Drop a PDF. We find the lines.
A heuristic + anchor-text scanner reads the underscore lines, the labels above and beside them — Signature, Signataire, Date, Daté, Nom — and snaps the right field type onto each one. Bilingual EN + FR pass, runs in the browser, zero manual placement.
Scanned anchors, snapped fields, zero manual placement.
Three layers cooperate — a vector pass for digital PDFs, an anchor-text reader for labels in EN + FR, and a snap engine that aligns to underscore lines.
Heuristic + anchor-text scanner.
Reads underscore lines, dotted leaders, and bracket placeholders. Cross-references labels above + beside each line so the field type is right the first time.
EN + FR pass.
Catches Signature, Signataire, Date, Daté, Nom, Paraphes in the same pass. No language toggle, no separate model — bilingual out of the box.
Snap to underscore lines.
Pixel-precise — the bottom edge of every signature field lands exactly on the underscore. No floating boxes, no inch-of-padding amateur look.
12 fields in under a second, zero misalignment.
Drop in, fields appear.
Drop a PDF into /app/new — the scanner sweeps it in 0.8s on a typical laptop, faster on Apple Silicon. The detected fields land color-coded by type (Signature is wax, Date is fountain, Name is green). Review, accept, send.
Every line, perfectly aligned.
No floating boxes, no fields drifting off the underscore. The signer sees a doc that looks like the sender hand-placed every field — because they didn't, but the snap engine did better.
Review every detection
A side rail lists every field the scanner placed — type, page, anchor it matched. One-click accept or reject per row. Reject one and the underlying line stays empty for you to ignore or hand-place.
Color-coded by type.
Signature fields are wax, Date is fountain, Name is green, Initials is gold. You can see at a glance whether the scanner mistook a date for a name (rare, but it happens on tight layouts).
Bilingual anchors.
The scanner indexes EN + FR labels in the same pass. A bilingual Quebec contract gets perfect placement even when half the page is in French and half is in English — no language flag, no separate workflow.
Manual review is one click.
Click "Accept all" if it looks right, or tap the X on the misfires. Average doc — 12 fields, 11 correct, 30 seconds of review. Five seconds if you trust the scanner on a familiar template.
No fields off the line
Every signature box sits on its underscore. Every date field aligns to its DD/MM/YYYY skeleton. The signer never has to wonder if a misaligned field means the doc is broken — it's not, because the scanner placed it.
Mobile-precise targets.
The snap is pixel-perfect on phones too — a 7px-tall signature line on a 6-inch screen still gets a tap-target the right size, in the right place.
Bilingual labels respected.
Sign next to Signataire on the French page, Signature on the English page. The signer's flow walks them through in document order — page 1 EN, page 2 FR, no language switching.
No "find the box" anxiety.
Older signers and rushed signers don't have to hunt for fields that drifted off the line. Every box lands where the eye expects — under the label, on the line, sized to the line.
Where AI Field Detection shines.
Anywhere a PDF arrives that someone else laid out — government forms, boilerplate contracts, scanned legacy templates.
Inbound government forms
CRA, Service Canada, RAMQ, provincial registrar forms — pre-laid-out PDFs with bilingual labels. The scanner places fields without you ever leaving the canvas.
Standard NDAs
Mutual NDAs in any of the common templates floating around the internet. Drop it in, hit send. The first one's free.
Boilerplate contracts
MSAs, SOWs, retainers you receive from a counterparty's legal team. Their layout, your signing flow — they never know you didn't hand-place the fields.
Real estate disclosures
OREA, BCFSA, GTAM forms — multi-page property disclosure packages with 30+ fields. The scanner places them all in under a second; you spot-check the rare miss.
Healthcare consents
PIPEDA + PHIPA-aligned intake and consent forms with bilingual labels. The scanner respects French anchors so your Quebec patients get the same flow as Ontario.
Cross-border bilingual deals
A contract drafted half in English and half in French — common in Quebec commercial deals. One scan handles both, no manual placement on either half.
Questions, answered.
How accurate is the field detection?
On clean, digital PDFs with underscore lines or bracket placeholders + standard EN/FR labels: ~96% precision, ~93% recall across our test corpus of 800 Canadian commercial + government forms. On scanned/image PDFs the numbers drop — see the next question. We tell you up-front when accuracy is likely to suffer with a paper-icon warning at upload.
Does it work on scanned (image) PDFs?
Partially. Image-only PDFs lose the vector-text pass, so accuracy drops to ~70% — the scanner falls back to pure edge-detection of underscore lines without anchor labels to disambiguate field type. You'll spend more time in manual review. We recommend running scanned PDFs through OCR first (we don't bundle one — Adobe Acrobat or your scanner's built-in OCR works fine).
Does it support languages beyond EN and FR?
Not yet. EN + FR is the v1 scope because they're the two official Canadian languages and the bilingual pass is the headline differentiator vs the US incumbents. Spanish + Portuguese are on the v2 list for late 2026 (driven by cross-border LATAM customer requests). Tell us what you need at hello@inksign.ca — we prioritize by request volume.
What does the scanner typically miss?
Three things: (1) handwritten or freeform signature lines without underscores or labels, (2) signature blocks inside complex tables where row borders confuse the line detector, (3) non-standard label phrasings like 'X-marks-the-spot' or icon-only fields. When a miss happens, you hand-place the field — the canvas still has the snap-to-line magnet so even manual placement stays pixel-perfect.
What's the correction workflow when I disagree?
Side rail lists every detected field. Tap the X to reject a placement (the line stays empty for you to ignore or hand-place). Tap a field on the canvas to change its type with a dropdown (Signature ↔ Date ↔ Name ↔ Initials). Drag to reposition with magnetic snap. No re-scan needed — your edits commit instantly.
What model is this, and where does my PDF get sent?
It's not a transformer model — it's a heuristic scanner built on pdf.js + a tuned line-detection + anchor-matching pipeline. The entire pass runs in your browser. Your PDF never leaves the device for the detection step; bytes only travel to Supabase ca-central-1 when you save the draft. PIPEDA + Quebec Law 25 compliant by construction, not by promise.
Is the bilingual pass Quebec-compliance ready?
Yes. The detection layer is language-agnostic at the snap level — the underscore line is the underscore line, regardless of the label above it. The anchor lexicon was built with Loi 25 in mind: French labels (Signataire, Daté, Nom, Paraphes, Témoin) are first-class citizens, not an afterthought. Pair with Witness Flow for in-person + witnessed signings that satisfy Quebec civil-law requirements.
You won't find this anywhere else?
Correct. DocuSign's 'AssistAI' is a paid Enterprise add-on and English-only. Adobe Sign's auto-detect is English-only and requires a cloud round-trip. Most lower-tier platforms don't ship in-browser detection at all. Ink. AI Field Detection runs in your browser, ships with Team tier (no add-on), and the French pass is built-in — not an afterthought. It's the headline reason Canadian Team customers pick Ink. over the US incumbents.
Still have questions?
Email us →Stop placing fields by hand on contracts someone else laid out.
AI Field Detection ships with Team. Try it free — no card, no trial timer.