Auto Detect Signature Fields in PDF With Vision AI (Simply Explained)

Signing Isn't the Slow Part. Getting Ready to Sign Is.

Signing a document takes 30 seconds. Click the box, draw your name, done.

The slow part comes before that. Someone has to prepare the document first. They drag a signature box onto the right line, a date box next to it, a name box below it. Then they do it all again for the second signer on a different page.

On a 12-page contract with two signers, that fiddly box-dragging eats 5 to 10 minutes. Every single time.

I hit this problem when I built my own e-signature tool to stop paying for DocuSign. The signing part was easy. The setup part was miserable. And paying a monthly subscription just to dodge that hassle felt like renting a solution to a problem I could solve myself.

So I gave myself a clear target. Read the document, find every blank a person needs to fill in, figure out which signer each blank belongs to, line the box up perfectly, and get it right more than 90% of the time. The last 10% would be a quick drag-fix that takes a couple seconds.

Most People Think AI Can't Do This Kind of Work

Here's the assumption baked into all of this. Most people, even smart operators who've already put plenty of their business on autopilot, assume AI can't handle precise layout work like this.

Boxes on lines. Telling Signer A's signature apart from Signer B's. It feels too exact for a tool that sometimes makes things up.

But it's very doable. The trick is using two ways of looking at the document instead of one, and letting plain old reliable code make the final decision, not the AI.

Let me walk you through how it works, without the technical stuff.

Two Sets of Eyes Are Better Than One

The first thing I do is read the words already in the document.

Most documents have hidden text inside them, and that text comes with location info. So when the document says "Signature:" or "Print Name," I know exactly where those words sit on the page. From there, I can put a signature box right next to the right label.

That works great for clean, well-organized contracts. It catches maybe 60 to 70% of the blanks for free.

But it has blind spots. It misses blanks that are just a bare line with no label next to them. It gets confused by complicated tables. Plenty of real legal templates have a row of underscores and zero text telling you what they're for.

So I add a second set of eyes. I turn each page into an image and hand it to an AI that can actually look at the page the way a person would.

I ask it one simple thing. Find every spot where a human is expected to write, and point at it. That's it. I'm not asking the AI to make big decisions. I'm asking it to spot the blanks.

This catches what the text reading misses. A bare underline with no label. A weird table layout. The AI doesn't need a nearby label, because it can see the line.

Now, the AI isn't perfect. Sometimes it points at the same blank twice. Sometimes it imagines a blank in empty space because the page just felt like it should have one. That's fine. It's not the final answer. It's just one of my two opinions.

Letting the Math Make the Final Call

Now I have two lists of blanks. One from reading the text, one from the AI looking at the page. They overlap constantly, because both keep pointing at the same signature line.

If I just combined them, every blank would show up twice. So I clean it up with simple geometry, not guesswork.

I look at how much two boxes overlap. If they're sitting basically on top of each other, they're the same blank, and I merge them into one. If they're close but not overlapping (the text found the label, the AI found the line just below it), I check how close their centers are. Close enough means same blank.

When both methods agree on a spot, I know it's almost certainly real. When only one of them found it, I keep it but flag it so a human can double-check.

None of this is the AI guessing. It's reliable math that runs the same way every single time. The AI's job is to see. The math's job is to be exactly right.

Putting Each Blank With the Right Signer

This is the hardest part, and it's where most simple versions fall apart.

A two-signer contract has a signature line for each person. Put the wrong signer on the wrong line and you didn't save time. You ruined the document. And you might not catch it until it's already signed wrong.

So I use a list of rules, tried in strict order, from most reliable to least.

First, does the signer's full name appear near the blank? If yes, easy, assign it. If not, does part of their name show up? Then I check first names only. Then I check if the blank sits near a company name that matches someone's email. (The company rep gets the company line.) Then I check if the blank is right next to one I've already assigned.

Only as a last resort do I fall back to simple order: first blank goes to the first signer.

That order-based guess is what most people would reach for first. I put it dead last. Order is just a guess, and it quietly mislabels everyone the moment a template lists signers in a different order than the lines. I only use it after every smarter clue has come up empty.

On standard legal templates, this gets the right signer on the right line almost every time. And when something slips through, it's obvious in a quick review, not buried.

The Detail That Makes It Feel Finished

Here's the small thing that decides whether the whole tool feels polished or broken.

Even with the right blank and the right signer, a signature box floating a few pixels above the actual line looks wrong. People don't trust it. It reads as sloppy.

So I snap the box right onto the line. I find the underscores that make up the line you sign on, measure them, and place the box exactly where a pen would land.

There's some tricky math behind that (documents and screens measure positions in opposite directions, which trips up almost everyone who works with PDFs). Once I handled it, the box lands correctly on more than 90% of templates. The rest take a two-second drag to fix. Exactly the target I set at the start.

Why This Matters More Than One Feature

Step back from the PDFs for a second.

The real point was never that I dodged a DocuSign bill, though I did. The point is this is exactly the kind of fiddly, precise work people assume AI can't touch.

It can. The trick was never asking one AI to do everything. I used two sets of eyes that cover each other's blind spots. I let reliable math make the final call. And I refused to let a confident guess quietly mislabel anyone.

If you've got a manual document process eating hours every week (contracts, intake forms, onboarding packets, anything where someone drags boxes or retypes the same fields), this is buildable today. Not someday. Now.

Want to explore what AI could do for your business?

Book a free 30-minute strategy call. No pitch deck, no sales team, just a real conversation about your operations and where AI fits.

Book a Discovery Call