Built for research teams

Extract structured data from papers with AI.

Show PDFSifter’s AI three examples of what you need. It reads every PDF in the folder, pulls out those values in minutes, and shows you the exact quote behind each one.

Start free See how it works

Free plan: 200 pages a month · No credit card needed

Trial facts · 48 papers · 47 done

Page 3

Field	Value	Confidence
Sample size	150150	98
Design	Double-blind RCTDouble-blind RCT	95
Follow-up	24 weeks24 weeks	91
Dropout rate	12.4%12.4%	72

Sample size · page 3

“A total of 150 participants were randomised, 75 to each arm.”

1 value is waiting for reviewReview →

Every value cites its source
Page, quote and a 0–100 confidence
Scanned pages included
OCR runs automatically
CSV, Excel, JSON or PDF
Exports your team can use
Files stored in the EU
Deleted on your schedule

How it works

From a folder of PDFs to a table you can trust

No prompts to write and no rules to maintain. You describe the data once, check a few examples, and the AI does the rest.

1
Describe what you need
Write a short brief in plain English and list your fields: text, numbers, percents or dates.
Sample size · numberFollow-up · text
2
Train the AI on three examples
Add PDFs with the answers you expect. “Label in PDFSifter” drafts them for you to check. The AI writes its own instructions and shows how often it matched you.
Measured on your examples
3
Run, review, export
Drop in a folder or a zip. Progress updates live, uncertain values are queued for review, and results export in one click.
CSVExcelJSONPDF report

See every step →

Features

AI extraction you can check

Every value shows where it came from. Checking what the AI found shouldn't mean re-reading every paper: PDFSifter tells you which values to trust and which ones to look at.

Quote, page and confidence
Each value links to the sentence it came from and the page it’s on, with a confidence score from 0 to 100.
Review only what’s uncertain
Values below your threshold are flagged. Step through them with J and K, edit with E, accept with Enter.
Scans and charts, too
Pages without a text layer, like scans and photos of charts, are read with OCR. Nothing to switch on.
Retry only what failed
If a few documents fail, run just those again in the same job. You only pay for their pages.
Gets better as you review
Turn a reviewed document into a training example in one click. Retrain, then compare every version in History.
Exports that keep their types
Excel keeps numbers, percents and dates typed and highlights values still waiting for review.

For teams and developers

Fits the way your lab already works

Invite your team, give everyone the right role, and plug PDFSifter into your own pipeline when you’re ready.

Roles and permissions for every workspace
An audit log of who changed what, and when
Search names, values and the full text of every PDF
Resumable uploads for big folders and zips
API keys and signed webhooks on Lab and Research

# 1. Upload PDFs with any tus client
tus upload papers/*.pdf  →  /files/

# 2. Start a job with a trained template
POST /api/v1/workspace/jobs
Authorization: Bearer sk_live_…

# 3. Hear back when it's done
←  job.finished
   signed with HMAC-SHA256

# 4. Read the records, or export them
CSV · Excel · JSON · PDF report

# POST to your endpoint, signed with HMAC-SHA256
{
  "type": "job.finished",
  "data": {
    "job": {
      "name": "March batch",
      "status": "completed",
      "documents_done": 48,
      "documents_failed": 0,
      "records_count": 48,
      "review_count": 3,
      "records_api": "/api/v1/workspace/jobs/…/records"
    }
  }
}

Security and privacy

Your papers stay yours

Unpublished manuscripts, patient reports, internal studies: PDFSifter is built to handle documents you can’t afford to leak.

Files stored in the EU
Uploaded PDFs live in EU-jurisdiction storage.
Zero data retention for AI
Model requests go only to providers that don’t keep your data.
Deleted on your schedule
Source PDFs are removed after your retention period. Results and search keep working.
Two-step verification
Authenticator-app codes and one-time recovery codes for every account.
Sign in the way you already do
Google, Microsoft or ORCID, next to email and password.
A 30-day safety net
Deleted jobs and templates wait in the Trash for 30 days before they’re gone.

Read how we handle your data →

Pricing

Start free. Pay when it’s part of your week.

Priced by the page, so a 3-page abstract and a 40-page report cost what they should.

Free

Try PDFSifter on a handful of papers.

$0forever

No credit card needed

Start free

200 pages a month
1 seat
2 templates
Train on PDFs of up to 20 pages
1 GB of stored PDFs
Source PDFs kept 30 days

Starter

For one researcher working through a review.

$19per month

or $180 a year

Get Starter

2,000 pages a month
2 seats
5 templates
Train on PDFs of up to 50 pages
5 GB of stored PDFs
Source PDFs kept 60 days
Extra pages with credits, $0.012 each

Lab

For a small team extracting every week.

$79per month

or $756 a year

Get Lab Request a trial

10,000 pages a month
5 seats
Unlimited templates
Train on PDFs of up to 120 pages
API access and webhooks
20 GB of stored PDFs
Source PDFs kept 90 days
Extra pages with credits, $0.009 each

Research

For labs running extraction at scale.

$300per month

or $2,880 a year

Get Research Request a trial

50,000 pages a month
10 seats
Unlimited templates
Train on PDFs of up to 300 pages
API access and webhooks
50 GB of stored PDFs
Source PDFs kept 90 days
Extra pages with credits, $0.004 each

Prices in US dollars. A busy month? Top up with credits instead of upgrading. Need more seats or an invoice? Talk to us.

FAQ

Questions researchers ask us

Something else? The in-app Help page has step-by-step guides, or you can write to us.

Do I need to write prompts?: No. You write a short brief, list the fields and add a few examples. PDFSifter’s AI writes the extraction instructions itself and tests them against your examples. You can still edit its summary and rules if you want to.
How accurate is it?: Every template shows its measured accuracy on your own examples. Every value has a confidence score, and anything below your review threshold (80 by default) is flagged for a person to check.
What kinds of PDFs work?: Papers, reports and forms, with one record per PDF or many. Scanned pages are read with OCR. Password-protected files need to be unlocked first.
What happens to my files?: They’re stored in the EU and deleted after your retention period: 30 days on Free, up to 90 on Lab and Research. The values, quotes and text stay, so your results and search keep working.
What counts as a page?: Each page of each PDF you run. If a document fails and you retry it, only that document’s pages are counted again.

Turn your next stack of papers into a table

Your first 200 pages every month are free. Set up a template in about 15 minutes.

Start free Sign in

Extract structured data from papers with AI.

From a folder of PDFs to a table you can trust

Describe what you need

Train the AI on three examples

Run, review, export

AI extraction you can check

Quote, page and confidence

Review only what’s uncertain

Scans and charts, too

Retry only what failed

Gets better as you review

Exports that keep their types

Fits the way your lab already works

Your papers stay yours

Files stored in the EU

Zero data retention for AI

Deleted on your schedule

Two-step verification

Sign in the way you already do