Open a drawing PDF from your CAD system, press Ctrl+F, and search for a
dimension you can plainly see on the screen. Nothing. The text is right there, crisp at
any zoom, and the PDF swears it doesn't exist.
This surprises people, and it has a straightforward cause: by the time the PDF was written, those characters had stopped being characters.
Three kinds of PDF that look identical
| Type | What's inside | Ctrl+F |
|---|---|---|
| Text | Real character codes with an embedded font | Works |
| Vector line art | Each character drawn as outlines and strokes — geometry, not letters | Finds nothing |
| Raster | One flat image of the whole sheet, usually from a scan | Finds nothing |
All three look sharp on screen. Only the first knows what it says.
Telling them apart in ten seconds
- Try to select a dimension with the text cursor. If you can highlight it, it's text.
- Zoom in hard — 800% or more. Vector line art stays perfectly crisp. Raster goes soft and blocky.
- So: crisp but unselectable means vector; blurry and unselectable means raster.
Why CAD does this
It isn't a bug, and it usually isn't carelessness. Converting text to geometry solves real problems for a drawing office:
- The drawing looks the same everywhere. No font substitution on a machine that doesn't have your CAD fonts — and a substituted font can change line lengths, overlap a callout, or push text out of a title block.
- GD&T symbols survive. Position, flatness, the modifiers — many of these live in specialised symbol fonts. Turned into outlines, they can't be mangled by a viewer that lacks the font.
- It's often the default in the plot or export configuration, set years ago by someone solving exactly the font problem above.
A drawing is a legal document. Guaranteeing that it renders identically on every machine is a defensible reason to give up searchable text, and plenty of aerospace drawing offices made that trade deliberately.
Scanned drawings are a different problem with the same symptom. A drawing that has been printed, signed, and scanned back in is a raster image — and often a skewed, speckled one. Everything below about extraction applies, but expect worse results and more verification.
What it means for reading dimensions automatically
If there's no text layer, no tool can simply read the drawing. The only route left is optical character recognition — treating the sheet as a picture and inferring characters from shapes. That has consequences worth being honest about:
- Numerals and plain text read well. Vector-rendered text is
unusually clean input for OCR — no scan noise, no skew — so values like
3.982orØ3.625 ±0.001come through reliably. - GD&T symbols do not. Standard OCR has no glyphs for position, flatness, profile, or the material-condition modifiers. It will either skip them or guess something wrong. Feature control frames need to be entered by hand or built with a proper GD&T editor.
- Ambiguity is real. A worn
8and aB, a0and anO, a decimal point and a speck of toner. On a compliance record, a plausible-looking wrong value is worse than no value.
OCR output on a drawing is a pre-fill, never an input. It saves typing; it does not save checking. Any tool that quietly drops recognised values straight onto an inspection form without showing you what it read is doing you a disservice, however good its recognition rate is.
What to ask your CAD administrator
If you'd rather have searchable text, the fix belongs upstream — at export, not after the fact. Worth asking:
- Does our PDF export convert text to geometry, and can it preserve text instead? (In some packages this is a "capture fonts", "preserve text" or "do not convert to outlines" option; in AutoCAD-family plotting it's tied to the PDF driver's font handling.)
- If we preserve text, do the GD&T symbols still render correctly on a machine without our fonts? Test on a clean laptop before changing site-wide.
- Can we embed the fonts rather than reference them?
- Can we produce a searchable PDF in addition to the controlled one, for internal use only — so the released drawing stays exactly as it is?
That last option is often the easiest sell. The drawing office keeps its guarantee of identical rendering; quality gets a copy it can work with.
If you can't change the export
Frequently you can't — the drawing came from a customer, or it's a released document nobody will re-issue for your convenience. In that case the workable approach is the one that treats extraction as assistance rather than automation:
- Select the region of the drawing you're reading from.
- Let OCR propose the value.
- Confirm or correct it before it becomes part of the record.
- Enter GD&T through a symbol builder rather than hoping recognition catches it.
That's slower than a promise of full automation and faster than typing every characteristic by hand — and it's the only version of this that belongs underneath a compliance document.
Export options vary by CAD package and version, and by how your site has configured them. The settings named here are starting points for a conversation with whoever owns your CAD configuration, not a universal recipe.
Reading callouts from a flattened drawing
FAIRmark's OCR assist reads dimensions from drawings that have no text layer — then asks you to confirm before anything lands on a form.