Code research methods
Can you use ChatGPT for local code and permitting research?
For residential solar and storage teams, getting AHJ and permitting requirements right is the difference between a project that moves and one that stalls at permit. Here's an honest comparison of the methods teams use — general AI tools, manual research, FOIA, and dedicated regulatory data platforms — and exactly where each one breaks.
Last updated: July 13, 2026
From Fordje — AI code and regulatory data for residential clean-energy projects.
Part of the Residential Solar & Storage Guide.
Can I use ChatGPT to look up my AHJ or local permitting requirements?
Not reliably, and the failure mode is specific. General AI tools struggle to disambiguate jurisdictions with similar names and aren't tied to the current adopted code, so they return requirements from the wrong place or an outdated cycle.
They also confuse a conditional use (allowed with a discretionary permit) with a prohibited one, or invent a pathway that doesn't exist. For a determination that gates a real permit submission, an unsourced answer you can't trace to code text is a liability. Teams that have tried it describe the same pattern: the tool is confident, fast, and wrong in ways that are hard to catch. It can be a useful "better Google" for orienting yourself, but it can't be the source of record for a requirement you're going to build against.
Why does a general AI tool get jurisdiction questions wrong?
Two reasons. Its underlying data is the open internet, which mixes jurisdictions, editions, and outdated documents, so it can't reliably tell which AHJ or code cycle applies to a given site.
The classic version of this failure is not distinguishing between similarly named or overlapping places. See the AHJ database guide for a real example: a property with a Fairfax, Virginia address could fall under Fairfax County or the separate, independent Fairfax City — two different building departments with two different permitting processes — and Baltimore City and Baltimore County split the same way. A general AI tool has no reliable way to know which one actually governs a given parcel. And local requirements often aren't on the public internet in searchable form at all — they live in scanned PDFs, supplementary documents, or records that must be requested. Even a perfect model can only answer from what it can access, and a large share of local code lives outside what's been cleanly published online, so it fills the gap with a plausible guess.
Why isn't manual research or Googling enough on its own?
Manual research is the most common fallback and, for many jurisdictions, the only thing that actually works, but it's slow and incomplete.
Codes are often non-searchable PDFs, requirements are scattered across the ordinance, fire jurisdiction, utility tariff, and ancillary documents, and proving that a requirement doesn't exist means reading the entire relevant code. The reliable manual method is going document by document by hand, which doesn't scale. Experienced researchers tend to arrive at the same conclusion: they tried AI, they tried searching, and the only approach that held up was manual, by-hand review of every document. That's defensible and accurate, and it's exactly why it becomes the bottleneck when a team tries to scale across more projects or move into a new market, where the same hours-per-jurisdiction effort has to repeat dozens of times.
ChatGPT vs manual research vs FOIA: which is best for AHJ lookup?
Each common method trades off accuracy, speed, sourcing, and how well it scales. The honest summary: general AI is fast but unsourced and error-prone; manual research is accurate but slow; FOIA is authoritative but lengthy; a dedicated regulatory data platform is fast and sourced within the jurisdictions it covers.
| Method | Strength | Where it breaks |
|---|---|---|
| General AI (ChatGPT, Copilot) | Fast, conversational, good for orientation | Wrong jurisdiction/cycle; confuses conditional vs prohibited; no citations |
| Manual PDF / web research | Accurate; the reliable way to prove a negative | Slow; doesn't scale across jurisdictions or markets |
| FOIA request | Authoritative, direct from the jurisdiction | Lengthy; must be repeated per jurisdiction |
| Calling the AHJ or utility | Fast when you reach the right person | Inconsistent, undocumented, person-dependent |
| Dedicated regulatory data platform | Fast, sourced, comparable across jurisdictions, kept current | Limited to the jurisdictions it has built out |
What makes a dedicated regulatory data platform different?
A dedicated regulatory data platform differs from a general AI tool in what it's built on and how it answers. Instead of searching the open internet at query time, it works from a maintained, jurisdiction-specific corpus — gathered from ordinances, amendments, state code, fire criteria, utility tariffs, and ancillary documents — validated, and kept current.
Answers ladder to the correct AHJ and code cycle and cite the exact source text, so they can be verified rather than taken on faith. See the permitting requirements guide for what that looks like applied to a specific plan set.
Where Fordje fits. Fordje is an AI code and regulatory data platform built specifically for this problem in residential clean energy. It doesn't query the open internet — it maintains a validated, jurisdiction-specific corpus gathered from ordinances, amendments, state code, fire-jurisdiction criteria, utility tariffs, and the ancillary documents that explain how a code is applied, all laddered to the right AHJ and code cycle. Every answer cites its source text, so a requirement is one a team can verify and defend, and the data updates as codes and utility tariffs change.
Related questions
Is any AI reliable for code research, or none of it?
The distinction isn't AI versus no AI — it's what the AI is grounded in. A general model answering from the open internet is unreliable for local code. An AI system grounded in a maintained, validated, jurisdiction-specific corpus with citations is a different thing: the AI handles language and retrieval, but the answer traces to real source documents rather than to the open web.
How do I verify an answer about an AHJ or code requirement?
The test is whether you can trace it to the exact section of the adopted code or document it came from, for the correct jurisdiction and cycle. An answer you can't source can't be defended to a permitting office, which is why citation back to source text — not just a confident response — is the thing that makes a code answer usable.