AI Stock Analysis Prompts: Six Copy-Paste Templates That Avoid the Usual Traps
Most bad AI stock analysis comes from a bad prompt, not a bad model. A vague question gets a vague, agreeable, differently-shaped answer every time. These six templates are built to get description instead of prediction, a fixed structure you can compare across charts, and an honest list of what the model could not see.
The short answer
A good chart analysis prompt does four things: it gives the model the context it cannot see (symbol, timeframe, indicator settings), it demands a fixed output order, it bans predictions and trade suggestions, and it asks the model to list what it could not see. The six templates below cover a first read, levels, pattern checks, two-timeframe comparison, a devil's advocate review, and a journaling summary. Paste them into ChatGPT, Claude or Gemini with a chart screenshot attached.
Why the prompt matters so much
General-purpose models are good at reading a chart image and bad at three things that a prompt can fix. They answer in a different shape every time. They do not know what is not in the picture. And they lean toward agreeing with the framing of the question. A prompt that addresses all three turns a chatty assistant into a fairly disciplined first-pass technician.
Each template below uses square brackets for the parts you fill in. Attach the chart screenshot in the same message. If you have not already, read how to screenshot a chart; no prompt rescues a capture with the price axis cropped off.
1. The structured first read
Use this as your default. It is the template to save somewhere you can paste from.
Analyse the attached chart. Describe it; do not tell me what to do with it.
Context: [SYMBOL], [TIMEFRAME] candles, roughly [NUMBER] bars visible. Indicators shown: [LIST WITH SETTINGS, e.g. 50 and 200 SMA, RSI 14].
Answer in exactly this order, with these headings:
1. Trend and structure: the sequence of highs and lows, and whether it is intact or has broken.
2. Key levels: each zone described by where it formed and how many times it was tested. Approximate prices only, labelled approximate.
3. Patterns: name any chart or candlestick patterns, and say whether each is complete or still forming.
4. Indicators: read only what is visible in the image.
5. Legibility: how clearly could you read this chart, low, medium or high, and why.
6. What you cannot see: anything cropped, unreadable or missing.
Do not predict direction. Do not suggest entries, exits, stops or targets.
The legibility line keeps the confidence honest: it asks how clear the picture is, not how likely a trade is to work. The last section is the most important. It forces the model to surface blind spots instead of quietly filling them in.
2. Levels only
When you want a focused list of support and resistance zones without a full read.
From the attached [TIMEFRAME] chart of [SYMBOL], list the horizontal zones where price has reacted more than once.
For each zone give: approximate price range, whether it acted as support, resistance or both, roughly how many reactions, and how recent the last reaction was.
Order them from nearest to current price to furthest. If you cannot read the price axis precisely, say so and describe zones by position instead (for example "the March lows").
Asking for a range rather than a single number matches how levels actually behave, as covered in how to draw support and resistance.
3. Pattern check
When you think you see a pattern and want an independent opinion. Note that you do not name the pattern in the prompt.
Look at the attached chart of [SYMBOL] on the [TIMEFRAME]. Are there any recognised chart patterns or candlestick patterns in the most recent [NUMBER] bars?
For each one: name it, describe exactly which swings or candles form it, say whether it is complete, and state what would invalidate it. If nothing qualifies cleanly, say that rather than forcing a pattern.
"If nothing qualifies, say that" is the key line. Without it, models tend to find a pattern because you asked for one.
4. Two-timeframe comparison
Attach two screenshots of the same symbol: one higher timeframe, one lower. This is the AI version of the top-down method in multi-timeframe analysis.
Image 1 is [SYMBOL] on the [HIGHER TIMEFRAME]. Image 2 is the same symbol on the [LOWER TIMEFRAME].
1. Describe the trend and structure on each separately.
2. Say whether they agree or conflict.
3. Identify which higher-timeframe level, if any, the lower-timeframe chart is currently trading near.
4. List anything visible on one chart that is invisible on the other.
Description only, no trade suggestions.
5. Devil's advocate
The most useful prompt for someone who already has a view. It counters the model's tendency to agree with you.
I have a [BULLISH / BEARISH] read on the attached [SYMBOL] [TIMEFRAME] chart because [YOUR REASONS].
Argue the opposite case as strongly as the chart allows. What on this chart contradicts my read? Which of my reasons is weakest? What would I expect to see if I were wrong?
Base everything on what is visible in the image.
Here you deliberately state your view, but you ask the model to attack it rather than confirm it. That reverses the agreeableness problem and turns it into something useful.
6. Journal summary
For keeping a trading journal of what charts looked like at the time you looked at them.
Summarise the attached [SYMBOL] [TIMEFRAME] chart in five bullet points for a trading journal: trend, nearest level above, nearest level below, any pattern in progress, and one thing that is unclear. Keep each bullet under fifteen words. Include today's date: [DATE].
Short, dated, fixed-format notes are what make a journal useful later. You can look back and compare what you thought the chart showed with what happened next, which is the fastest honest feedback loop available.
Lines to never put in a prompt
- "Should I buy?" It pushes the model from description into prediction, where it will still answer confidently.
- "Give me a price target." Same problem, with a specific number attached, which feels more authoritative and is not.
- "This is bullish, right?" A leading question. You will get reasons you are right.
- "Be accurate." It changes nothing. Asking it to list what it cannot see does far more.
When typing prompts stops being worth it
Templates work well for a handful of charts. The friction appears at volume: filling in brackets, re-attaching context, and keeping reads in a consistent format across sessions. That workflow problem, rather than any extra intelligence, is what dedicated tools such as ChartCheck are built around: you set up how you trade once, and every screenshot comes back in the same structure with a legibility-based confidence level and a "what I couldn't see" list. If you only read a few charts a week, the prompts above are enough.
Frequently asked questions
What is the best prompt for AI stock chart analysis?
One that supplies context the model cannot see (symbol, timeframe, indicator settings), asks for a fixed output order, bans predictions and trade suggestions, and asks the model to list what it could not see. Template 1 above does all four.
Do these prompts work in ChatGPT, Claude and Gemini?
Yes. They are plain-language instructions and work in any assistant that accepts an image. Output quality depends mostly on the screenshot: price axis visible, timeframe visible, and enough bars for context.
Can a prompt make AI predict stock prices accurately?
No. A prompt can make the description of a chart more structured and more honest, but it cannot give the model information about the future that is not in the image. Prompts that ask for predictions get confident answers, not reliable ones.
Why should I avoid telling the AI my position?
General-purpose assistants lean toward agreeing with the framing of a question. If you say you are long, the analysis tends to find reasons to support that. Ask neutrally, or use the devil's advocate template to have it argue against your view.
Related Articles
- Using ChatGPT for Chart Analysis: What Works and Where It Breaks
- Free AI Chart Analysis: What You Actually Get, and the Hidden Costs
- How to Screenshot a Trading Chart So AI Can Actually Read It
Scope
This article is educational and is not financial, investment, or trading advice. Nothing here is a recommendation to buy, sell, or hold any security, cryptocurrency, currency, commodity, or derivative. Technical analysis describes what a chart has already done; it does not predict what it will do, and every pattern described here fails a meaningful share of the time. Trading involves risk of loss. Do your own research and consult a licensed financial professional before making any trading decision. ChartCheck is made by the author of this site.