AI Trading Bots: What They Really Do

An AI trading bot is usually one of two things: a fixed-rule program sold with an "AI" label, or a program that uses machine learning to find patterns in past market data and turn them into trades. Neither can reliably predict prices. Machine learning can find real patterns, but it is also very good at finding false ones in noisy market data, so its results need even more careful checking than a simple rule-based bot.

AlgolaneBots's five strategies are fixed-rule bots, not AI. This post explains what the AI label usually covers, how rule-based and machine-learning approaches differ, and how to judge an AI claim.

What does "AI" usually mean in trading bot marketing?

"AI" has no agreed meaning in trading-bot marketing. In practice the label covers a wide range:

• Plain rules, relabelled. "If the price crosses above its 50-day average, buy." This is ordinary programming. Calling it AI is marketing.
• Optimised rules. A computer tries thousands of combinations of settings on past data and keeps the best. This is a search, not intelligence, and it is a direct route to overfitting.
• Machine-learning models. A model is trained on past prices and other data to forecast something, such as whether tomorrow's price will be higher. The model's decision logic comes from the data rather than from a person.
• Language models reading text. Some tools scan news, company filings or social media, score the mood, and feed that score into trading rules.
• Chatbots giving picks. A general-purpose chatbot asked which stocks to buy produces fluent text, not a tested strategy.

When a product says it uses AI, the first question is which of these it means. A provider that cannot or will not say is telling you something.

Rule-based vs machine learning: what is the difference?

Both kinds of bot follow instructions. The difference is where the instructions come from.

A rule-based bot follows rules a person wrote, such as: "Each month, if the market is healthy, hold technology leaders; otherwise hold short-term bonds and gold." You can read the rules and predict what the bot will do.

A machine-learning bot follows rules a model learned from data. A person chooses the inputs, the type of model and the training period, but the final decision logic may involve thousands of numbers that nobody can read as plain rules.

Neither is automatically better. Some professional firms use machine learning heavily, backed by large teams, specialist data and big computing budgets. For an individual, the practical question is whether you can understand, test and stop what you are running. Rule-based vs discretionary trading covers the human side of this choice, and what a trading bot is explains the basic loop every bot runs.

Why is overfitting a bigger risk with AI trading bots?

Overfitting means a model has learned the noise in past data rather than a real pattern. It looks excellent on the data it was built on and disappoints on new data. Market data makes this especially likely:

• It is noisy. Day-to-day price moves are mostly random. The real signal, if there is one, is small.
• There is not much of it. Decades of daily prices sounds like a lot, but it contains only a handful of crashes and long bull markets. A model cannot learn reliably from a few examples of the events that matter most.
• It changes. Markets shift as interest rates, regulation and participants change. A pattern from one decade may not exist in the next.
• Models are flexible. The more settings a model has, the more easily it can fit any history, including the parts that were pure luck.

An illustrative example: test 1,000 random rules on the same ten years of prices and some of them will look impressive by chance alone. Pick the best one and you have a great backtest with no edge behind it. Machine learning automates this kind of search on a much larger scale, so the risk grows with it. Our post on overfitting explains the warning signs.

Language models add a newer problem. A model trained on text up to a certain date has, in effect, read about what happened in markets before that date. Testing it on those same years can give flattering results, because it may already "know" the outcomes. That is a form of hindsight bias; see hindsight and survivorship bias.

Fixed rules can overfit too

Rule-based bots are not immune. Each time a person adjusts a rule after looking at past results, the backtest becomes a little less trustworthy. AlgolaneBots's own Leveraged Growth strategy is an example we publish openly. It reads a set of market signals after each close, holds leveraged growth funds such as TQQQ and SOXL when conditions look calm, and moves to bonds, volatility funds or inverse funds when they do not. Leveraged funds aim for a multiple of one day's move and reset daily, so over longer periods, especially choppy ones, they can drift far from that multiple and lose value quickly.

In a historical backtest (a computer test on past prices) from January 2011 to November 2026, it returned +129.1% a year with a worst fall of -53%. Over the same dates the S&P 500 returned +13.9% a year with a worst fall of -34%, and the Nasdaq 100 +18.2% a year with -35%. Its deepest fall, in early 2020, recovered within 3 months of the bottom. Nobody earned these figures, and future results will differ.

Its caveat: its rules may be fitted to the years after 2010, and leveraged funds can lose value quickly. A simulated three-fund version run through the 2000-2002 dot-com crash fell 91% against 47% for the S&P 500, and took 115 months to regain its pre-crash value against 78 months for the index. A number like +129.1% a year should make anyone more cautious, not less. The same goes for any AI claim.

What can AI genuinely help with in trading?

Machine learning has real, well-established uses in finance: spotting fraud, estimating trading costs, executing large orders efficiently, cleaning data and summarising long documents. Large firms rely on it for these jobs.

Forecasting prices well enough to trade profitably after costs is one of the hardest uses, because the signal is weak and competition is intense. For an individual, AI is more likely to add value in the tools around trading, such as summarising research or explaining how a strategy works, than in a black box that claims to predict prices.

How should you judge an AI trading bot?

Ask the provider these questions, and expect clear answers:

1. What exactly does the AI do? Which of the categories above does it fall into?
2. Can you explain why it made a particular trade? "The model decided" is not an explanation.
3. How was it tested? Was the test period kept completely separate from the training period?
4. What has it done since the model was frozen? Forward results on paper or real money matter more than any backtest.
5. How often is it retrained, and is each new version tested before it trades? A model that changes constantly has no stable track record.
6. What are the benchmark, worst fall and recovery time? Over the same dates, side by side. See how to read a backtest.
7. What costs were assumed, and what does the provider admit could go wrong?
8. Can you watch it on paper for free?

Treat "AI that predicts the market", promised returns, secret algorithms and missing drawdowns as red flags. Trading bot red flags covers more.

Why does AlgolaneBots use fixed rules instead of AI?

AlgolaneBots's five strategies follow written rules, the same for every customer, that can each be described in a few sentences. Gold Breakout waits for gold to go quiet, then trades a breakout in the direction of the long-term trend. Big Tech Trend decides each month whether the market is healthy. Steady Growth reweights seven fixed names each week. There is no model learning from your account or quietly changing its behaviour between tests.

That is not a claim that fixed rules beat AI. It is a choice to keep every strategy explainable and checkable. You can read what each one does, see its historical backtest next to a benchmark, and read its main weakness in the strategies section. The safety section explains the kill switch and order checks that apply whatever the rules decide.

Related reading: can ChatGPT be your trading bot?.

Key takeaways

• "AI trading bot" can mean anything from relabelled simple rules to machine-learning models; ask which.
• No bot, AI or not, can reliably predict prices.
• Machine learning is especially prone to overfitting because market data is noisy, limited and always changing.
• Fixed rules can overfit too; look for published caveats, benchmarks and worst falls.
• AlgolaneBots's strategies are fixed-rule, not AI, so you can read what each one does.

This is general education, not personal financial advice.

Watch fixed rules work on paper

Instead of trusting a label, watch what a bot actually does. Start a free 90-day AlgolaneBots trial on a $10,000 paper account, with no card needed, and see every decision the five strategies make on real prices.