Why You Can't Find That Drawing — and How Document Search Should Actually Work

You know the drawing exists. You remember reviewing it. You can picture the sheet. But twenty minutes into digging through folders, you give up and ask the person who might still have it in their email.

This is the most expensive problem in construction document management, and it's rarely a filing problem. It's a search problem — and it usually comes down to four specific failures.

1. Your scanned drawings are pictures, not text

This is the big one. A PDF exported from CAD contains real text. A PDF produced by scanning a paper sheet — which describes most as-builts and nearly every archive older than a decade — is just an image wrapped in a PDF container. There is no text in it to find.

Search that file for "roof drain" and you get nothing, forever, no matter how good the search box is. The sheet says "roof drain" in the title block; the file says nothing at all.

The fix is OCR (optical character recognition): running the image through text recognition and storing what it reads, so the words on the sheet become words you can search. Any system holding scanned drawings without OCR is storage, not document management.

Quick test: open a scanned drawing in any PDF viewer and press Ctrl-F. If you can't find a word you can plainly see on the page, that file is invisible to search.

2. It only searches filenames

Plenty of tools — and every shared drive — search the name of the file and nothing else. That works only if whoever saved it happened to type the term you're now searching for.

Real search has to cover several layers at once:

• Filename — E-201-Second-Floor-Power.pdf
• Sheet number and sheet title — the identifiers people actually use out loud
• Metadata — building, floor, trade, project, status
• Body text — the words on the sheet, via OCR for scans
• Comments and tags — the context your team added after the fact

The system should search all of it in one box. Asking someone to remember which field their term lives in defeats the point.

3. Whole-word matching quietly drops your results

This one is subtle and catches people out constantly. Most database full-text search breaks documents into whole words and matches those. Type "spec" and you will not match "Specification" — different words as far as the index is concerned.

Users never conclude "my search term was too short." They conclude "the document isn't in here" and go back to email. You lose the document and you lose their trust in the system.

Good search handles partial words, so spec finds Specification, mech finds Mechanical, and a fragment of a sheet number finds the sheet.

4. The metadata was never captured

Search can only filter on what was recorded. If nobody set building, floor, or trade at upload, "all Level 2 electrical drawings for Building C" is not a query — it's an afternoon.

This is why capture and search are the same problem. Metadata entered once at upload pays for itself every time anyone searches afterwards, and it's the difference between finding a file and pulling a set.

Five tests before you trust any system

Bring your own worst documents to any evaluation — not the vendor's clean demo data:

1. The scan test. Upload a scanned as-built. Search a word that appears only in the drawing body, not the filename. Does it come back?
2. The fragment test. Search a partial word (spec, mech, elec). Do you get the full-word matches?
3. The sheet number test. Search a sheet number the way you'd say it out loud. Does the right sheet rank first, or is it buried under files that merely mention it?
4. The set test. Ask for everything matching a combination — Level 2 + Electrical + Building C. Can you filter, or only search?
5. The ranking test. Search a common term. Are the most relevant sheets at the top, or is it just reverse-chronological?

Test 5 matters more than people expect. A search returning 400 results sorted by upload date has technically found your drawing and practically hasn't.

How ArchiveView approaches it

ArchiveView runs OCR on scanned drawings automatically, so the text on the sheet is searchable even when the file has no text layer. Every document is indexed across filename, sheet number, sheet title, metadata, comments, tags, OCR text, and AI-generated summaries — searched together from one box.

Results are weighted rather than flat: a match in a sheet number or sheet title outranks a passing mention buried in a comment thread, so the drawing you meant is at the top instead of somewhere on page four. Partial words match too — spec finds Specification. And because building, floor, trade, and project are captured as real fields, you can narrow to a set instead of scrolling a list.

Curious how it handles your worst scans? Request a demo and bring the drawing you can never find — that's the useful test.