Does ICT Trading Actually Work? Evidence and a Real AI Trade

ICT is not a myth and not a guarantee. Here is what can be tested, why many traders lose with it, and what a measurable edge looks like in a real, journaled AI trade.

LSG Club · 16 October 2026 · 11 min read

Does ICT trading actually work? The honest answer is that the concepts are only a vocabulary for reading price, and an edge exists only when you turn them into fixed rules and measure the results over many trades. Nobody can promise it will work for you, but you can test it cheaply, which is exactly what a journal such as the LSG Club trading journal is built for. This article shows what can be tested, why many traders fail, and what one real, journaled trade looks like.

Does ICT trading actually work? What can and cannot be tested

ICT (Inner Circle Trader) is the body of ideas taught by Michael J Huddleston, and Smart Money Concepts (SMC) is the version of it that spread across the community. Asking whether it "works" is like asking whether chess openings work. An opening is not a result. A result comes from a player who follows a plan, move after move, and is measured over many games.

What you can test

Anything that can be written as a rule that two different people would apply the same way. For example:

Only trade inside the London Kill Zone (02:00-05:00 New York time).
Only take longs after sell-side liquidity is swept and price breaks the last swing with displacement that leaves a Fair Value Gap.
Only buy in discount, below the 50% of the dealing range.
Enter with a limit at a fixed point of the gap, put the stop beyond a defined swing, and target a defined liquidity level.
Cancel the order if it is not filled within a set number of candles.

A rule like this produces a list of trades with an entry, a stop, an exit and a result in R. That list is data, and data can be judged.

What you cannot test

"The market maker is hunting my stops", "smart money is accumulating here" and "this looks like a good order block" are stories, not rules. They may be useful as context, but if you cannot say in advance exactly what qualifies, you cannot count it, and what you cannot count you cannot prove. This is the first reason the internet argues about ICT without ever settling it: one side shows hand-picked winners, the other side shows hand-picked failures, and neither side shows a full, rule-based sample.

Is there proof that ICT works?

There is no public, audited proof that "ICT" as a whole is profitable, and you should be suspicious of anyone who claims there is. Screenshots of winning trades prove nothing, because every method produces winners. What counts is the complete list of trades taken under one set of rules, including the losers.

There is, however, a fair reason to take the ideas seriously. Liquidity, structure shifts, imbalances and premium and discount describe real features of how price moves between obvious levels: stops rest above highs and below lows, and price often trades through them. That makes the concepts a reasonable starting hypothesis. A hypothesis is not an edge, though. The edge is a specific version of it that survives your own testing, with your own costs, on your own instrument.

Does ICT trading work for beginners?

It can, but beginners usually meet it in the worst possible form: hundreds of hours of free videos, dozens of overlapping terms and no single rule set. The concepts are learnable. The hard part is the discipline of picking one setup and taking it the same way every time.

If you are new, narrow the scope first. Learn one model from start to finish, for example the sequence of a liquidity sweep, a market structure shift and a Fair Value Gap entry, using the guide to trading Fair Value Gaps with entry rules. Ignore the other forty concepts until that one has a track record in your journal.

Why many traders lose with ICT

Losing with ICT rarely means the concepts are wrong. It usually means one of these things is going on. The full list is in 10 mistakes ICT traders make in their first year, but these are the ones that matter most for the question of whether it works.

No fixed rules. Every chart looks like a setup if you are allowed to choose the definition afterwards. This is hindsight with a nice vocabulary.
Vague definitions. One trader calls a move a CHoCH, another an MSS, a third a BOS. If you cannot say which one you require, your trades are not comparable. Pick one written definition and keep it.
Too small a sample. Ten trades in a row, good or bad, say almost nothing.
Cherry-picked memory. You remember the clean sweeps that worked and forget the ones you skipped, widened a stop on or closed early.
Risk that is too big. Even a real edge cannot survive position sizes that turn a normal losing run into a blown account.
Changing the method after every loss. If you switch setups every week, you never collect enough trades to learn anything.
Ignoring costs and execution. Spread, slippage and late entries are part of the result, and a chart screenshot hides them.

Notice that almost none of these are about the concepts themselves. They are about measurement and discipline, which is good news: both can be fixed.

Expectancy and sample size: the maths behind an edge

An edge is a positive expectancy: the average result per trade, in R, after losses. R is your initial risk on the trade, so a trade that loses its stop is -1R and one that reaches a 3R target is +3R.

For example, imagine a setup that wins 40% of the time, with an average win of 2R and an average loss of 1R. The expectancy is 0.4 x 2 minus 0.6 x 1, which is +0.2R per trade. The same setup with a 30% win rate would be negative. These numbers are only an illustration, not a claim about any real strategy. The point is that the win rate alone tells you nothing without the size of wins and losses, as explained in win rate vs risk-reward.

Sample size is the second half. A short run of trades is dominated by luck. As a rough example, a sample of 100 trades taken under identical rules says far more than 10, and even 100 can mislead if the market conditions were all the same. So your conclusion should always be provisional: "so far the expectancy looks positive or negative", never "it works".

How to test ICT yourself before risking real money

Pick one setup. For example: a sweep of sell-side liquidity in the London Kill Zone, a market structure shift, then an entry on a Fair Value Gap.
Write the rules on one page: bias, session, what counts as the sweep, what counts as the shift, the entry point, the stop, the target and the cancel condition. A complete ICT trading plan gives you a template.
Replay the past honestly. Scroll the chart bar by bar, take only the trades your rules allow and record every one, including the ugly ones.
Forward test on a demo or with very small risk. Past charts are easy to fit in hindsight. Live bars are not.
Journal every trade with the same fields: time, session, entry, stop, target, result in R and a screenshot. The ICT trading journal template lists what to track.
Review after a fixed number of trades that you decided in advance, and only then ask whether the expectancy is positive. If it is not, change one rule at a time and start a new count.

This process is boring, and that is the point. It is also the same process whether you trade by hand or hand the execution to a program, as covered in what an AI trading bot can and cannot do.

Case study: a live GBPUSD trade of the club's AI

Here is what one measured trade looks like. This was a live trade: the club's AI planned it before the session and then followed it on real bars until it closed. The result was a loss, and it is shown for exactly that reason.

The plan

Instrument and session: GBPUSD, London Kill Zone (02:00-05:00 New York time), Monday 12 October 2026.
Bias: long.
Setup: price entered the M15 Fair Value Gap between 1.32392 and 1.32416 and closed above it, at 1.32438. That close was the planned confirmation.
Entry: a long limit order at 1.32404, the consequent encroachment (the midpoint) of that gap.
Stop: 1.32333, beyond the protected M15/H1 low at 1.32347.
Target: 1.32683, the Asia high that had not yet been taken. A London / M15 impulse high at 1.32616 sat on the way.
Planned risk-reward: 3.93.
Invalidation: an M15 close below 1.32347 cancels the long, and so does a limit order that is not filled within 8 candles.

What happened

The limit order was filled at 07:45 UTC (03:45 New York time), inside the Kill Zone. Price moved in favour first: the best excursion was +0.93R, so the trade came close to one full risk unit in profit. It then turned back, and the trade closed at 09:33 UTC (05:33 New York time) at 1.32358, for a result of -0.7R. It did not reach the target at 1.32683, and it closed before the stop at 1.32333.

What this trade does and does not show

It does not prove that ICT works, and it does not prove that it fails. One trade is a sample of one. What it shows is what a testable trade looks like: the level, the entry, the stop, the target and the cancel condition were all written down before the fill, and the result can be measured in R without any argument. A loser like this one is not a failure of the method. It is a normal data point. The question is whether the full list of such trades, taken under the same rules, adds up to a positive expectancy.

What a measurable edge looks like in a journal

If you log every trade the same way, patterns appear that no memory can give you. You can filter by session and see whether your London trades differ from your New York ones. You can see whether trades with a full confirmation do better than the ones where you jumped in early. You can check how often the best excursion was large and the trade still ended at a loss, which is a signal about exits, not about the setup. You can read the average R of each setup separately and drop the ones that cost you money.

A journal also removes the most common self-deception, which is judging the method by your last three trades. Treat the weekly review as the moment where the numbers, not your mood, decide what you do next.

One plain warning, once: trading is risky, losses happen, and nothing here is financial advice. This article is education.

FAQ

Is ICT a scam or a real method?

The ideas describe real price behaviour, such as stops resting above highs and below lows, so it is not nonsense. But the label does not guarantee anything. Whether a rule set makes money for you is something only your own tested and journaled trades can show.

How many trades do I need before I can judge an ICT setup?

There is no magic number, but a few trades prove nothing. As a rough example, aim for around 100 trades taken under the same written rules, in different market conditions, and decide that number before you start. Treat any conclusion as provisional and keep tracking.

Why do many traders lose with ICT even when they know the concepts?

Knowing the concepts is not the same as having rules. Many traders define setups differently each time, risk too much, judge the method after a handful of trades and skip the journal, so they never see their real numbers.

Is ICT different from SMC when it comes to results?

The results do not depend on the label. They depend on the exact rules you apply. Whichever name you prefer, pick one defined setup, write it down and test it.

Can an AI trade ICT setups more consistently than a person?

A program follows written rules without hesitation, which removes some emotional errors, and every trade is logged in the same way. It does not remove risk or the need for a positive expectancy, so its trades must be judged by the same journal and sample size as anyone else's.