Which Sections of a SWPPP AI Can Draft vs Which Need a PE

A Stormwater Pollution Prevention Plan is a stack of layers. Some are the same on every site. Some are unique to the ground the project sits on. That split is why AI can draft part of a SWPPP without putting compliance at risk.

Every compliant SWPPP needs a site identification section and a site map covering facility and operator details, the project description, emergency contacts, topography, drainage patterns, structural control locations, surface waters, and material and washout storage. It needs erosion and sediment control BMPs chosen for the site, the receiving water identified and classified, an inspection and maintenance schedule, and signed certification blocks.

The trigger is settled: construction disturbing one acre or more, or a smaller site that is part of a larger common plan of development, needs a SWPPP before a Notice of Intent goes in under an NPDES Construction General Permit. The Clean Water Act sets the federal floor, and state and local agencies add stricter rules on top. Any tool drafting a SWPPP has to reflect all three layers at once.

How state rules shape what AI can and can't do

The variation between states is wide enough that no single template works everywhere, and patterned enough that the structural parts can be drafted reliably from documented rules.

A few examples:

• North Dakota requires the NOI at least seven days before construction. The SWPPP has to be prepared and kept on site but isn't submitted with the NOI.
• Kansas requires the NOI at least 60 days before construction, and work can't start until KDHE authorizes it. KDHE's construction permit says the SWPPP must be prepared under the supervision of a Kansas-licensed professional engineer, geologist, architect, or landscape architect, or a Certified Professional in Erosion and Sediment Control.
• South Carolina splits the plan into two documents: a Comprehensive SWPPP, which for sites disturbing more than 2 acres must be prepared by a South Carolina-licensed engineer, landscape architect, Tier B land surveyor, or federal employee, and a condensed On-Site SWPPP that must be available to inspectors within 15 minutes.
• Local layers stack on top. Georgetown County, South Carolina, requires third-party inspection reports submitted electronically every week.

NOI timelines, required-element checklists, and BMP menus all follow written rules, so they can be applied consistently. What doesn't follow a pattern is whether a specific discharge point reaches an impaired water body, whether a site falls inside a Coastal Zone, or whether a structural control needs real hydraulic design. Those are decisions about one site, and that line is where AI stops.

The sections AI can draft

These are the sections built from regulatory language, standard BMP menus, and project data the firm already has.

Regulatory language. AI can pull the applicable federal CGP text, the state permit conditions, and local add-ons, and place them in the structure the state expects.

Administrative fields. Facility details, operator information, the project description, and emergency contacts come straight from project data.

BMP selection from the menu. Choosing silt fence, inlet protection, or secondary containment based on site type, slope, and soil class means working from documented practices, not designing something new.

Site map elements. Topography, drainage patterns, and surface-water locations can come from Civil 3D surfaces, GIS layers, FEMA, USGS, and NRCS soils data.

Inspection schedules. These follow from permit type, acreage, and state rules. South Carolina, for example, requires weekly inspections on permitted construction sites, and sites disturbing more than 2 acres need qualified inspectors, either the C-SWPPP preparer or a certified inspector.

NOI and Notice of Termination paperwork. Filing timelines and checklists that tie the SWPPP to permit coverage.

Gradon handles this layer by writing the SWPPP from inside the project folder, working in Civil 3D, pulling public data from FEMA, NOAA Atlas 14, USGS, NRCS Soils, USFWS Wetlands, and county GIS, and filling in the firm's own state template. What comes out is a strong first draft of these sections, organized and ready for review. It is not a document ready to leave the office.

The sections that need a PE's judgment

Several parts of a SWPPP are engineering decisions. No AI output stands in for them.

Receiving-water classification. Whether a discharge point flows into a water body listed as impaired under the Clean Water Act is a factual question with enforcement consequences. A PE should verify it against current listings.

Structural BMP design. Any control that needs hydraulic sizing, like a sediment basin or detention structure, is design work, and it belongs with a licensed engineer.

Sensitive-resource risk. How close a site sits to drinking water supplies, wetlands, or other sensitive resources, and what harm that could mean, is a risk call that comes from experience.

Conflicting requirements. When federal, state, and local rules point different ways, someone has to decide which standard governs. That is professional interpretation.

Certification. The certification block is a legal statement that the plan is accurate and complete. In states with licensed-preparer rules, like South Carolina for sites over 2 acres or Kansas's supervision requirement, the licensed professional's involvement is required by law.

Even the drafted sections need a second look. A BMP that's right in the template can be wrong for a specific slope, a specific soil, or a site closer to a receiving water than the template assumed. PE review is what turns an accurate draft into a plan the firm can stand behind.

Why getting the split wrong lands on the firm

When a SWPPP is incomplete or wrong, the exposure falls on the operator and the engineer of record, not the software. A blank certification block or a missing receiving-water classification can delay permit approval and invite enforcement.

The SWPPP stays a living document through construction, stabilization, and the Notice of Termination, so an error at the start keeps compounding.

The riskiest case is a draft that looks complete. A clean, well-organized plan with the wrong receiving-water classification or an undersized BMP is more likely to slip through internal review than one that's obviously unfinished. That's the exact case PE review exists to catch.

The AI-drafts, PE-reviews workflow

The model that works keeps the two roles separate from the start.

1. The PE sets site parameters first. Receiving-water classification, impaired-water status, proximity to sensitive resources, and any engineered BMP requirements. AI works from those inputs.
2. AI drafts the rule-following layers. Regulatory language, BMP menu, site description, administrative sections, and inspection schedule, pulled from the project folder, the firm's state template, and public data, with output labeled as AI-produced.
3. The PE reviews every element. Not just the sections everyone knows need a look. BMPs against real site conditions, the receiving-water call, and the certification blocks before anything is signed.
4. Amendments follow the same split. AI updates standard language and BMP sections; a PE reviews any change that touches an engineering decision.

Gradon is set up this way. Output is labeled and routed to an engineer, and every deliverable is reviewed and signed by a licensed professional before it leaves the firm. It runs on the firm's own computers and is never trained on client data. Hand cross-referencing regulations becomes PE-directed review of an organized draft.

What to check before using AI on SWPPP work

On the state side:

• Does the state require a licensed professional to prepare or supervise the plan? Kansas and South Carolina do, and other states may bury similar rules in their CGP.
• Is there a separate comprehensive document requirement, like South Carolina's C-SWPPP and On-Site SWPPP split?
• What's the NOI lead time? Seven days in North Dakota, 60 in Kansas. A late-found error may not leave time to fix.
• Do local jurisdictions add BMP standards, inspection formats, or third-party reporting?
• Is the project near impaired waters or inside a Coastal Zone?

On the tool side:

• Does it work from state-specific templates, or produce a generic federal document someone has to adapt by hand?
• Does it pull site inputs from current public data instead of making an engineer type every parameter?
• Is its output labeled as AI-produced and structured for PE review?

Confirm the state details with the regulatory authority itself before relying on any draft.