Air quality permitting is one of the most technically demanding areas of environmental law. The regulations are dense, the cross-references are numerous, the permit conditions are facility-specific, and the consequences of getting it wrong range from compliance deviations to enforcement actions. It is exactly the kind of domain where AI should be able to help, and exactly the kind of domain where general-purpose AI tools have consistently fallen short.
This article explains why general AI struggles with air quality work, what purpose-built AI does differently, and what the practical implications are for consultants and facility EHS teams.
Why general-purpose AI gets air quality wrong
General-purpose AI models are trained on broad datasets that include regulatory text, technical documents, and legal materials. They can describe what BACT means, explain the Title V permit program, and summarize how NSR applicability works. For general questions, they perform reasonably well.
The problem is that air quality compliance is not a general question. It is a facility-specific, permit-specific, regulation-specific question where the details are the answer. The difference between a 12-month rolling average and a calendar year average is not a nuance. It is the compliance period. The difference between a permit limit of 0.10 lb/MMBtu and 0.15 lb/MMBtu is not a rounding error. It is the basis for whether a stack test passes or fails.
General AI models do several things that are dangerous in this context:
- They cite regulations from memory, which may be outdated or misremembered
- They apply generic regulatory knowledge rather than the specific conditions in a facility's permit
- They produce confident answers even when the information is incorrect
- They do not distinguish between a federal rule and a state-specific requirement that supersedes it
- They cannot access current CFR text and do not flag when their knowledge may be stale
The hallucination problem in regulatory work: AI hallucination, which means generating plausible but incorrect information, is a known limitation of large language models. In most contexts it is an annoyance. In air quality compliance, where a wrong CFR citation or an incorrect permit condition interpretation can produce a compliance deviation, it is a genuine risk. A tool that is confident and wrong is more dangerous than a tool that is obviously uncertain.
What purpose-built AI does differently
Purpose-built AI for air quality compliance addresses the failure modes of general AI through a different architecture and a different set of constraints.
General-purpose AI
- Answers from training data memory
- May cite outdated regulations
- No facility-specific context
- Confident even when wrong
- No structured output for professional review
- No verification checklist
Purpose-built AI
- Retrieves live CFR text before answering
- Cites current regulations, not memory
- Answers in context of specific permit
- Flags uncertainty explicitly
- Structured output for licensed professional review
- Verification checklist with every response
The most important difference is regulatory retrieval. AirComply retrieves current text from the Code of Federal Regulations before answering any question that involves a regulatory citation. The answer is grounded in what the regulation actually says today, not what a model learned about it during training. When regulations change, the answers change. The model does not have to be retrained.
The professional liability dimension
Air quality compliance work is signed by licensed professionals. The engineer or scientist who puts their name on a permit application, a BACT analysis, or a deviation report carries professional responsibility for its accuracy. Using a tool that produces unreliable outputs and then signing the work product is a professional liability exposure, not just a quality concern.
Purpose-built AI is designed for the professional workflow, not around it. Every output from AirComply includes a verification checklist: the specific facts the licensed reviewer needs to confirm before the output is finalized. The AI does the drafting and the initial analysis. The professional confirms the reasoning, checks the cites, and signs. That division of labor preserves the professional relationship with the work product while capturing the efficiency gains from AI assistance.
What the workflow actually looks like
The practical change is not that AI replaces the consultant. It is that the consultant's time moves from extraction and drafting to review and judgment. The tasks that required 30 to 40 hours of reading and organizing information now take minutes. The tasks that require professional expertise, client judgment, and regulatory interpretation remain entirely with the licensed professional.
For a permit deconstruction, that means the consultant reviews a structured extraction rather than producing it. For a BACT analysis, it means the consultant reviews a documented top-down analysis with current RBLC comparables rather than assembling the framework from scratch. For a SAMR, it means the consultant reviews a pre-filled report rather than drafting it line by line.
The output quality is the same. The time investment is a fraction. The professional's role shifts from production to quality assurance, which is where their expertise actually adds value.
The skeptic's question
The most common skeptical question from experienced air quality professionals is: how do I know the AI got it right? It is a legitimate question and the right one to ask.
The answer is that you review it, the same way you review work from a junior staff member. The AI produces a first draft. The licensed professional applies their expertise to verify the reasoning, confirm the citations, check the facility-specific details against the permit, and sign off. The work does not go out the door without professional review. What changes is that the first draft took three minutes instead of three days.
For professionals who have spent careers developing expertise in air quality compliance, the value of that expertise does not diminish when AI handles the drafting. It concentrates in the review, the judgment, and the client relationship, which is where it was always most valuable.
See purpose-built AI for air quality compliance
AirComply retrieves live regulatory text and produces structured outputs for professional review.
Request demo accessWhere this is heading
The question for air quality consulting firms and facility EHS teams is not whether AI will change compliance work. It is whether they will be using the right AI when it does. General-purpose tools applied to specialized regulatory work produce unreliable results. Purpose-built tools designed for the specific domain, with the specific failure modes addressed, produce outputs that licensed professionals can actually use.
Air quality compliance is technically demanding enough that it requires AI built specifically for it. That is not a limitation of the field. It is what makes purpose-built solutions durable. A general tool can be applied to any domain by anyone. A purpose-built tool is designed for the professionals who do this work and built to meet the standard they are held to.