Turn Health Screenshots Into Pattern Reports
Upload a week's health app screenshots to Google Gemini 2.0 Flash on AI Studio and get back a structured day-by-day table with cross-metric pattern observations — free with a Google account, no credit card, 1,500 requests per day resetting at midnight Pacific Time.
Collect your screenshots
Open the health apps you already have — Apple Health, Google Fit, Oura, MyFitnessPal, a mood tracker, a sleep app, whatever you actually use. Take screenshots of the daily views for the past 7 days.
Use daily views, not monthly overviews — the per-day numbers are larger on screen and far easier for the model to read from an image. You don't need every metric; grab whatever you naturally check.
Open Google AI Studio
Go to aistudio.google.com and sign in with a Google account. No credit card — the free tier activates on sign-in. You'll see a prompt interface.
The default model should be Gemini 2.0 Flash, which is the free-tier model you want. If a different model is selected, switch to Gemini 2.0 Flash in the model dropdown at the top.
Upload your screenshots
Click the image upload icon in the prompt area (or drag-and-drop your files). Upload all 7–10 screenshots in a single request — they'll appear as thumbnail previews.
The model processes all of them in one go thanks to its multimodal input and 1M-token context window. You do not need to send one screenshot per request.
Paste the prompt template and run
Copy the template below into the text area alongside your uploaded images, then click Run. The model reads each screenshot, extracts the visible metrics, builds a day-by-day table, and lists cross-metric patterns it observes in your data.
Save what matters
Copy the output table into a note or spreadsheet. The real value is in the "patterns" and "unusual days" sections — that's the insight your individual apps can't surface because each one only sees its own slice of your data.
Gemini sees all your metrics at once, which is the entire point of this workflow.
What breaks
- Safety filters block medical-advice phrasing. "What patterns do you observe?" works. "Should I be worried about this heart rate?" gets blocked. This is the correct boundary for a non-medical tool — work within it. The prompt template is already phrased to stay on the safe side.
- Screenshot OCR is imperfect. Small text, graph lines, overlapping chart elements, and low-contrast numbers get misread. If a number in the output looks wrong, check the original screenshot. Don't make decisions based on a single extracted number you haven't verified.
- 1,500 requests/day resets at midnight Pacific Time. Not a rolling 24-hour window — a fixed clock reset. Hit the limit at 11:30 PM PT and you're back in 30 minutes. Hit it at 9 AM PT and you wait 15 hours.
- Your prompts and images go to Google. The free tier may use inputs to improve models. Check ai.google.dev/policies for the current data retention policy. Don't upload anything you wouldn't store in any cloud service.
- No memory between sessions. Each prompt starts fresh. If you want to compare this week's patterns to last week's, paste both summaries into a single new prompt — the 1M-token context window handles it easily.