How to Screenshot a Trading Chart So AI Can Actually Read It
The quality of an AI chart read is capped by the quality of the screenshot. This is the least discussed and most fixable part of the whole process, and after looking at a great many submitted charts, the same handful of mistakes account for most of the bad output.
The short answer
Keep the price axis, the time axis, the symbol and the timeframe label inside the frame. Show 100 to 200 candles - enough for structure, not so many that individual candles become unreadable. Capture at full resolution rather than a photo of a screen, turn off drawings you do not want interpreted as structure, and never crop the right edge, because the most recent price action is the part that matters most.
Why this matters more than the model
An analysis engine can only describe what is in the picture. Everything you leave out does not exist, and the failure is silent - you do not get told "your price axis was cropped", you get a read that quietly guessed at the scale. That guess then propagates into every level it quotes.
The good news is that this is the one part of the process entirely under your control, and fixing it takes about ten seconds per chart.
The seven things that must be in frame
Capture checklist
1. The price axis. Without it there is no scale, and every level becomes a guess. This is the single most common omission.
2. The time axis. Establishes how much history you are showing and lets the read reference "early August" rather than "on the left".
3. The symbol. Crypto, forex and equities behave differently enough that knowing the instrument changes the read.
4. The timeframe label. A 5-minute and a daily chart of the same asset look similar and mean completely different things.
5. The right edge. Never crop the most recent candles. That is where the actionable structure is.
6. Indicator panels, with settings visible. An RSI panel is readable. An unlabelled oscillator is not - the read cannot tell a 14-period from a 7-period.
7. Enough bars for context. See below.
How many candles to show
Between 100 and 200 is the practical sweet spot, and the reasoning is symmetrical.
Too few and there is no structure to read. Thirty candles cannot establish a trend; it shows you a fragment, and a fragment of a pullback inside an uptrend looks exactly like the start of a downtrend. Reads from short frames are the ones most likely to be confidently backwards.
Too many and the candles compress into noise. At 800 bars on a phone screenshot, individual candles are two pixels wide, wicks disappear entirely, and candlestick pattern recognition becomes impossible. You also flatten recent action into irrelevance - a 3 percent move that matters to your position becomes a rounding error on a two-year chart.
If you genuinely need both, take two screenshots and analyse them separately. The higher timeframe establishes context, the lower one shows the setup. That is how a human technician does it too.
Capture method, in order of preference
| Method | Quality | Use when |
|---|---|---|
| Native screenshot on the device showing the chart | Best - pixel-exact, full resolution | Always, if possible |
| Platform export (TradingView "save chart image") | Excellent - clean, labelled, no UI clutter | Desktop platforms that offer it |
| Screenshot then crop | Good, if you crop carefully | Removing unrelated UI, never removing axes |
| Photo of a monitor with a phone | Poor - moire, glare, skew, soft text | Avoid. Axis labels frequently become unreadable |
The bottom row is worth stressing because it is common and the damage is invisible. A photographed screen introduces moire patterns across dense candle regions and softens small text exactly where precision matters - the axis labels. The read comes back looking normal, built partly on misread numbers.
Clean up the chart before you capture
Anything drawn on the chart is part of the image, and anything in the image may be interpreted as meaningful.
Old trendlines from a setup you abandoned last month, half-finished Fibonacci retracements, position markers, alert lines and stray annotations all get read as structure you intentionally marked. If you left a trendline sitting across a region that no longer matters, it will very reasonably be treated as a level you care about.
Delete what you are not using. Keep drawings only where you specifically want them considered - a level you have marked deliberately is useful signal, and worth keeping for exactly that reason.
Dark mode, light mode, and colour
Either theme reads fine. What matters is contrast between the candle bodies and the background, and between the two candle colours.
The one genuine problem is a custom colour scheme where up and down candles are similar in value - two mid-tone colours that differ in hue but not brightness. That is hard for a human to scan quickly and it is hard for a model too. If you have built a subtle low-contrast theme, it is worth switching to a default scheme before capturing.
The mistakes that cause the worst reads
Ranked by how wrong the output gets
Cropping the price axis. Produces confident, specific, wrong levels. Worst possible failure because the output looks completely normal.
Too few candles. Produces trend reads that are frequently the reverse of the real trend.
Photographing a screen. Misread axis numbers, lost wick detail.
Cropping the right edge. Analysis of a setup that has already resolved.
Leaving stale drawings on. Phantom levels treated as real.
Unlabelled indicator panels. Indicator read with assumed default settings that may not be yours.
Every one of these takes seconds to avoid, and between them they account for the large majority of reads that come back plausible-sounding and wrong.
Frequently asked questions
What is the best way to screenshot a chart on iPhone?
Use the native screenshot - side button and volume up together - with the chart in landscape orientation. Landscape gives you substantially more horizontal candles at readable width. Check before submitting that the price axis on the right edge is inside the capture, since some platforms let it sit partly off-screen in portrait.
How many candles should be visible in the screenshot?
Between 100 and 200 for most purposes. Fewer than about 50 and there is not enough history to establish trend, which is where reversed trend reads come from. More than about 400 on a phone-sized image and individual candles compress until wick detail and candlestick patterns are no longer readable.
Does it matter if my chart is in dark mode?
No. What matters is contrast - between candles and background, and between up and down candles. Both standard dark and light themes are fine. Custom low-contrast colour schemes where the two candle colours have similar brightness are the one setup genuinely worth changing before capture.
Should I remove my drawings and indicators first?
Remove stale drawings, keep deliberate ones. Anything on the chart may be read as structure you marked on purpose, so an abandoned trendline from last month becomes a phantom level. Indicators are worth keeping if their settings are labelled and visible, because then they can actually be read rather than assumed.
Related Articles
- AI Chart Analysis: What It Actually Does, and What It Can't
- Using ChatGPT for Chart Analysis: What Works and Where It Breaks
- How to Read a Stock Chart, in the Order You Should 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.