LLMs for trading: what ChatGPT can and cannot do with a chart
What ChatGPT can and cannot do with a trading chart: it explains concepts and checks strategy code, yet it cannot predict prices, read live data, or trade.
Someone screenshots a 5-minute chart of the Nasdaq, pastes it into ChatGPT, and types "should I go long or short?" The reply comes back in seconds, fluent and confident, with a suggested entry, a stop and a target. It reads exactly like a trader talking. That fluency is the thing to be careful with, because the model has no idea whether it is right.
Large language models are genuinely useful around trading, and they are also being sold as something they are not. This is the honest split: what an LLM like ChatGPT can do with a chart and a trading idea, and what it cannot do no matter how the prompt is worded.
What a language model actually is
An LLM is a text predictor. Trained on an enormous amount of writing, it estimates the most plausible next word given everything so far, then does it again, one token at a time. That is the whole mechanism, and it is astonishingly good at producing text that sounds like a knowledgeable human, because that is exactly what it was built to do.
Notice what is missing there: a price feed, a model of supply and demand, a memory of what the market did five minutes ago, any notion of being correct. The model is not reasoning about the Nasdaq. It is completing a passage of text about the Nasdaq. Most of the confusion about "AI trading" comes from mistaking the second thing for the first, and the fuller version is in what is real in AI trading.
An LLM predicts the next word, not the next candle. It can explain a strategy beautifully and still have no ability to tell you where price is going. Fluent is not the same as informed.
What it genuinely helps with
Used as what it is, a language tool, ChatGPT earns its place in a trader's week. The honest list:
- Drafting and checking code. It can write a first pass of a NinjaScript indicator, spot a bug in a loop, or turn a rule you describe in plain English into something closer to code. You still test it, but it clears the blank page.
- Summarizing. Paste a long strategy document or a dense forum thread and it compresses it. Paste a description of an economic release and it tells you what that release usually moves.
- Learning faster. It is a patient tutor that never sighs. For a beginner working through how a trading algorithm works, it turns a two-hour search into a ten-minute conversation.
Every one of those is a language task: explain, rewrite, summarize, teach. That is the model working well inside its actual competence. Now the other column.
| What an LLM does well | What it cannot do |
|---|---|
| Explain a concept like profit factor or drawdown | Predict the next price move |
| Draft and debug strategy code | Read a live chart or the order book |
| Summarize a long document or thread | Access real-time market data on its own |
| Translate a rule from English toward code | Know whether its own answer is true |
| Teach a beginner patiently | Follow a rule set with discipline in the market |
Where it reads a chart, and where it only appears to
Modern versions of ChatGPT accept images, so you can upload a chart and it will describe it. It reads the axis labels, names the pattern it thinks it sees, and comments on the trend. Some of that is genuinely accurate, because it can often read text and obvious shapes off an image.
But "describe this picture" and "tell me what happens next" are different requests, and the model quietly slides from the first to the second. It cannot read exact price levels off a screenshot, it does not know what happened just outside the frame, and it cannot see the order book or the next tick. When it says "this looks like a bullish flag, consider a long," it is pattern-matching against the chart descriptions in its training data, not analyzing your market. The label may be right. The prediction attached to it is a guess dressed as analysis.
The two hard limits
Two limits matter more than all the others, and neither goes away with a better prompt.
No real-time data. A base language model has no live market feed, and knows nothing after its training cutoff. Even tools that browse the web or call an API are fetching delayed, summarized data, not trading off the tape. Ask "where is MNQ right now" and the honest answer is that it does not know. The dangerous versions answer anyway.
Hallucination. Because the model generates the most plausible-sounding text, it will invent a confident, specific, wrong answer as readily as a correct one, and it cannot tell the two apart. It will cite a price that never printed, describe an indicator value it did not compute, or explain the "reason" for a move it made up. In casual writing a hallucination is embarrassing. Attached to a trade, it is money.
Why "predict the price" is the wrong request
Set the tool aside for a second, because this part is not really about ChatGPT. Short-term price movement in a liquid futures market is close to unpredictable in the strict sense: the next move is dominated by new information that, by definition, is not yet known when the model answers. Expecting any model, language-based or otherwise, to reliably call the next candle is asking for something the market does not offer. Anything that appears to do it in a screenshot is either cherry-picked or fooling you. The same point is argued in can AI predict the stock market and in machine learning in trading.
How to use ChatGPT around trading well
The productive stance is simple: let it handle language, and keep the decisions yourself. Ask it to explain, to summarize, and to draft code you will test. Do not ask it to predict, and treat any confident market call as the model doing what it always does, producing plausible text. Verify anything factual it tells you, because it will state a wrong number in the same tone as a right one. A practical walkthrough of that workflow is in ChatGPT for trading. The model is a superb assistant and a poor oracle, and the trouble starts the moment you ask the assistant to be the oracle.
Frequently asked questions
Can ChatGPT predict the stock market?
No. A language model predicts the next word in a passage of text, not the next move in a market, and it has no live price data of its own. Short-term futures moves are driven by information that is not yet known when the model answers, so any specific forecast it gives is a plausible-sounding guess rather than analysis.
Can I upload a chart screenshot and get a trade from it?
You can upload a chart and the model will describe what it sees, sometimes accurately for labels and obvious shapes. It cannot reliably read exact price levels, it does not know what happened outside the image, and it cannot see the order book or the next tick. When it names a pattern and attaches "consider a long," the label may be fair but the prediction is not something it can actually compute.
Educational content, not investment advice. Trading futures involves substantial risk of loss and is not suitable for every investor. Any reference to Algomyth results is hypothetical, simulated over historical data, and shown in full on the performance page; past performance does not guarantee future results.
Twenty years, drawdowns included
Every engine publishes its full history repriced at today's values, with the losing years and the deepest drawdowns printed as plainly as the profits. Verify first, decide later.
More from AI & machine learning
AI & machine learningHow Hedge Funds Actually Use Machine Learning (Not What the Ads Say)Hedge funds use machine learning for execution, risk and data research, not price prediction. What that takes, and why the retail AI bot rarely holds up.
AI & machine learningNeural networks for price prediction: why they overfit so easilyNeural network trading models fit price history almost perfectly and predict the next week badly. Why non-stationarity, noise and data hunger break them.AI & machine learningAI vs Rule-Based Trading Systems: Which Wins, and WhenAI vs rule-based trading, compared on interpretability, overfitting and debugging: when a black box wins, and when plain rules win.