Cited AI, Consistency, and Defensibility

The credibility problem with “uncited intelligence”

Most general-purpose AI tools generate responses by predicting what text sounds correct based on patterns in their training data. That works surprisingly well—until it doesn’t.

The problem isn’t that these tools are always wrong. It’s that when they are wrong, they’re often wrong in ways that are hard to detect.

Two issues consistently raise concern in public-sector use.

Hallucination

AI can produce answers that are fluent, confident, and incorrect—without any indication that uncertainty is present.

Opacity

Even when an answer happens to be correct, the system often cannot show why it arrived there or point clearly to the governing source.

In planning and zoning, that combination is especially risky. Staff are accustomed to answering questions with context, caveats, and citations. AI systems that present a single, confident response—without visible sourcing—break that norm.

When a resident challenges an answer—
“Where does it say that in the code?”
“Who decided that interpretation?”
“Is that still current?”

An uncited response offers no clear path forward.

“Because the AI said so” isn’t just unnerving. In fact, in a regulatory environment, it’s indefensible.

Why citations change the game

Cited AI systems tie each answer to a specific, verifiable source—an ordinance section, an adopted policy, or a dated amendment. In planning, that connection isn’t optional. It’s how information earns legitimacy.

Citations matter because they support three expectations that already exist in local government.

Transparency
Planning departments are expected to show their work. When an answer includes a clear reference to the governing text, staff and the public can see how an explanation was derived and what authority it rests on.

Auditability
Planning guidance must be reviewable. Citations allow staff, supervisors, attorneys, and auditors to trace an answer back to the adopted language and evaluate whether it was applied appropriately.

Trust
Public trust in planning doesn’t come from polished explanations. It comes from consistency, traceability, and the ability to confirm information independently. Answers that point directly to adopted documents reinforce confidence that guidance is grounded in law, not interpretation alone.

Without citations, AI-generated responses may sound helpful but remain detached from the regulatory framework that gives them meaning. With citations, AI becomes a navigational aid—helping users find and understand adopted authority rather than replacing it.

Consistency is a public obligation

Human staff vary. That’s normal. AI variability is a different problem.

Uncontrolled AI systems can return different answers to the same question depending on phrasing, timing, or context. In consumer tools, that’s an inconvenience. In local government, it’s a risk.

Inconsistent answers create:

• Unequal treatment of residents
• Confusion at the counter and online
• Exposure to claims of arbitrariness or bias

This concern is not theoretical. The American Planning Association consistently emphasizes that planning decisions must be rooted in adopted authority and applied predictably. APA’s Ethical Principles in Planning state:

“A planner’s primary obligation is to serve the public interest and to provide clear, accurate information that is grounded in adopted plans and regulations.”
https://www.planning.org/ethics/ethicalprinciples/

Consistency is not just a professional ideal. It is a prerequisite for public confidence and legal defensibility.

Courts routinely examine whether similar cases are treated similarly. An AI tool that answers the same zoning question differently on Tuesday than it did on Monday undermines that standard.

How consistency is enforced in defensible AI systems

Consistency doesn’t come from better prompts. It comes from bounded systems.

Planning-grade AI relies on:

• Fixed, authoritative source documents
• Version-controlled content
• Deterministic retrieval before response generation

In other words, the AI doesn’t “remember better.” It’s constrained better.

This mirrors long-standing best practices in legal research tools, which prioritize authoritative sources and stable interpretations over creative inference. In regulatory environments, that’s not a limitation. It’s the point.

Defensibility is where AI adoption succeeds or fails

Most AI discussions fixate on efficiency. Public agencies ultimately care about defensibility.

A defensible AI-assisted answer can withstand:

• A public records request
• A legal challenge
• A council meeting
• A front-page headline

That bar is high—and it should be.

Legal scholars Danielle Citron and Frank Pasquale have warned that automated systems must preserve accountability structures rather than obscure them, noting that opaque tools often shift risk without reducing it (The Scored Society, Washington Law Review).
https://digitalcommons.law.uw.edu/wlr/vol89/iss1/2/

The Lincoln Institute of Land Policy reinforces this same principle specifically in the context of land-use governance. In its work on data and technology in planning, the Institute notes:

“Data and analytical tools must be transparent, well-governed, and understood by both decision-makers and the public if they are to support equitable and defensible land-use outcomes.”
https://www.lincolninst.edu/topics/data-technology

The Institute further emphasizes that:

“Consistency in the application of land-use regulations is central to public confidence in planning institutions.”

AI that obscures the chain of responsibility does not reduce risk. It redistributes it—often onto frontline staff who didn’t choose the tool.

What this means for planners right now

This is not an argument against AI. It’s an argument against careless AI.

Used well, AI can:

• Improve public access to complex codes
• Reduce staff time spent on repetitive questions
• Support clearer, more consistent communication

Used poorly, it can:

• Undermine trust
• Create legal exposure
• Complicate, rather than simplify, staff workflows

The dividing line is governance.

Cited sources. Controlled inputs. Consistent outputs. Defensible answers.

Anything less is not ready for public-facing use.

The Wrap Up

Planning has always been about interpretation anchored in adopted law. AI doesn’t change that. It raises the stakes.

Tools that respect authority, consistency, and defensibility can improve access and efficiency without increasing risk. Tools that don’t simply move uncertainty downstream—to staff, to applicants, and to the public.

As communities explore how AI fits into planning workflows, the most important question isn’t what the technology can do, but whether it operates within the same governance framework planners already rely on every day.

If your department is evaluating AI-assisted tools for planning and zoning, the place to start is with citation, consistency, and accountability—before efficiency.

To see how planning-grade AI can support research and public access while preserving adopted authority, book a consultation to see how enCodePlus approaches AI-assisted planning tools.

Need deeper context?