Most articles about artificial intelligence in urban planning focus on what the technology makes possible. This one takes the opposite approach, not out of pessimism, but because understanding a tool's limits is a precondition for using it responsibly in a professional setting. An urban planner who deploys AI without knowing its blind spots is exposed to errors that can have real consequences on land use decisions, on the quality of the documents produced and on the confidence citizens and elected officials place in the process.
This page is meant as much for the converted, who want to calibrate their expectations, as for the skeptics, who need an honest picture before deciding whether these tools belong in their practice. In both cases, naming the limits precisely is more useful than sidestepping them.
To place these limits in the broader context of possible uses, the page AI and the urban planner in Quebec: how to use it day to day offers a complementary overview.
Why addressing the limits is more useful than piling up use cases
The dominant discourse about AI in professional circles tends to highlight spectacular applications: a by-law drafted in minutes, a summary of a public consultation produced in an hour, an automated compliance analysis. These examples are real and the time savings they illustrate are legitimate. But presented without nuance, they create expectations that do not match reality on the ground.
Any urban planner who has already used an AI tool on a concrete case knows that results vary considerably depending on the type of task, the quality of the instructions provided, the specificity of the regulatory context and the tool's ability to handle information rooted in a local reality. Naming these variations allows professionals to deploy these tools with clear eyes rather than navigating blindly between excessive promises and avoidable disappointments.
Technical limits
The AI tools available in 2026, even the best performing ones, come with technical constraints that directly affect their usefulness in a Quebec municipal urban planning context.
AI does not know your territory
A general purpose AI tool has no knowledge of the ground you work on. It does not know that the north sector of your municipality is crossed by a hydrogeological constraint that systematically complicates construction applications. It does not remember the political decisions of recent years that shape how certain regulatory provisions are interpreted. It does not feel the tension between residential development and heritage protection in a given older neighbourhood.
This embodied knowledge of the territory is at the core of what a competent urban planner does. It is acquired through practice, through exchanges with citizens and elected officials, through direct observation of the environment. No algorithm, however sophisticated, can acquire it by proxy. AI outputs on questions that assume this local knowledge must be treated with caution proportional to that blind spot.
Training data does not reflect the Quebec reality
Large language models have been trained on massive text corpora coming mainly from English and from general North American or European contexts. Quebec municipal law, with its particularities stemming from the Act respecting land use planning and development, the structure of the RCMs, land use planning schemes, interim control by-laws and government land use orientations, is very poorly represented in that training data.
In practice, this results in outputs that look plausible on the surface but contain approximations or errors on precise points of the Quebec regulatory framework. A by-law article that cites a provision of the planning act incorrectly, a reference to an approval process that does not match the actual procedure in Quebec, or a confusion between concepts that have distinct meanings in Quebec municipal law: these errors are not rare and are not always easy to spot without solid knowledge of the field.
Hallucinations in regulatory documents
The hallucination phenomenon, meaning the tendency of language models to produce factually incorrect statements with apparent confidence, is particularly problematic when drafting or analyzing regulatory documents. An AI tool can cite a section of law that does not exist, attribute a provision to a by-law that does not contain it, or describe an administrative procedure incorrectly while keeping an assertive tone that betrays no doubt.
In everyday use, a hallucination is an inconvenience. In an official regulatory context, it can become a source of legal error. That is why any legislative or regulatory information produced by an AI tool must be verified directly against the source text before being included in an official document.
Professional and ethical limits
Beyond technical constraints, there are limits that stem from the very nature of professional practice in urban planning. These limits will not disappear with the next generations of AI tools: they are structural.
Professional responsibility cannot be delegated to an algorithm
An urban planner who is a member of the Ordre des urbanistes du Québec practises under a professional liability regime governed by the Professional Code and the OUQ Code of Ethics. They are responsible for the quality and rigour of the professional acts they perform, whether they carried them out entirely themselves or relied on tools to prepare them.
A document produced in whole or in part with the help of an AI tool remains the full responsibility of the urban planner who validated and signed it. If that document contains an error, whether AI generated it has no legal or ethical relevance. What matters is that the urban planner exercised, or should have exercised, professional judgment on the content before adopting it.
The professional act remains human
Signing an urban plan, a professional opinion or an analysis report constitutes a professional act within the meaning of the Professional Code. That act assumes independent judgment, certified competence and personal accountability. No AI platform can sign a professional act, and none can assume the responsibility attached to it.
This reality should practically guide how urban planners integrate AI into their workflow: as a tool for preparation and efficiency, not as a substitute for the professional judgment that must precede any official act.
Contextual and political judgment
Urban planning is a practice carried out at the interface of technical constraints, social dynamics and political intentions. A skilled urban planner knows how to read a room during a municipal council meeting, anticipate resistance to a densification project, and frame a professional recommendation so that it can be received in a particular political context. That kind of contextual judgment cannot be reduced to an algorithmic calculation. AI can produce a technically coherent recommendation; it cannot assess whether that recommendation is politically viable in the specific context of a given municipality at a given moment.
Ethical and social limits
The ethical issues tied to the use of AI in urban planning are not abstract. They have concrete effects on the fairness of processes and on the quality of the decisions that follow.
Representation bias in urban data
AI tools learn from data. In urban planning, that data reflects past decisions often made in contexts where certain populations were underrepresented in decision making. An AI tool trained on that data risks reproducing, or even reinforcing, the biases encoded in it.
This can show up in subtle ways: a densification analysis that fails to account for the specific needs of low income households, a consultation summary that weighs contributions differently depending on their register of language or their length, a permit application prediction tool that performs less well in areas whose characteristics are less well represented in its training data. The urban planner using these tools has a responsibility to keep a critical eye on the results and to test them against direct knowledge of the ground.
The digital divide in citizen participation
Integrating AI into public consultation processes raises a fairness question that land use professionals cannot ignore. Digital tools, however accessible they may seem in theory, are not equally accessible to every segment of the population. Older adults, people unfamiliar with digital interfaces, those without reliable Internet access and those who express themselves in a language other than French or English can find themselves at a disadvantage in a consultation process that leans too heavily on technology.
The issues specific to public consultation and the ways to navigate them responsibly are covered in the article onAI and public consultation in urban planning.
What these limits mean for your day to day practice
Knowing the limits of AI should not lead to rejecting these tools, but to using them with a rigorous professional posture. Two practical principles follow directly from the above.
Treat AI as an efficiency tool, not a decision tool
The distinction is fundamental. AI can speed up the preparation of a document, make it easier to summarize large volumes of contributions, generate a draft that serves as a working base. It should not be positioned, internally or in public communications, as the author of an official decision or recommendation. The decision always remains the act of an accountable professional.
This distinction also has change management value within teams: it reassures professionals who fear seeing their role reduced by technology, and it properly frames the expectations of those who hope AI will solve problems that are first and foremost organizational or political.
Develop a critical eye for AI outputs
The most valuable skill an urban planner can develop in relation to AI is not knowing how to use the tools, it is knowing how to assess their outputs. That means systematically asking the same questions of any result produced by an AI: is this text factually accurate? Does it correctly reflect the applicable regulatory framework? Does it account for the local context the tool could not know? Would a competent person in this field have produced the same analysis?
This critical eye is not improvised. It develops through practice, through exposure to examples of typical errors and through a basic understanding of how the tools work. That is precisely what is covered by artificial intelligence training designed for Quebec urban planners and municipal teams : not learning which buttons to press, but developing the professional judgment that makes responsible use of these tools possible.
Frequently asked questions
Will AI eventually replace urban planners?
No, and the reasons set out on this page explain why that fear, though understandable, is not based on a realistic reading of what AI does and does not do. The tasks that could be automated are those that are repetitive, high volume and not very dependent on local context. Yet the core of urban planning practice, namely territorial judgment, mediation between conflicting interests and professional responsibility, does not fall into that category. What is more likely is that urban planners who master these tools will have a concrete advantage over those who ignore them, and that some administrative tasks currently performed by professionals could be handed to technicians better equipped thanks to AI.
How do you know whether an AI tool is reliable for a regulatory task?
By testing it on cases where you already know the correct answer. Give the tool an excerpt from a by-law you know well, ask it a question whose exact answer you know, and assess the accuracy of the result. Run that test with several types of tasks and documents before deploying the tool in a context where an error could have consequences. This approach, simple and fast, gives you an empirical basis for calibrating how much you trust the tool.
Has the Ordre des urbanistes du Québec set rules for using AI in professional practice?
The reflection is under way. The OUQ has published content on the ethical and professional conduct issues tied to AI in its magazine Urbanité and has begun setting out professional responsibilities in this context. To date, there is no formal directive specific to AI in the OUQ Code of Ethics, but the general provisions on competence, integrity and professional responsibility apply fully to work produced with the help of these tools. Members are advised to follow the Order's communications as more precise guidelines take shape.
Does using AI on confidential municipal documents breach Law 25?
Potentially yes, depending on the tool used and the data submitted. Law 25 governs the collection, use and disclosure of personal information held by public bodies, including municipalities. Submitting data containing citizens' personal information to an external AI platform without having assessed that platform's privacy practices exposes the municipality to a compliance risk. Before any deployment, a privacy impact assessment is strongly recommended, particularly for tasks involving permit application files, citizen complaints or property assessment data.
How do you raise the subject of AI's limits with enthusiastic elected officials who want to deploy these tools quickly?
By validating the enthusiasm before framing it. Elected officials who want to modernize their administration's practices are right to take an interest in these tools. The point is not to slow that momentum but to make sure deployments are done in a structured way, with adequate staff training, an assessment of privacy risks and expectations calibrated to what these tools can and cannot produce. A short presentation to elected officials, backed by concrete examples drawn from the Quebec municipal context, is often more effective than a technical document for establishing that common framework.
