AI trade import lets you snap a photo of your broker's positions screen and get structured trades back in seconds, no typing required. But not every screenshot parses cleanly, and blindly saving what the AI returns is how bad cost basis gets baked into your portfolio. This guide breaks down what Claude reads well, where it fails, and the exact checks that separate a clean import from a corrupted one.
What is AI trade import and how does it actually work?
AI trade import is a feature that uses a vision-capable language model, like Claude, to read a screenshot of your broker account and convert the visible positions or order history into structured trade records. Instead of typing every ticker, quantity, and price by hand, you upload an image and the model extracts the data into fields your tracker can save.
The process runs in three stages. First the model performs optical parsing, reading the text and layout of the image. Then it maps that raw text to a trade schema, matching columns to fields like symbol, quantity, entry price, and date. Finally it returns structured output your app validates before saving.
- Vision parsing: Claude reads pixels into text, including tables, colored gains, and small fonts.
- Field mapping: it decides which number is quantity versus market value versus cost basis.
- Normalization: tickers get standardized (e.g. AAPL, BTC-USD, EMAAR.AE) and dates get formatted.
This is different from a direct broker API connection. If you want fully automated syncing instead of image uploads, read our guide on how to connect your brokerage account to a portfolio tracker. Screenshot import is the fallback when your broker has no API or you only have a photo from a friend's screen.
What broker screenshots does Claude parse reliably?
Claude parses clean, high-contrast screenshots of positions tables with near-perfect accuracy. The best results come from images where each row is one holding and columns are clearly labeled.
Screenshots that import cleanly
- Positions tabs from Schwab, Interactive Brokers, and Fidelity web dashboards with visible headers.
- Order history from Alpaca showing filled quantity, average fill price, and timestamp.
- Crypto holdings from Binance or Coinbase where the pair (e.g. ETH-USD) sits next to the amount.
- UAE market screens from ADX or DFM apps listing EMAAR.AE or FAB.AE with share count.
In practice, a full-desktop screenshot of a positions table with 10 to 15 rows and labeled columns will parse with over 95% field accuracy. The model handles thousands separators, currency symbols, and even red or green profit coloring without confusion.
What Claude adds automatically
Claude also infers missing context that a raw CSV would not. It recognizes that a Binance screenshot uses USDT quotes, that a DFM screen prices in AED, and that a US broker defaults to USD. If you are tracking both asset classes together, our walkthrough on tracking crypto and stocks in one portfolio covers how currencies and pairs get normalized.
Where does broker screenshot parsing fail?
Screenshot parsing fails most often when the image is ambiguous, cropped, or shows values that the model cannot distinguish from each other. These are the failure modes to watch for.
- Market value vs cost basis confusion: if a table shows both, the model can grab current value instead of your entry price, corrupting your average cost.
- Cropped columns: a screenshot cut off at the edge loses the quantity or date column entirely.
- Aggregated positions: a single row showing a blended average across multiple buys hides the individual lots.
- Low resolution or glare: a phone photo of a monitor with reflection turns 8 into 0 or drops decimal points.
- Ambiguous tickers: a bare symbol like ABC could match a US stock or a foreign listing without an exchange suffix.
The most dangerous failure: silent errors
The worst failures are the ones that look correct. A misread quantity of 150 instead of 1,500 produces a plausible trade that quietly understates your position by 90%. Because the output is well-formatted, it passes a casual glance. This is exactly why verification before saving is non-negotiable.
How do you verify AI-extracted trades before saving?
Verify AI-extracted trades by cross-checking three anchor numbers against the original screenshot: the symbol, the quantity, and the entry price. If those three match for every row, the rest of the record is almost always correct.
Run this checklist every time before you commit an import:
- Count the rows. The number of extracted trades should equal the number of holdings in the image. A missing row means a cropped or skipped line.
- Spot-check the largest position. Your biggest holding by value is where errors hurt most, so confirm its quantity and price first.
- Validate the price magnitude. If Claude returns $4.20 for a Schwab S&P 500 ETF trading near $580, that is a decimal misread.
- Confirm the currency and exchange suffix. A UAE holding should read EMAAR.AE in AED, not a bare EMAAR in USD.
- Check the date. Missing dates default to today, which distorts your holding period and any realized gain math.
PortfolioTrackr handles this by showing the parsed trades in an editable review table side by side before anything saves. You confirm or correct each field, and nothing touches your portfolio until you approve it. Our overview of logging trades with AI from broker screenshots walks through that review step in detail.
Screenshot import vs manual entry vs API sync: which should you use?
Screenshot import is the middle ground between slow manual entry and fully automated API sync. Each method fits a different situation, and most active investors end up using all three.
| Method | Speed | Accuracy risk | Best for |
|---|---|---|---|
| Manual entry | Slow | Typos, fatigue errors | One or two trades |
| AI screenshot import | Fast | Misread digits, needs review | Brokers with no API |
| Broker API sync | Instant | Very low, auto-updates | Supported brokers |
The reason so many retail investors reach for screenshot import is fragmentation. If you hold US stocks at Schwab, crypto at Binance, and UAE equities at an ADX broker with no API, screenshots are the only way to unify them quickly. For a broader look at how these approaches stack up against a manual sheet, see our comparison of a portfolio tracker versus a spreadsheet.
How do you take a screenshot that imports cleanly?
Take a screenshot that imports cleanly by capturing the full positions table at native resolution with all column headers visible. The quality of your input image is the single biggest driver of extraction accuracy.
Do this
- Screenshot the app directly instead of photographing a monitor. Native captures have no glare or angle distortion.
- Include the header row so the model knows which column is quantity versus value.
- Scroll and capture in sections if you have 30+ holdings, then import each image.
- Use the widest layout your broker offers so no columns are truncated.
Avoid this
- Cropping out the ticker or price column to fit a phone screen.
- Uploading a compressed, blurry chat forward where digits blur together.
- Screenshotting a chart or summary card that shows total value but not per-lot detail.
Why does verification matter more than extraction accuracy?
Verification matters more than raw extraction accuracy because a corrupted cost basis compounds silently through every future calculation. A single wrong entry price distorts your average cost, your unrealized gain, and eventually your tax reporting when you sell.
Consider the chain of effects from one bad number:
- A misread entry price throws off your average cost per share.
- That wrong average feeds a false unrealized profit figure on your dashboard.
- When you sell, the tracker calculates the wrong realized gain.
- Your portfolio allocation percentages shift, which can trigger misguided rebalancing.
AI is a drafting tool, not an auditor. The model's job is to save you the typing, and your job is the final sign-off. This same principle applies across every AI feature in a tracker, which we cover in our roundup of what AI can do for investors in 2026. Extraction gets you 90% of the way in seconds; verification protects the 10% that actually matters.
The bottom line
Claude can read a clean broker screenshot into structured trades with over 95% accuracy, turning a tedious data-entry chore into a 30-second upload. The technology genuinely works for positions tables, order histories, and multi-currency holdings across US, crypto, and UAE markets.
The catch is that the best AI import is the one you check before saving. Confirm symbol, quantity, and entry price on every row, especially your largest positions, and never let an unreviewed extraction touch your portfolio. Do that, and screenshot import becomes the fastest reliable way to keep a fragmented, multi-broker portfolio accurate.
Track your portfolio in real time: free for 3 days
Live P&L across stocks, crypto, and global markets. WhatsApp and Telegram price alerts. AI trade import. Unified dividend tracking. No brokerage connection required.
Start Free Trial
Frequently asked questions
Can Claude read a broker screenshot into trades accurately?
Yes, Claude reads clean, high-contrast positions screenshots with over 95% field accuracy. It reliably extracts ticker, quantity, and price from labeled tables. Accuracy drops with blurry phone photos, cropped columns, or aggregated positions, which is why you should always review the extracted trades before saving them.
What broker screenshots fail to import correctly with AI?
Low-resolution photos of monitors, cropped columns missing quantity or date, and rows that show market value instead of cost basis fail most often. Aggregated positions that blend multiple buys into one average also hide individual lot detail. Always capture the full labeled table at native resolution.
How do I verify AI-extracted trades before saving them?
Cross-check three anchor numbers against the original screenshot: symbol, quantity, and entry price. Count the rows to confirm none were skipped, spot-check your largest position, and validate that price magnitude and currency look right. PortfolioTrackr shows an editable review table so you approve each field before it saves.
Is AI trade import safe for my cost basis data?
AI trade import is safe when you review the output, because a misread entry price corrupts your average cost and every downstream gain calculation. The model drafts the trades but you sign off. PortfolioTrackr never saves an extracted trade until you confirm or correct it in the review step.
Should I use screenshot import or connect my broker API?
Use broker API sync when your broker supports it, since it updates automatically with near-zero error risk. Use AI screenshot import for brokers without an API, such as some ADX, DFM, or regional platforms. Most investors combine both to unify a fragmented multi-broker portfolio in one place.