What an AI trading bot actually does — and what it can't

The three jobs inside every trading bot

Strip away the branding and every automated trading system does three separable things. Evaluating a bot means evaluating each one on its own:

• Analysis — reading market data (price, volatility, structure) and forming a view: buy, sell, or stay out. This is where 'AI' actually applies: a model weighs more context than a fixed indicator rule can.
• Risk decision — deciding whether that view is allowed to become a trade: how much is at stake, where the stop-loss goes, whether current conditions permit trading at all. This should be rules, not AI — you want it deterministic and inspectable.
• Execution — placing, managing, and closing the order on your broker account. This is plumbing, and the questions here are reliability and transparency, not intelligence.

What the AI part genuinely adds

A language- or reasoning-model can hold more context in one decision than a classic indicator strategy: recent price action, volatility regime, session timing, and the relationship between them. Where a moving-average crossover fires blindly, a model can conclude 'this setup is technically valid but the conditions are poor — hold.'

That last output matters more than people expect. The single biggest difference between a disciplined system and an account-burner is how often it declines to trade. 'Hold' and 'no trade' should be common outputs, not error states.

What no AI can do

No model — however large — predicts the market reliably. Markets are moved by information that does not exist yet: news, order flow, other participants' reactions. Any product that claims its AI 'knows' where price is going is describing something that cannot exist.

What a well-built system can honestly claim is narrower: it applies a consistent process, it never revenge-trades, it never moves a stop out of fear, it sizes every position the same disciplined way, and it stops when its rules say stop. Those behaviors — not prediction — are where automation earns its keep, because they are exactly the behaviors humans fail at under pressure.

The claims that should end the conversation

Some marketing lines are not exaggeration — they are disqualifying. If you see any of these, close the tab:

• A promised return ('2% daily', 'passive income guaranteed'). Returns depend on markets nobody controls; a promise is a fabrication.
• A published win rate with no audited track record behind it. Win rate alone is meaningless anyway — a 90% win rate loses money if the tenth trade gives it all back.
• 'No losses' or 'risk-free'. Every real trading system loses regularly. The honest question is whether losses are capped and expected, not whether they exist.
• Pressure to fund a live account immediately, skip the demo, or deposit with a specific unregulated broker.

The questions worth asking instead

Flip the evaluation: instead of asking 'how much will this make?', ask how it behaves when things go wrong — because they will.

• What happens when the AI provider is down or returns garbage? The only safe answer is 'no trade'. A fallback signal is a hidden second strategy you never agreed to.
• Can I run it on a demo account first, with the identical logic that would run live?
• Can I see every decision it makes and why — before or after execution?
• What are the hard limits — per-trade risk, daily loss cap, drawdown stop — and can I set them?
• Is going live an explicit, deliberate step, or the default?