracy scoring shows how hard that is. Across 10,478 races our model reads 0.098 on the standard forecasting score against the market's 0.090, where lower is better. The market is the better forecaster. A model that is slightly worse at forecasting than the market, and pays the charge on top, has no route to profit.

The chatbots have a further problem: they are largely reproducing what the market already thinks. They read the same form, and their selections cluster around the front of the market. Gemini backs the favourite 12% of the time. A tipster that agrees with the market gets the market's accuracy and pays the market's charge.

What would actually change the result

Two things, neither of which is a better model.

The first is a cheaper market. The same selections in small-field races rather than big handicaps would face a charge around a third of the size. That is not a hypothetical: it is the largest single number we have measured.

The second is a better price for the same selection. If a tipster's pick is available at a bigger price somewhere than the one it eventually starts at, the difference goes to the bettor rather than the book. That is the one area where we have seen a real signal in our own testing, set out in how some people make a living at this.

What would not change the result is a cleverer chatbot. We have five of them, from five different laboratories, and the spread between best and worst is a few percentage points inside a much larger loss. The variable that dominates is not which model you ask.

A word on all of this

None of these pages is a tip, and none describes a way to win. They describe what betting costs, which is a different and more reliable subject. If your betting has stopped being fun, BeGambleAware has free, confidential help, and the National Gambling Helpline is on 0808 8020 133.

Common questions

Why do AI tipsters lose money?

They pay the same charge as everyone else, and they are not more accurate than the market. Being a good judge is not enough when the price already reflects a good judgement.

Is the AI worse than a human tipster?

We have not tested human tipsters. We have tested a baseline that backs the favourite mindlessly, and the AI models are in the same range as it.

Does more data fix it?

Our own model reads far more data than a chatbot does and still returns −14.4% over 10,478 races.

Would a better prompt help?

We changed the prompt for the informed arm partway through and the standings did not transform. The gap to break even is much larger than prompt wording moves.

So is AI useless for racing?

Useless for beating the market, on this evidence. Genuinely useful for measuring it, which is what this site actually does with it.