AI prompts for ICT/SMC trading: how to ask AI for analysis that's actually useful

Published July 3, 2026 · AI + ICT/SMC · 7 min read

You write to ChatGPT or Claude "analyze the NQ chart and tell me if I should buy or sell" and get three paragraphs of generalities: "consider the trend", "use a stop loss", "manage your risk". It's not that the AI doesn't know — it's that the prompt didn't give it a framework. The quality of the analysis you get is proportional to the structure you provide.

In this guide: why vague prompts fail, what a serious ICT/SMC prompt has to include, and a complete example you can copy and adapt.

Why "analyze the chart" produces garbage

A vague prompt forces the AI to decide for you everything an institutional trader defines before looking at the chart: which timeframe rules? which concepts matter? which session do you trade? how much do you risk? Faced with that vacuum, the model answers with the average of everything it ever read about trading — which is, by definition, generic.

The solution is to treat the AI the way you'd treat a new analyst on your desk: you hand it your complete trading framework and ask it to work inside it, not to invent one.

The 5 blocks of an ICT/SMC prompt that works

1. Instrument and session context

Exact instrument, style (scalping, day trading, swing) and the session you trade. For ICT, the session isn't decoration: the London and New York killzones concentrate the liquidity moves the method looks for. An analysis without a session is an analysis of some other system.

2. Explicit multi-timeframe structure

Top-down analysis is the skeleton of ICT/SMC: one timeframe for the bias and another for execution. Tell the AI which ones are yours — "analysis on 15m, entries on 5m" — and ask it to respect the order: structure and bias on the higher timeframe first, then confirmation on the lower one. Without this, the model mixes timeframes and the result is incoherent.

3. Concrete concepts, not "use ICT"

"Use ICT concepts" is as vague as saying nothing — the repertoire has dozens: Fair Value Gaps, Order Blocks, Breaker Blocks, BOS, CHoCH, liquidity pools, equal highs/lows, premium/discount, OTE, Power of Three, Judas Swing... Pick the 3–6 that your strategy actually uses and name them. The difference in the answer is dramatic: the AI goes from reciting theory to hunting for specific setups.

4. Risk management with numbers

Minimum acceptable R:R, risk per trade, maximum trades per day and the stop criterion (structure? below the order block? ATR?). This turns the answer from "an opinion about the market" into "an executable plan": the AI can discard setups that don't reach your R:R instead of recommending whatever.

5. Output format

Ask for the answer in a fixed structure: bias → key levels → setup with entry/SL/TP → invalidation. If you don't define the format, you get an essay; if you define it, you get a plan.

Example: complete prompt ready to use

Act as an institutional trader specialized in ICT/SMC.

INSTRUMENT: NQ (Nasdaq futures) | SESSION: NY Killzone
ANALYSIS: 15m for bias → ENTRY: 5m

MY FRAMEWORK (work ONLY with this):
- Structure: BOS / CHoCH on 15m to define bias
- Zones: Order Blocks and Fair Value Gaps on 15m
- Liquidity: equal highs/lows and pools above/below prior
  highs/lows
- Entry: on 5m, after a liquidity sweep + displacement
  that leaves an FVG, enter on the pullback to the FVG
- I only trade in the direction of the 15m bias, in premium
  (sells) or discount (buys)

RISK MANAGEMENT:
- Minimum R:R 1:2 (discard setups that don't reach it)
- Risk per trade: 1% | Max. 3 trades/day
- SL: below/above the structure that validates the setup

RESPONSE FORMAT:
1. 15m bias and why (structure, not indicators)
2. Key zones marked (OB, FVG, liquidity) with prices
3. Concrete setup: entry, SL, TP, resulting R:R
4. What invalidates it (if X happens, the analysis dies)

With the same question as before ("should I buy or sell NQ?"), this framework produces an analysis you can execute, audit and — more importantly — invalidate. If the AI can't tell you what kills the setup, it wasn't a setup.

The most common mistakes

Asking for predictions: "where is price going?" is astrology. Ask for conditional scenarios: "if it sweeps the liquidity at X and displaces, then...".
Mixing frameworks: asking for ICT + RSI + Elliott + fundamentals in the same prompt produces a salad. One framework per prompt.
Trusting without validating: the AI structures your analysis, but it doesn't know whether your strategy is profitable. That gets answered with data: log the trades and validate with Monte Carlo, not with how confident the text sounds.
Rewriting the prompt every day: the value is in consistency. The same framework, every day, is what lets you compare results and adjust.

Generate your ICT/SMC prompt in 2 minutes

The PromptDesk Pro strategy generator builds this kind of prompt for you with a multi-timeframe wizard: you pick instrument, session, timeframes, your ICT/SMC concepts (25 available: FVG, OB, BOS, CHoCH, liquidity pools, PO3, OTE...) and your risk management, and it delivers the complete prompt, consistent and ready for ChatGPT, Claude or whichever AI you use. You can also describe your strategy in your own words or upload an image of your setup.

Create my strategy →

This content is educational and does not constitute financial advice. No prompt or AI analysis guarantees results; always validate your strategy with real data before risking capital. Trading involves risk of loss.