Share
If you've never used an AI-powered spec scanning tool before, the process can feel like a black box. You upload a document and something comes out the other end, but what's happening in between, and how do you know the output is reliable? This post walks through exactly how SpecSwift processes your spec book or product manual from the moment you upload it to the moment you're looking at organized, actionable extraction results.
Step 1: Upload Your Document
The process starts with your spec book or product manual PDF. SpecSwift accepts standard PDF files, the format that virtually every spec book, product manual, and project manual is delivered in today.
A few things that affect upload and processing:
File size. Most commercial spec books run between 5 and 50 MB as PDFs. SpecSwift handles files across this range without issue. Very large files, project manuals that combine drawings and specs, may take slightly longer to process.
Text vs. scanned PDFs. Spec books produced in word processing or spec writing software (like MasterSpec or SpecLink) are text-based PDFs that process immediately. Spec books that have been printed and re-scanned, common with older projects or when paper documents are digitized, require OCR processing to convert the image to readable text. SpecSwift handles this automatically.
Addenda. If your spec book has been modified by addenda issued during the bid period, upload the addenda separately or use a version of the spec that incorporates the addenda changes. Requirements modified by addenda are contractually current, the original spec language is not.
Step 2: Document Parsing and Structure Recognition
Once uploaded, SpecSwift parses the document to identify its structure. This is more than just reading text. It's understanding the organizational hierarchy of the spec book.
SpecSwift identifies:
- CSI division numbers and titles: Division 03, Division 07, Division 26, etc.
- Section numbers and titles: 03 30 00 Cast-in-Place Concrete, 07 54 23 TPO Roofing, etc.
- Three-part section structure: Part 1 General, Part 2 Products, Part 3 Execution within each section
- Subsection headings: Submittals, Quality Assurance, Materials, Installation, etc.
This structural recognition is what makes the extraction accurate. The same phrase "ASTM C150" means something different in a quality assurance context (it's a standard the contractor's process must follow) than in a materials specification context (it's a standard the product must meet). SpecSwift uses document structure to contextualize what it extracts.
Step 3: Category Extraction Across All Sections
With the document structure identified, SpecSwift extracts the specific categories of information that matter for bid prep and project management, simultaneously, across all spec sections.
Material standards extraction pulls every ASTM, UL, FM, ANSI, NFPA, and AWS reference in the document, tagged with the spec section and part where it appears and a plain-language description of what it governs.
Submittal requirements extraction identifies every action submittal and informational submittal required across all spec sections (shop drawings, product data, samples, certificates, warranties, and O&M documentation) with the responsible party and any timing requirements specified in the section.
Approved manufacturer extraction pulls every approved manufacturer list and substitution clause in the document, organized by spec section and product category, flagging sections where substitutions are restricted or prohibited.
Trade responsibility extraction identifies scope assignment language (who furnishes, who installs, who coordinates) and flags items where responsibilities cross trade lines or where language may create scope gaps.
Testing and inspection extraction pulls every testing requirement, special inspection trigger, mockup requirement, and performance verification obligation in the document, with the applicable spec section and any specified testing standards.
Warranty and compliance extraction identifies warranty duration requirements, extended warranty obligations, and compliance documentation requirements that have cost or administrative implications.
Step 4: Organized Output by CSI Division
The extraction results are organized by CSI division and section number, mirroring the structure of the spec book itself and the way contractors think about project scope.
Each extracted item includes:
- The spec section number and title it came from
- The part of the section (Part 1, 2, or 3) where it appeared
- The extracted requirement in plain language
- The original spec language for reference and verification
This organization means a plumbing sub can go directly to their Division 22 extraction results without sorting through irrelevant output from other divisions. A GC reviewing the full project can move through divisions sequentially, the same way they'd work through a manual spec review, but in a fraction of the time.
Step 5: Review, Verify, and Act
The SpecSwift output is a starting point for your bid prep work, not a replacement for it. The final step is yours: review the extracted requirements, apply your estimating judgment, and use the organized output to build your bid, your submittal log, and your scope of work documentation.
Because everything is traceable back to a specific spec section and part, verification is straightforward. If an extracted requirement seems unusual or you want to confirm the exact language, you go directly to the referenced section rather than searching through the full document.
The result is a bid prep process that's faster, more complete, and more defensible, because everything you've priced can be traced to a specific requirement in the spec.
Further reading: How to Use AI to Extract Requirements from a Construction Spec Book and How AI Reads a PDF Spec Book Faster Than Your Estimating Team and What Are Trade Responsibilities in Construction.