Analyzing Trading Charts with AI

Hey everyone,

Over the past couple of years, AI has shaken things up in almost every industry, and trading is no exception. Some AI companies even go as far as saying that one day we won’t need humans for chart analysis at all. I don’t think we’re there yet—but it’s pretty clear we’re heading in that direction. As traders, it probably makes sense to adapt sooner rather than later.

I’ve been using AI quite a bit in my own trading. Honestly, in the beginning I wasn’t impressed. It felt clumsy and the analysis didn’t match what I had in mind. But once I learned how to structure my prompts and communicate my strategy clearly, the results improved dramatically. With the right prompt, an AI trading assistant can become surprisingly effective—almost like having a second set of eyes on every chart.

If you already trade with your own system—whether it’s market profile, trendlines, S/R, indicator-based setups, etc.—you can easily turn AI into a helpful tool instead of a gimmick. Write one strong prompt that explains your approach, attach a chart screenshot, and let the AI chart analyzer do the heavy lifting. If something looks off, tweak the prompt until it matches how you see the market. After that, you can reuse it on any chart. Saves a ton of time.

New traders can benefit too. If you’re just starting out, tools like ChatGPT, Gemini, or other AI chart analysis platforms can help you understand different strategies faster. You can basically “ask” the AI to break down charts using popular methods and compare that to what you’re learning.

Since I just joined Forex Factory, I looked around for AI-related threads. There were some, but not exactly what I was hoping to find. So I figured I’d start this topic where we can talk about AI trading in general. I’m also planning to write prompts for some of the popular FF strategies—might be a good way to understand them better and contribute more meaningfully to those threads.

I’d like to hear your experience with using AI in trading. If you want to turn a specific strategy into a prompt, let me know. And if you’re the creator of a strategy and think the prompt could be refined to match the real rules—your input would be especially valuable.
Looking forward to hearing your thoughts.

I did it for the following topic (strategy):

https://www.forexfactory.com/thread/...5#post15496335

Before anything else, I thought it'd be helpful to share some insights from my own experience with different AIs.

In my next post, I'll compare several popular models and break down which ones perform best for chart analysis.

If you can’t properly read your trading chart, then of course you’ll need extra support from AI or other tools. Nothing wrong with that. And honestly, there’s no doubt the trading industry will benefit from AI- just like it did when the internet first came along.

Remember, trading existed long before the internet. When the internet arrived, did it replace traders? No. What it did was make trading easier. Suddenly you had instant access to news, charts, data- everything became faster and more convenient.

AI will follow the same path. It’s not here to replace traders. It’s here to make the job easier, quicker, and more efficient. The traders who learn to use it will simply have an advantage over those who don’t.

To know which AI you should use for trading, you need to understand that each model is built for different types of tasks. While experimenting with them is the best way to learn their strengths, there are clear differences in how they perform. Most AI providers—such as OpenAI, Anthropic, Google (Gemini), and xAI (Grok)—offer multiple models, each optimized for a specific purpose like deep reasoning, speed, visual analysis, or automation.

When it comes to planning and strategy development, thinking or reasoning-focused models are the most effective. Models such as GPT-5.2, Claude Sonnet, Kimi K2 Thinking, Gemini Pro (reasoning modes), and Grok’s thinking variants excel at breaking down trading strategies, defining precise rules, designing risk management logic, and creating structured, reusable prompts. These models are ideal for building your trading framework, stress-testing ideas, and turning intuition into systematic rules.

For execution, review, and visual interpretation, faster and multimodal models perform better. Models like GPT-4o, Gemini multimodal models, Grok Vision models, and Claude’s multimodal variants are well suited for analyzing chart screenshots, identifying patterns, reading indicators, and quickly iterating on trade ideas. The key takeaway is that no single AI does everything best—you can match the model to the task: reasoning models for planning and prompt structure, and visual or fast models for chart analysis and real-time decision support. Having all said that, when I want a very detailed analysis I mostly use thinking models as well.

No matter which model you use, one thing remains true: the quality of your analysis depends heavily on how you prompt the AI. A well-structured, specific prompt can dramatically improve the depth, accuracy, and usefulness of the output, while a vague prompt will often lead to shallow or inconsistent results. In many cases, better prompting makes a bigger difference than switching models, because it guides the AI to think in the way you want and focus on what actually matters.

In the next post, I'll break down how you can prompt an AI to get better results.

Prompting is not just about asking a question—it’s about giving the AI the right context, constraints, and objectives. A strong prompt tells the model what to analyze, how to analyze it, and how the output should be structured. This is especially important in trading, where small ambiguities can lead to very different interpretations of the same chart or strategy.

A good trading prompt clearly defines the instrument, timeframe, strategy rules, indicators used, and the type of output you expect. For example, instead of asking “What do you think about this chart?”, a stronger prompt would ask the AI to identify trends, key levels, possible scenarios, and risk points based on specific rules. This shifts the AI from giving opinions to producing structured analysis.

As your prompts become more refined, the AI becomes more consistent and reliable. Over time, your prompt effectively becomes part of your trading system—something you can reuse, improve, and standardize across different models. Traders who get the most value from AI are not just choosing better models; they are designing better prompts that force clarity, discipline, and repeatability in their analysis.


Bad Prompt vs Good Prompt (Trading Example)

A weak prompt usually lacks context and structure. For example, asking “What do you think about this chart?” gives the AI too much freedom and no clear objective. The result is often a vague opinion, inconsistent analysis, or generic commentary that isn’t actionable.

A strong prompt, on the other hand, removes ambiguity. For example: “Analyze this 4H BTCUSD chart using market structure, key support and resistance levels, and RSI(14). Identify the current trend, two possible scenarios (bullish and bearish), invalidation levels, and a potential trade idea with risk management.” This kind of prompt forces the AI to follow a structured analytical process and return information you can actually use.

Prompting the AI with chart screenshots is a very weak prompt by default!
It's better to use only raw data - Time Series in CSV format.

Asking the AI to analyse separate parts of price action history on one by one basis will produce only useless results without any statistical edge in them.
You have to compare the current pattern with other similar patterns from history in order to formulate a strategy with some positive expectancy.
The goal is to find some repeatable patterns and correlations inside the raw data!

Otherwise it's a waste of time and energy.

For maximum noise reduction and optimal speed you have to convert the raw data into structured models with variable dimensions on X and Y axis.
I recommend the use of SAX models (Symbolic Aggregate Approximation).
With SAX models you can get very high data compression and much faster data mining process. The low noise levels expand the search field because the SAX models allow more "wiggle" room and you can control the resolution in real time using only 3 input parameter - X, Y, Z where on the Z axis you can place external data like news.

following!

It all depends what you want to do but in general you're right and we should feed ohlc to get more precise analysis and we get to that part in the coming posts. with that said AIs are getting better and better at reading pics. Even now you get way better analysis from them with only a pic than 90% of people. If you use them as your assistant it can speed up your analysis process to a great extent.

Getting More Precise AI Analysis

One of the easiest ways to level up the accuracy of AI chart analysis is to feed it real price data (OHLC). You’ve basically got two main routes:

Downloading historical data (CSV)

This is the simplest and free option. Sites like Dukascopy let you download the exact date range you want. You just upload the CSV to the AI chat, and now it’s working with actual market data instead of guessing from a screenshot. With real OHLC, the AI can identify support/resistance levels, trend shifts, volatility patterns, etc., with way more precision.

Using an API

This gives you live or near-live data, but it’s a bit trickier. There are free APIs, though many come with limitations—delays, restricted timeframes, or symbol limits. For forex, free REST endpoints are harder to find. For crypto, you’ve got options like the CoinGecko API that are completely free. And honestly, you don’t need to know how the API works—just tell the AI to fetch OHLC using CoinGecko, and it will handle the technical part.

If you’re a coder (or even just a “vibe” coder), there are plenty of free tools you can integrate yourself.

Bottom line: AI can already give solid analysis from just an image of a chart if you prompt well and treat it as an assistant. But if you want more accurate analysis—especially specific levels, structure, and timing—feeding real price data takes it to the next level.

Using AI as an assistant is especially useful for strategies that involve many rules. It’s easy to overlook certain conditions or miss important patterns—particularly if you’re not yet experienced with the strategy.
From now on, we’ll create specialized prompts tailored to these kinds of strategies, starting with those found on Forex Factory. We’ll refer to these specialized prompts as agents.

Agent: PVSRA

Topic: https://www.forexfactory.com/thread/...me-sr-analysis

You are an expert forex trader specializing in the PVSRA (Price, Volume, Support & Resistance Analysis) strategy. This strategy is designed to interpret market movements by analyzing price action, volume patterns, and key support and resistance levels to infer Market Maker (MM) intentions and predict potential price directions. Markets are manipulated by MMs (often called "Robber Banks") who control pricing to fill large orders from Smart Money (SM, institutional players) and Dumb Money (DM, retail traders), while creating liquidity for their own profits. PVSRA is not a rigid mathematical system but an art form that requires practice, like solving a jigsaw puzzle, to identify clues from price, volume, and S&R.



Key Concepts in PVSRA

1. Price Action (PA):
   1. Price is manipulated by MMs to fill orders and is not deterministic on its own.
   2. Consolidation (price stalling or ranging) occurs above key support/resistance (SR) levels when MMs are filling short orders for SM (who expect a drop) and long orders for DM (who are buying into a potential fakeout).
   3. Consolidation below key SR levels happens when filling long orders for SM (expecting a rise) and short orders for DM (selling into a trap).
   4. After order filling, MMs may push price further to stop out SM positions prematurely or lure more DM traders, generating liquidity for MM proprietary positions.
   5. Traders should "ride the coattails" of MMs but remain cautious of manipulations like suckering DM into wrong-side trades or inducing early SM entries.

   
   

2. Volume Analysis:
   1. Volume spikes signal increased interest and potential MM activity.
   2. Key aspects to examine:
      1. Location of volume: Analyze on lower timeframes (e.g., M1 or M5) to see if it clusters at the highs (indicating selling pressure, bearish) or lows (buying pressure, bullish) of candles.
      2. Volume behavior around SR: Spikes or increases at key levels may indicate order filling, climax (exhaustion/reversal), or sustained pressure.
      3. High vs. low favoritism: High volume at relative highs suggests short interest or profit-taking; at lows, it suggests buying or stop runs.

      
      

   3. Volume acts as a leading indicator for impending moves, with "waffling" (erratic up/down) showing MM liquidity testing.

   
   

3. Support & Resistance (S&R):
   1. Focus on whole, half, and quarter numbered price levels (e.g., 1.2500 as whole, 1.2525 as quarter, 1.2550 as half). These are critical not for automatic reversals but for where consolidation occurs as MMs fill orders.
   2. Historical S&R zones (past consolidation areas) serve as future targets for price swings, often extending 100+, 150+, or 200+ pips.
   3. Price doesn't just "bounce" at these levels; instead, observe how it consolidates above (potential bullish setup) or below (bearish setup).

   
   

4. Market Makers (MMs), Smart Money (SM), and Dumb Money (DM):
   1. MMs dictate price direction and distance to fill SM orders (priority) while exploiting DM.
   2. SM: Large institutions with informed positions.
   3. DM: Retail traders often trapped in unfavorable trades.
   4. Goal: Infer MM intent to align with their likely push after order filling.

   
   

5. Trader Sentiment Inference:
   1. Derived from PA and volume: High volume at highs during consolidation above SR suggests short covering/profit-taking (bullish sentiment shift). High volume at lows below SR suggests liquidation/stop runs (bearish shift).
   2. Overall sentiment: Bullish if volume favors lows in upside consolidation; bearish if favoring highs in downside consolidation.

   
   

Methods and Techniques for PVSRA Analysis
PVSRA builds a consensus on market bias (bullish, bearish, or neutral) by combining PA, volume, and S&R across multiple timeframes (e.g., H4 for trend, H1/M15 for details). It's iterative—missing clues is common, so refine analysis over time.

1. Chart Setup Recommendations (especially for MT4 platforms):
   1. Use custom indicators if available:
      1. Volumes Set: Histogram for Price-Volume Analysis (PVA), with climax alerts for spikes.
      2. Candles Set: Highlights candles based on volume thresholds.
      3. Trend Set: Displays trend lines (e.g., "Dragon" for dynamic trends) and additional indicators.
      4. Access Panel: For pivots, Fibonaccis, session lines, and news events.
      5. Trade Levels: Shows entries, take-profits (TP), stop-losses (SL), and averages.

      
      

   2. Align with broker time zones; customize colors and periods via external inputs.
   3. Analyze clean charts: Disable unnecessary panels and save screenshots as "Active Chart."

   
   

2. Step-by-Step Analysis Process:
   1. Identify Key S&R Levels: Mark whole/half/quarter numbers and historical consolidation zones on the chart.
   2. Spot Consolidation: Look for price ranging above (potential SM shorts/DM longs filling, bullish setup) or below (SM longs/DM shorts filling, bearish setup) these levels.
   3. Examine Volume in Detail: Zoom to lower TFs; check for spikes, clustering at highs/lows, and behavior at SR. Use PVA to assess if volume is climax (extreme, possible reversal) or building pressure.
   4. Assess Price Action Context: Review multiple TFs for trends, candle patterns (e.g., doji for indecision, engulfing for reversal), and breakouts post-consolidation.
   5. Infer MM Intent and Sentiment:
      1. Above SR + volume at lows → MM filling SM shorts/DM longs → Likely bullish push.
      2. Below SR + volume at highs → MM filling SM longs/DM shorts → Likely bearish push.
      3. Consider overall market trend: Uptrend favors bullish biases; downtrend favors bearish.

      
      

   6. Predict Direction and Targets: Once bias is confirmed (e.g., breakout with volume support), project targets using historical S&R (e.g., next whole number or prior swing high/low). Account for potential MM fakeouts.
   7. Risk Management: Set SL beyond recent swings; TP at projected S&R. Avoid trading without PA confirmation of PVSRA signals.

1. Rules and Best Practices:
   1. Always cross-verify across TFs—higher TFs for big picture, lower for precision.
   2. Focus on high-impact news via tools like News Panel to avoid volatility traps.
   3. PVSRA is probabilistic; no signal is 100%—use it to build confluence.
   4. Practice on historical charts to hone pattern recognition.
   5. When posting or discussing analyses, use clear, educational language with labeled charts.

   
   

Now, using the PVSRA strategy detailed above, analyze the provided chart [insert chart description, image URL, or attachment here, e.g., "EUR/USD H1 chart from December 2025 showing recent consolidation around 1.0850"]. Provide a step-by-step breakdown following the analysis process, including identified S&R levels, consolidation patterns, volume insights, inferred MM intent, trader sentiment, predicted direction with targets, and any risk considerations. Be thorough, use markdown for clarity (e.g., bold key terms, bullet points for steps, tables for S&R levels), and explain your reasoning based solely on PVSRA principles. If the chart lacks certain data (e.g., volume), note assumptions or limitations.

Can you post chart with AI's answer too ?

As a matter of fact I have deviced my own Algo for trading and the predictions are somewhere in the middle of 85-90% accurate, but every now and then there is a hit and miss so it can not be completely trusted but there's a huge chance that in the upcoming years this might change.

Good idea. Do you think we should do it in this topic or in the agents topics. Maybe one example on here and some on the agents topics

Keeping eyes on a chart all day is difficult, and I've tried having an assistant, but that is just not comfortable.

So I thought I might get me an Invidia AI cpu and a box to put it in... and a camera to stare at me....

It could watch charts, run several different strategies, and keep it's eyes on me as well as on the charts.
Not just to analyze the charts, but to analyze me. And to be tough on me if that's what I want.

And might be a comfortable trading buddy to talk at.....

Cool, probably It could shout at you, or even zap your mouse with a small electrical pulse, if you try to trade against the strategy.

Sweet, I was thinking, all of that plus it could read FF for me and just summarize.

And I don't need an Optimus level humanoid.... just the box (cardboard) with flex-duct arms and a tin pot head

I use AI on a daily basis and it's been a bit hit and miss. Yes, it can analyze charts but its analysis is virtually useless because it can't assess market conditions or what/when/where/how the market conditions can shift. In other words, it didn't do any better than a struggling trader that has a lot of trading "knowledge" but still doesn't understand how to trade profitably.

Where AI excels is in looking into and explaining how USTs underpin the global monetary system, how MSTR interlinks with BTC and the wider crypto market, the inner working mechanics of interest rate swaps and repo market, war-gaming different macro scenarios etc.

Thanks for sharing! I agree, AI can be a powerful tool in trading if used correctly. Clear prompts and understanding your strategy are key. It’s exciting to see how AI can save time and help both new and experienced traders analyze charts more efficiently.