What is an AI trading bot—and what does it actually automate?

The four layers of an automated trade

A system collects market data, computes a signal, applies risk rules and—if permitted—sends an order to a broker. Data quality and order execution remain critical even when the signal model is sophisticated. A forecasting tool with no order connection is an analytical assistant, not a hands-free trading system.

MetaTrader’s official documentation describes Expert Advisors that perform analysis and trading operations based on algorithms. A program can therefore be fully automated without machine learning.

What an actual AI component would do

A machine-learning model may estimate probability distributions, classify market states or rank candidate trades. Its usefulness must be tested on data it has not seen during development. The word “AI” alone tells us nothing about model quality, execution cost or financial performance.

Due-diligence questions

Ask for the exact supported instruments, who maintains the model, how signals become orders, the full cost stack and how you can disable trading. Request live, time-stamped evidence that includes losses and open exposure rather than a screenshot of past gains.

• Is the system analysis-only, signal-only or allowed to execute orders?
• Which broker account and permissions are required?
• Are risk controls enforced independently of model decisions?
• Does live reporting include trading costs, drawdown and failed orders?

A practical buyer rule

If the seller cannot distinguish the forecasting process from execution and custody, the product is not ready for a favorable review. Fraud warnings about guaranteed AI returns are particularly relevant in an industry where backtests can be selectively presented.