Is AI taking over trading?
The honest answer is more useful than the scary one.

You have probably seen the number: “AI now drives 89% of global trading volume.” It gets repeated constantly, usually by someone selling a bot.

Here is a different number, and this one was actually measured. The Bank for International Settlements surveys the FX market every three years. In April 2025 it found that 59% of FX trading was executed electronically — virtually unchanged since 2022. Voice trading is still, in their word, vital.

Electronic isn't even the same thing as AI. So the honest answer to “is AI taking over trading?” is: it is taking over parts of it, at speed, and almost none of those parts are the one you are worried about.

What the institutions actually run

Adoption is real and it is high. Industry surveys in 2026 put roughly 78% of financial institutions using AI somewhere in trading decisions, and about 70% of hedge funds running machine-learning models somewhere in their pipeline.

Then you read the next line of the same research, and it changes shape:

Around 18% of hedge funds rely on AI for more than half of their signal generation. Everyone else is using it for something else entirely — execution, slippage reduction, position sizing against volatility regimes, risk monitoring, compliance, research triage.

That is the pattern underneath every honest account of this: professional money uses AI to execute better, not to decide what to buy. The prediction problem is the hardest one in the building and the one they trust it with least.

What the retail tools actually do

The retail end looks nothing like that, and the regulators have been unusually direct about it.

• The CFTC issued a customer advisory whose blunt summary is that AI will not turn trading bots into money machines, and which recommends human oversight of automated systems.
• The SEC, FINRA and NASAA put out a joint investor alert on AI and investment fraud. The SEC has since made “AI washing” — overstating what your AI does — an enforcement priority.
• The FCA maintains a warning list, and unauthorised AI trading bots appear on it by name.

On performance, the estimates that exist suggest 70–80% of retail trading bots fail to produce sustainable returns — overfitting, thin risk management, strategies that cannot bend when the regime changes. Treat that as an estimate, because almost nothing in this corner of the market is independently verified. That is rather the point: the gap between what these tools claim and what anyone has checked is as wide as anywhere in retail finance.

So where is AI genuinely changing things for retail traders?

Not in the entry. In everything around it.

The most interesting movement in 2026 is on the prop-firm side, where firms have started running AI over trader behaviour rather than over price. Firms doing it report double-digit improvements in evaluation pass rates and meaningfully fewer accidental daily-loss breaches. Those are the firms' own numbers and nobody has audited them — but the direction is telling, and it is not “the machine picks the trades.”

It is: the machine watched you for six weeks and noticed you lose money on Thursdays after a losing Wednesday.

That is a genuinely different product from a signal bot, and it is the one with evidence behind it. A model reading a market it has never seen is guessing. A model reading your own trading history is doing something closer to what it is actually good at — finding a pattern in a dataset, where the dataset is you.

The uncomfortable bit

That only works if the dataset exists.

Most retail traders have no usable record of their own trading. A broker statement tells you what filled — not what you were thinking, not which rule you broke, not the four setups you saw and talked yourself out of. Ask any AI to review that and it will produce something that reads well and means nothing, because there is nothing underneath it.

So the real answer to “is AI coming for retail trading” is a bit deflating: it cannot get to you yet, because you haven't written anything down. The traders who will get the most out of AI over the next few years are the ones who spent this year keeping a proper record.

What to actually do about it

• Stop evaluating AI on whether it predicts. Judge it on whether it tells you something true about your own trading that you did not already know.
• Be suspicious of any win rate in marketing. Regulators are prosecuting exactly this, which tells you how common it is.
• Keep the record first. Entries, exits, the rule you were following, the setups you skipped. Everything useful downstream depends on it.
• Use AI where it has an edge: pattern-finding across hundreds of your own trades, which is tedious, and which humans are measurably bad at.

Exhibit A is built on that premise. It is a trading journal with an AI coach that reads your own history and writes you a performance review — including the trades you didn't take, which nearly every other journal ignores. It does not predict anything. It tells you what you already did, in a form you can act on.

Common questions

Will AI replace retail traders?

Not on current evidence. Institutions overwhelmingly use it for execution and risk rather than direction, and the minority who let it generate most of their signals are a small slice even of hedge funds. The pressure on retail traders comes from costs and discipline, as it always did.

Are AI trading bots worth buying?

The available estimates say most fail, regulators on both sides of the Atlantic have issued specific warnings, and independent verification is close to non-existent. If you buy one, buy it knowing you are the one carrying the risk.

What is AI actually good at in trading?

Reading large amounts of your own data and finding patterns in it — behavioural leaks, timing, sizing, the difference between your rules and your behaviour. That requires a record. Most traders don't have one.