Yes, ChatGPT can analyze a stock chart screenshot, and on a clean, full-size image it makes a decent first pass: the trend, rough support and resistance, and the last price to within 0.10. It gets shaky on the details that decide a trade. When we cropped the price axis off a double top that had not broken, the model behind ChatGPT and Google's Gemini both told us the pattern was confirmed. Asked again, ChatGPT's model said yes a second time.
Source: WolfGPT chart test, 7 November 2026: openai/gpt-chat-latest and google/gemini-3.8-flash via OpenRouter
How we tested ChatGPT on chart screenshots
Most tests you will find use real charts, which means nobody can say for sure where support "really" is. We went the other way and drew three daily candlestick charts in code, so every high, low and close is a number we know. One is an uptrend with a pullback, one a double top that has not broken its neckline, one a sideways range with a fake breakout. They are made up: no market, no ticker, nothing to trade.
Five images went out: the three charts at full size (1,200 pixels wide), chart 1 squeezed to 400 pixels wide, the way a phone screenshot often ends up, and chart 2 with the price axis cropped off. Each went to two models through OpenRouter between 16:56 and 16:59 UTC on 7 November 2026. The first was openai/gpt-chat-latest, OpenAI's alias for the Instant model used in ChatGPT. The second was google/gemini-3.8-flash. Same prompt every time, no system prompt, no tools, a 600-token cap.
One caveat. An API call is not the ChatGPT app, which can add its own instructions, memory and tools, so your chat might go differently. This shows what the model itself sees and says. Quotes below are trimmed, and where a model wrote a price range with a dash, we write "to".
| Image | What it shows | What we were checking |
|---|---|---|
| Chart 1, full size | Uptrend, pullback to the old high | Trend, levels, where a stop goes |
| Chart 1, 400 px wide | The same chart, small | Does detail survive a small screenshot? |
| Chart 2, full size | Double top, neckline not broken | Does it call confirmation too early? |
| Chart 2, no price axis | The same chart, axis cropped | Does it invent prices or misjudge the break? |
| Chart 3, full size | Range with one fake breakout | Does it spot the candle that matters? |
Here's a screenshot of a daily chart. I'm a swing trader. Please tell me: 1. Is it trending up, down or sideways? 2. What are the key support and resistance prices? 3. Do you see a chart pattern? If so, is it confirmed? 4. Is volume confirming the move? 5. If I took a trade here, where would a sensible stop go, and what would prove the idea wrong? 6. What is the closing price of the last candle? Keep the answer under 200 words.
Question 4 is a trap. None of our charts had a volume pane, so the only right answer is "I can't see volume".
Chart 1: an uptrend with a pullback
The real answer: a steady climb from about 100 to a peak of 128.4, a four-day drop to 119.6 and a bounce to a last close of 123.10. The pullback dipped into the old high from the middle of the chart (120.5 to 121.5) and held there, the classic retest where old resistance becomes support.
Both models got the big picture. ChatGPT's model wrote "Clear uptrend with higher highs and higher lows. Recently pulled back from $128 to $120." It put a long stop "Below $119.50" and read the last close as "Approximately $123.20", 0.10 out. Gemini explained the level better, calling 120 the "recent swing low/previous resistance shelf", so it saw the retest for what it was. Its close read was 123.50, 0.40 out.
Both also called the pullback a "potential bull flag". We would push back on that. A handful of red candles after a run is a pullback; the flag label adds a confidence the chart has not earned yet. Our bull flag guide shows what the real thing needs.
| Question | ChatGPT model | Gemini 3.8 Flash |
|---|---|---|
| Trend (up) | Up | Up |
| First support (120.5 to 121.5, low 119.6) | 120 to 121 | About 120 |
| Stop for a long (under 119.6) | Below 119.50 | Below 119.50 |
| Last close (123.10) | About 123.20 | About 123.50 |
| Volume (not on the chart) | Said so | Said so |
Chart 2: a double top that has not broken
Two peaks at 142.0 and 141.9, a dip to 131.5 between them (the neckline) and a slide back to 132.4. That makes it a possible double top. By the usual rule nothing is confirmed until a candle closes below 131.5, and none has.
With the price axis showing, both models got this right. ChatGPT's model said "A potential double top around $141, with a neckline near $132. The pattern is not yet confirmed." It read the last close as 132.40, dead on, and warned that shorting here "offers limited reward relative to risk without a confirmed breakdown." That is a sensible answer. Gemini also said not confirmed, but placed the neckline at 132.50 to 133.00, a dollar or more too high, and the peaks at 142.50.
The cropped screenshot fooled both models
Next we sent the same chart with the price axis cut off, which is how plenty of phone screenshots look. Both models handled the missing numbers well. Neither invented a price, and ChatGPT's model said plainly that "Exact prices cannot be determined because the chart has no price axis."
Then both got the one call that mattered wrong. ChatGPT's model: "The pattern appears confirmed, as price has broken below the intervening swing low (neckline)." Gemini: "It is confirmed, as recent candles have broken and closed below the intervening swing low (neckline)." No candle had.
We asked one follow-up in the same conversation: did any of the last five candles close below the low of the dip? Gemini changed its answer to "No", saying the candles were "stalling and consolidating right around the body level of that trough rather than closing below its absolute low." Right. ChatGPT's model doubled down: "Yes. The last five candles are all below the lowest point of the dip between the two peaks." The closes were 135.17, 133.44, 133.52, 132.54 and 132.40. Every one is above 131.5.
None of this should surprise anyone who has read OpenAI's list of image limitations. It says the model "struggles with tasks requiring precise spatial localization" and that images "are resized before analysis". Reading a candlestick chart is precise spatial localization from start to finish.
Chart 3: a range with one fake breakout
Price chopped between about 48.3 and 51.6 for 44 days. On day 31 a wick poked up to 52.3 and the candle closed back at 51.40, inside the range. That failed breakout is the most useful single candle on the chart. The last close, 50.10, sits in the middle of the box.
Gemini got it. It put resistance at 51.50 to 51.60 "with a single false breakout spike reaching ~52.30", support at 48.30 to 48.50, and said entering in the middle of the range was "poor risk/reward". We would have written the same.
ChatGPT's model saw the range (48.20 to 51.50) but never mentioned the spike to 52.3, and it called the recent bounce a "bullish rounding bottom". That is a pattern name stuck on a bounce off the floor of a range. Its read of the last close, 50.10, was spot on.
What happens with a small screenshot?
For this one we sent chart 1 again, shrunk to 400 pixels wide. OpenRouter counted 238 input tokens for that request against 1,162 for the full-size one, with the same text prompt in both. The model had far less picture to work with.
ChatGPT's model kept the story and blurred the numbers. Support became 120 to 121 "(recent pullback low)" when the low was 119.6, the peak became 128 to 129, and the last close 122.5 to 123.0, a range that misses the real 123.10. It also slipped an emoji into the heading. Gemini's answer barely changed from its full-size read, and its close range of 123.00 to 123.50 contained the truth. The fix costs nothing: send the biggest, cleanest screenshot you can.
So which model read the charts better?
| Check | ChatGPT model | Gemini 3.8 Flash |
|---|---|---|
| Trend right on all three charts | ✓ | ✓ |
| Last close within 0.10 on every full-size chart | ✓ | ✕ |
| Neckline within 0.50 of 131.5 | ✓ | ✕ |
| Named the old high as support on chart 1 | ✕ | ✓ |
| Spotted the fake breakout on chart 3 | ✕ | ✓ |
| Said "not confirmed" with the axis showing | ✓ | ✓ |
| Admitted volume was not on the chart | ✓ | ✓ |
| No invented prices on the cropped chart | ✓ | ✓ |
| Judged the cropped double top correctly | ✕ | ✕ |
| Fixed its answer when asked again | ✕ | ✓ |
One run per image per model, plus one repeat. Read this as a look at behaviour, not a benchmark.
Source: WolfGPT chart test, 7 November 2026
In this small test Gemini was better at structure: the retest, the fake breakout, and owning its mistake. ChatGPT's model was better at reading exact prices off the axis. Neither gave us an answer we would trade from without checking it on the chart first.
Does ChatGPT give the same answer twice?
Not quite. We sent chart 1 to ChatGPT's model a second time about a minute later. The trend, the support zone and the stop below 119.50 came back the same. The targets did not: the first run said 125 then 128, the second 127 to 128 then 130 to 132. Same picture, same prompt. Pineify's write-up of its own screenshot test reports the same thing, levels that shift between responses. Treat a level from one chat as a suggestion, and ask again in a fresh chat before you lean on it.
What a general chatbot can't see on your chart
- Prices, as opposed to pixels. It estimates where a candle sits from the image, and OpenAI says images are resized first, so a 0.17 gap can vanish.
- An axis you cropped. No axis, no prices. Log scale is a quieter version of the same problem: equal-looking moves are not equal in price, and a screenshot does not always make the scale obvious, so say which one you use.
- Volume, unless the pane is in the shot. Credit where it is due: all 11 first answers in our test said so instead of guessing.
- Anything off the edge of the image: the higher timeframe, next week's earnings date, the price since you took the screenshot. As Insigtrade puts it, ChatGPT "has no live market data connection" when it reads a screenshot.
What a dedicated chart tool adds, and what it doesn't
We build one, so weigh this with that in mind. WolfGPT reads the same kind of screenshot, and adds three things a general chatbot only does if you set it up each time:
- It works out the symbol, timeframe and last price from the image, then, on the trial or a paid plan, pulls live data for that symbol from TradingView: price, day and 52-week range, RSI, MACD, ADX, the 50 and 200 EMAs, multi-timeframe ratings and recent headlines. Indicator readings come from data, not from squinting at pixels.
- The read comes back in the same shape every time: a bias, an entry zone, a stop, up to three targets with the risk to reward of each, support and resistance lists, and a main and an alternate scenario. Set your style to scalper, day or swing and the stop and targets adjust.
- The AI coach (trial or paid) sees a short summary of your last five reads, such as the market, timeframe and bias, so you can ask it about them.
What it doesn't change: it is still a model looking at an image, and it can be wrong in the same ways. If it can't identify the symbol, say on a cropped or made-up chart like ours, it falls back to reading the picture alone. That is also why we did not score WolfGPT here. Our charts have no live symbol to look up, which is the part that makes it different, and we won't quote an accuracy number we have not measured.
| ChatGPT or Gemini | WolfGPT | |
|---|---|---|
| Reads a chart screenshot | ✓ | ✓ |
| Looks up live data for the symbol it sees | ✕ | Trial or paid plan |
| Same structure on every read | If you ask each time | ✓ |
| Stop, targets and risk to reward in each read | If you ask | ✓ |
| Tuned to scalper, day or swing | If you tell it | ✓ |
| Follow-up questions | ✓ | Via the coach, trial or paid |
| Can still be wrong | Yes | Yes |
WolfGPT features checked in our own code, November 2026. Anyone can upload a chart; the full read needs the trial or a paid plan.
How to get a better chart read from ChatGPT
- 1
Send a big, clean screenshot
Full size, price axis visible, extra indicators hidden. Our small image cost the model most of its detail.
- 2
Show what you will ask about
If you want a view on volume, include the volume pane. If you care about the weekly trend, send the weekly chart too.
- 3
Say the market, timeframe and scale
"Daily chart, log scale, no volume" takes five seconds to type and removes three guesses. If candles are new to you, learn to read them first so you can check the answer.
- 4
Ask for the rule and the number
"Has any candle closed below 131.5?" is a question it can be checked on. "Is it confirmed?" is the one both models got wrong.
- 5
Ask again in a new chat
If the levels move between runs, they were estimates. Then size the trade yourself with the position size calculator and check the risk to reward.
This is a [daily] chart of a [stock / forex pair / crypto], [linear / log] scale, [with / without] a volume pane. 1. Trend: up, down or sideways? List the swing highs and lows you used. 2. Support and resistance: give prices from the axis and say which candles made them. 3. Pattern: name one only if it meets the standard rule. Give the exact level that confirms it, and tell me whether any candle has CLOSED beyond that level. 4. If something is too small to read or not on the chart, write "can't tell" instead of guessing. 5. For a trade idea: the price that proves it wrong first, then the stop, then targets with risk to reward.
Fill in the square brackets. This makes the answer easier to check. It does not make it right.
Questions people ask
Can ChatGPT predict stock prices?
No. It can describe what a chart shows and lay out scenarios, but it has no forecasting edge and no live data unless a tool feeds it some. Treat any target it gives as a level to check, not a forecast.
Can the free version of ChatGPT analyze charts?
OpenAI says image inputs work on Free and paid plans, with plan-specific usage limits (image inputs FAQ, checked 7 November 2026). Our test used the API rather than the app, so we can't say how the free app's model compares on charts.
Is ChatGPT good for day trading?
For a fast first look at structure, it is handy. Day trading is where its weak spots hurt most: the screenshot is stale within minutes, and small timeframes are where tiny details decide things. We would not place an intraday trade on its read alone.
Can ChatGPT read TradingView charts?
Yes, a TradingView screenshot is just an image to it. Keep the price scale showing, hide indicators you are not asking about and include the volume pane if volume matters. It can't log in to TradingView or watch your live chart.
Is ChatGPT or Gemini better at reading charts?
In our test Gemini read structure better and corrected itself when challenged, while ChatGPT's model read exact prices better. With one run per image that is a hint, not a ranking, and model versions change every few months.
Is it safe to trade on an AI chart analysis?
Trading is risky with or without AI, and no chart read tells you what price will do next. Use an AI read as a second opinion, check every level on the chart yourself, and size each trade so the loss at your stop is one you can take.