On this page (5 sections)
Upload a 1h chart screenshot to ChatGPT and ask what it sees, and the reply will usually name a trend and a support level and say where the idea fails, with prices to two decimals. Part of that comes straight off the image and part is estimated between the labels on the price axis, and the reply reads the same either way. So it can analyze a chart in the sense of describing one, and how much its numbers are worth depends on checks you run yourself.
What it can read from a screenshot
Given a clean screenshot with the timeframe label and the price axis visible, a general chatbot can describe what is on it, such as whether the visible swings step higher or lower, or a long wick that ran under a low and closed back above it. It can put trading names to those shapes, such as a liquidity sweep or a fair value gap, and explain the idea behind each in plain words as many times as you ask.
It also works as a reviewer for your own plan. Paste the plan next to the screenshot and ask what you haven’t defined yet, such as which timeframe’s close counts as the exit or what you do if price never reaches the entry.
Where a screenshot read goes wrong
- No clock and no live price. A screenshot is a still image. From it alone the model can’t tell when it was taken or where price is now. If the timeframe label is cropped, it has to guess which timeframe it is reading.
- Prices estimated from the axis. The price scale carries a handful of labels. A level between two of them is estimated from where it sits on the screen, and small fonts or a squeezed chart make that estimate worse. The reply still gives it to two decimals, so an estimate looks the same as a price taken from a candle.
- Nothing outside the frame. The swing that decides the trend may sit left of the visible window, and zooming in or out changes which swings the model gets to see. The higher timeframe isn’t on the image at all (more on pairing timeframes in the multi-timeframe guide).
- Confident numbers with nothing under them. Ask a chat model for a support level and it will name one, whether or not a candle marks that price, in the same tone it uses for a level that does exist.
- No fixed rules. Nothing in a general chatbot pins down how many candles confirm a swing or whether a wick counts as a break. Ask twice, or word the question differently, and the low it treats as the trend’s line can move. You can write your rules into the prompt, and it is still worth checking that it applied them.
How to check any AI chart read
These checks work the same whether the read came from a general chatbot or from a dedicated chart tool.
- Pin down the chart. Put the symbol, the exchange, the timeframe and the time of the screenshot in your message, and check that the reply names the timeframe you meant. Before you act, count the candles that have closed since the image was taken.
- Ask for the invalidation as a close. Ask “Which candle close, on which timeframe, would prove this wrong?” A usable answer names one price and one timeframe. An answer like “if support breaks” doesn’t tell you which close to act on.
- Find that price on the live chart. Put the crosshair on it. For a trend idea it should be the last confirmed swing on the other side of price. A swing low counts as confirmed once five candles on each side of it have higher lows (a swing high mirrors that), and a trend flips only on a close beyond the last confirmed swing (see BOS and CHoCH). If no candle has its high or low at that price, the number came from the axis.
- Check the other numbers the same way. Entry zones and targets should sit on candles you can point to.
- Build the stop and size yourself. The invalidation level is the close that proves the idea wrong, and the stop goes past it by a buffer at least as wide as recent wicks ran past their levels. 1R is the distance from entry to stop, and the size comes from the money you accept to lose at the stop, fees included.
A worked example: a SOL 1h screenshot
The numbers are invented. You take a screenshot of SOL on the 1h during the 13:00 UTC candle. The window starts at the 06:00 candle and the price axis is labelled every 2.00. A chatbot reads it as a 1h uptrend and says the bullish case holds while price stays above 151.00. Asked which 1h close would prove it wrong, it answers a close below 151.00.
Timestamp. You open the live chart at 16:05 UTC. Three 1h candles have closed since the screenshot (13:00, 14:00 and 15:00), and price is at 153.00.
The level. No candle has a low at 151.00. It sits between the 150 and 152 axis labels, about where the 12:00 candle’s low of 151.20 appears on the image. That low has three closed candles after it, two short of the five that confirm a swing. The last confirmed swing low is 149.60, from the 04:00 candle, two candles before the screenshot’s first one. A 1h close below 149.60 is what flips this uptrend, so that close is the invalidation. A stop tucked under the chatbot’s 151.00 would sit just below the unconfirmed 151.20 low, where stops cluster and a sweep can fill them while the 1h trend stays up.
Stop and size. Recent 1h wicks ran up to 0.40 past their levels, so the stop goes 0.50 under the line, at 149.10. With a long at 153.00 that is 3.90 away, and 3.90 is 1R. Say you accept losing 150 USDT at the stop and pay a 0.05 percent taker fee on each side. Each coin then loses 3.90 in price and about 0.15 in fees, about 4.05 in all, so the size is 150 / 4.05, about 37.0 SOL. The position size calculator does the same sum. On isolated margin, keep the leverage low enough that the liquidation price sits well below 149.10, which the liquidation price calculator estimates.
How Finora handles it
Finora, an AI trading assistant on Telegram, takes a symbol and timeframe you type, such as SOL 1h, and fetches the chart itself, so the read starts from the chart at the moment you ask. Its structure comes from a fixed, rule-based chart table, Finora Dow Trend v2.1, which confirms a swing with five candles on each side and flips the trend only on a close beyond the last confirmed swing, and each read names its invalidation level (more in how Finora reads a chart). The read is still an interpretation that can be wrong and is meant as education, and where to put the stop and how much to risk stays your decision.
Related reading
- AI crypto chart analysis: how Finora reads a chartThe steps behind every Finora read: structure first, then liquidity, fair value gaps, key levels and a plan with an invalidation level.
- What is an invalidation level? Decide where you’re wrong firstAn invalidation level is the price that proves your trade idea wrong. How to find it, how your stop and size follow from it, and a worked 1h example.
- Break of structure vs change of character, worked through on a 4h chartBOS: a close past the last confirmed swing with the trend. CHoCH: the first close past it against the trend. Worked through on a 4h ETH example.
- Crypto market structure screenerLive market structure for the top 100 Binance perpetuals: 1h and 4h trend, last swing high and low, the close that flips each trend, RSI and funding.