AI extraction · margin of error included

From pixels
to numbers.

Drop any chart image. Plotion places the axes, traces every series, and ships the dataset with a computed margin of error. Built for researchers who cite their sources.

Join the waitlist
Avg. error
±2.4%
Chart families
7
Datasets
486K+
199020002010202020300125250375500reading image…
0K
Datasets digitized
0K+
Researchers active
±0.0%
Mean extraction error
3 steps · 30 seconds

Drop. Describe. Done.

The whole loop, from a screenshot in someone else's paper to a CSV in your hand. Adjust anything mid-flight with a drag.

/01

Drop a chart image

PNG, JPG, SVG, GIF, WebP, BMP. Up to 10 MB. Crop right in the browser before any pixels leave your machine.

figure-3.png
/02

Describe what you see

A short sentence is enough. "x is years, y is in thousands, log scale on y." Plotion uses it to seed the axis calibration and the AI starting points.

"y is thousands of people"
/03

Verify, export, cite

Every point is draggable. Margin-of-error follows you in real-time. CSV, XLSX, JSON, Python, MATLAB. The σ column comes along.

7 chart families · 4 axis scales

If it's a chart, we read it.

Linear, logarithmic, date/time, category. Pick the scale per axis and Plotion calibrates accordingly. Multiple series on one chart? Mixed scales on X and Y? Yes and yes. XY, Bar, Pie, Polar, Ternary, Geo and Scale-bar are all live today.

XY · scatter · line
lin · log · date · cat
Bar
h · v · stacked
Polar
r · θ
Ternary
3-component
Pie · donut
percent
Geo
lat · lon
Scale-bar
ruler
computed live · per-extraction

An honest number,
not vibes.

Plotion computes ±% from axis-tick alignment, curve variance, and your image's effective resolution. Reviewers stop asking "how accurate is this?" because the answer is right there.

↓ image resolution slider
±3.1%
72 dpi150 dpi300 dpi600 dpi

Drag the slider.

Higher resolution images give Plotion more pixels to lock onto. Watch the margin shrink as the source quality goes up. The number you ship in your paper is the same number you see here.

  • → axis-tick alignment
  • → curve variance σ
  • → effective DPI of source
  • → AI-vs-manual disagreement δ
export · 6 formats · σ included

Ship it your way.

One click to CSV, XLSX, JSON, Python list, MATLAB matrix, or a clipboard-ready paste for Google Sheets. Every format ships with the σ (margin) column attached.

# figure-3-population.csv
year,population_thousands,sigma
1990,0,0.014
2000,114,0.015
2010,232,0.018
2020,392,0.020
2030,534,0.021
3 tiers, cancel anytime

Honest pricing.

Billed through Lemon Squeezy. Your datasets and exports stay yours.

Free
$0/forever
  • 10 extractions / month
  • Manual digitizer, all plot types
  • CSV + JSON export
  • Community support
Join the waitlist
Researcher
$12/month
  • Unlimited AI extractions
  • Our polished AI model, fully optimized
  • Margin of error on every dataset
  • All export formats incl. MATLAB
  • Priority email support
Join the waitlist
Lab
$48/month
  • Everything in Researcher
  • 5 seats, shared workspace
  • SSO, audit log, DOI metadata
  • Dedicated Slack channel
Join the waitlist
launching soon

Pricing opens with early access.

Join the waitlist, early respondents lock in a launch discount before pricing goes live.

Join the waitlist
questions · answers

FAQ.

Honest answers. For details not covered here, email support@plotion.app.

Across our benchmark of 400 scientific figures, AI-assisted extraction lands within ±2.4% of hand-digitized values on average. The ±% Plotion shows on each dataset is computed live from your specific image and axes, not a static average.