Writing Specifications and Schedules with AI

Specifications are where projects get built — or fought over. A clean Section 09 51 00 Acoustical Ceilings saves weeks of RFIs; a sloppy one creates change orders. Most firms still write specs by heavily editing a master (ARCOM MasterSpec, BSD SpecLink, or a proprietary firm library). AI will not replace that master, but it can cut spec editing time by 60-80% when used correctly. This lesson shows you how, using examples from MasterFormat Divisions 03 through 28.

What You'll Learn

• How AI speeds up editing your spec master without introducing errors
• The prompt structure that produces a usable three-part CSI section
• How to use AI to coordinate specs against the drawings and schedules
• When to trust AI spec output and when to stop and rebuild manually

The Spec-Writing Reality

Specifications come in two flavors:

• Performance specs — describe what the product must do (acoustic rating, U-value, fire rating). Short, flexible, common for design-build.
• Prescriptive specs — describe exactly what product to use, by manufacturer and model. Long, enforceable, common for federal, institutional, and high-end commercial.

AI is good at both, but the review burden differs. Performance specs can be drafted and lightly reviewed. Prescriptive specs require careful review of every manufacturer, model number, and standard reference.

The Spec Prompt Structure

A good spec-writing prompt includes six elements: project type, division/section, referenced standards, project-specific conditions, master spec baseline (if any), and output format.

Template:

Act as a CSI-trained specification writer. Draft CSI MasterFormat Section {section number and title} for a {project type and location}. Use three-part format (Part 1 General, Part 2 Products, Part 3 Execution). Reference the following standards: {ASTM, ANSI, NFPA, UL, ASHRAE as applicable}. Project-specific conditions: {climate zone, seismic category, acoustic criteria, sustainability target}. Flag any paragraph where I need to fill in a project-specific number (e.g., fire rating, STC rating, color). Do not invent manufacturer model numbers — use 'approved manufacturer' placeholders. Output as clean Markdown with CSI paragraph numbering (1.01, 1.02, etc.).

Worked Example: Section 09 51 00 Acoustical Ceilings

A working prompt for a typical office fit-out:

Act as a spec writer. Draft Section 09 51 00 Acoustical Ceilings for a Class A office tenant fit-out in Atlanta, GA. IBC 2021, Type II-B, fully sprinklered. Target NRC 0.70 in open office, STC 35 ceiling assembly rating where adjacent to enclosed offices. Use three-part CSI format. Reference ASTM E1264, ASTM E1111 (for NRC), and UL fire-rated assembly where required. For manufacturers, list 'acceptable manufacturers include Armstrong, USG, Rockfon, or approved equal' — do not pick model numbers. Output in Markdown.

What AI does well here:

• Structures the three-part format correctly
• References the right ASTM standards
• Includes typical submittal and warranty language
• Covers installation, testing, and cleaning scope

What AI does badly here:

• May invent a UL design number that does not exist
• May miss jurisdiction-specific amendments
• Will not know your firm's standard front-end or divisions 00/01 cross-references
• Will not catch conflicts with drawings unless you point them out

Always review the first and last lines of every article — that is where silent errors hide.

Coordinating Specs Against Drawings and Schedules

The highest-value AI spec task is coordination, not drafting. Paste your draft spec and your room finish schedule (as text or image) into Claude or ChatGPT:

I am going to paste (1) the Part 2 Products article from Section 09 65 13 Resilient Base and Accessories, and (2) the room finish schedule for our project. Check for: (a) products called out on the schedule but missing from the spec, (b) products listed in the spec but not used on the schedule, (c) color or profile callouts that disagree. Return findings as a table with row number, issue, and recommended fix.

A 20-minute coordination task becomes a 3-minute one. You still verify every finding.

Schedules and Tables — The Hidden Productivity Win

Door schedules, window schedules, finish schedules, and equipment schedules are tedious. AI is excellent at them.

Prompts that work:

• "Convert this messy list of doors into a clean schedule with columns: Mark, Type, Width, Height, Material, Finish, Fire Rating, Hardware Group, Notes."
• "From this room program list, generate a Room Finish Schedule with columns: Room Number, Room Name, Floor Finish, Base Finish, Wall Finish, Ceiling Finish, Ceiling Height, Notes."
• "Here is a messy 40-line electrical load list. Organize by panelboard, renumber circuits, total each panel, and flag any panel over 80% loaded."

For schedules, AI saves 70% of the grunt work. You verify the final numbers.

Division-Specific Tips

Each MasterFormat division has its own AI quirks:

• Division 03 Concrete: AI handles mix design language well but verify compressive strength, aggregate size, and admixture references.
• Division 05 Metals: Good at structural steel general requirements; careful on AWS weld specs and AESS categories.
• Division 07 Thermal & Moisture: Excellent at roofing warranty language; verify flashing details cross-reference the drawings.
• Division 08 Openings: Watch for hardware Builders Hardware Manufacturers Association (BHMA) numbers — AI often invents them.
• Division 09 Finishes: Strong overall; verify color callouts match schedules.
• Division 21-23 Fire, Plumbing, HVAC: Good at general requirements; always have a qualified MEP engineer verify performance criteria and equipment schedules.
• Division 26 Electrical: Watch NEC citations carefully; the edition adopted varies by jurisdiction.
• Division 27-28 Communications & Security: Rapidly evolving — AI's training data may be out of date by a year or more.

Front-End Documents (Division 00/01)

AI can draft Division 01 sections (General Requirements like 01 33 00 Submittal Procedures, 01 45 00 Quality Control) reasonably well for standard projects. But front-end documents (Division 00 Procurement and Contracting) depend heavily on your owner's contract form (AIA A101, ConsensusDocs, DBIA, EJCDC, federal, etc.). Do not let AI draft Division 00 without your firm's attorney reviewing.

A Personal Checklist for AI-Drafted Specs

Before the spec goes to QC:

• Every ASTM/ANSI/NFPA standard number verified
• No manufacturer model numbers that you cannot find on a current data sheet
• Paragraph callouts (e.g., "per paragraph 2.03 C 2") actually exist
• Referenced details (e.g., "per detail 5/A-501") exist on the drawings
• Fire-rated assemblies verified against UL listings
• Sustainability criteria (LEED, WELL, Living Building Challenge) properly tagged to the project target

Key Takeaways

• AI cuts spec editing time 60-80% for performance specs; more careful for prescriptive specs
• Always use the PROJECT framework plus standards, conditions, and output format
• Coordination (spec vs. schedule vs. drawings) is often the highest-ROI AI spec task
• Verify every ASTM, NFPA, and manufacturer reference against a current source
• Never let AI draft Division 00 contracting documents without attorney review