AI in trading: what it can really do (and what it can't)

Why talk about it without hype

Since ChatGPT, Claude and other assistants became mainstream, social media is full of promises: bots that "never lose", prompts that "find the perfect trade", automated strategies making 10% a month. Reality is simpler and, in my view, more interesting.

I have been trading for ten years. Over the last year I stopped publishing and spent the time testing AI seriously, on my own process, with my own mistakes. This article sums up what I saw working and what did not.

What AI can really do in trading

1. Turn a vague idea into a written plan

Most bad trades start from an idea that was never written down. An AI assistant is great at asking the right questions: why you enter, where you are wrong, how much you risk, when you exit. It does not decide for you, but it forces you to be clear.

2. Play devil's advocate

Asking AI to tear your idea apart is one of the most useful and least used tricks. Ask it for three reasons the trade could go wrong: it often finds at least one you had missed.

3. Calculate risk without rushed mistakes

Position size, risk in money, risk to reward: simple maths that goes wrong under pressure. AI does it in seconds, as long as you give it the right numbers and check the result.

4. Write and check code

Indicators, alerts, Pine Script for TradingView, small backtests: this is where AI saves hours. Always read and test the code, because it can contain errors you will not spot at first glance.

5. Summarise long documents

Earnings releases, central bank minutes, research reports: AI summarises them well. Use the summary to know where to look, not as a replacement for the original source.

6. Analyse your trading journal

If you keep a journal with entries, exits and reasons, AI can help you find patterns you cannot see alone: the hours when you make more mistakes, the setups that work, the emotions that come back before losses. This is the use that changed my workflow the most.

What AI can NOT do (and where it fails)

It does not predict the future

A language model has no crystal ball. Ask it "where is the price going tomorrow?" and it will still answer, confidently. A confident tone is not a forecast.

It can make up numbers

This is the biggest risk. An assistant can write a plausible but false price, date or financial figure. Every number you use to decide must be checked at the source: your broker, the company's official website, real market data.

It often lacks current data

If the assistant is not connected to live data, it works with old information. Always ask which data it is using and up to which date.

It can build backtests that look too good

With AI it is very easy to tune a strategy until the past looks perfect. This is overfitting: everything works on the past, much less on the future. A credible backtest has simple rules, includes costs and keeps a test period separate from the one used to build the strategy.

It does not know you

It does not know how much you can lose and still sleep well, your time horizon, or what you already hold. You have to provide this, and the responsibility stays with you.

5 rules to use AI in trading safely

1. Use it as a copilot, never as an oracle.
2. Check every number at the source before using it.
3. Decide your maximum risk before opening the chat, not after.
4. Write everything down: plan, reason, outcome. Without data, AI has nothing to analyse.
5. Be wary of anyone promising returns, with or without AI.

This article is for educational and informational purposes only and is not financial advice. Trading involves a high risk of losing capital.