Checking a project against the PLU with a ChatGPT plugin

Install the ChatGPT plugin: Planning rules check →

The outline design of "Maison Durand", a fictional detached house in a UB zone, is almost finished. Before preparing the permit documents, the architect wants to know whether the project holds up against the regulations: the south elevation measures 7.20 m to the eaves, the site plan shows a 3 m setback from the street and the area schedule states a 38% footprint. Asking ChatGPT "does my project comply with the PLU?" often produces a reassuring, all-encompassing answer, with no regulation version and no source for each measurement. What you need here is a list of traceable checks, row by row.

This article presents the "Planning rules check" ChatGPT plugin that Educasium provides to participants in its training courses: what it produces, how it tests the rules of a PLU against dimensioned drawings, how to use it, what sets it apart from a plain prompt, how to install it in Codex and where its capabilities end.

Contents

1. What the plugin does and what it produces
2. Written rules, graphic rules and measurements: the context of the check
3. Using the plugin step by step
4. What changes compared with a plain ChatGPT prompt
5. Installing the plugin in Codex
6. When this plugin is not enough
7. Training to check your projects with AI
8. Frequently asked questions

What the plugin does and what it produces

Each relevant article of the regulations becomes a check: the plugin looks for the matching measurement in the documents it has actually read, then classifies the result. No BIM needed: dimensioned PDF drawings, an area schedule and the zone regulations are enough.

A control matrix, row by row

Each row of the matrix states the subject of the check, the type of requirement (regulatory, contractual or client objective), the source with article, page and version, the threshold and its unit, the project value with the document and page where it was read, the precision, then the status, the reason and the proposed action. An unknown column stays empty, never filled with a plausible-looking value.

The statuses are deliberately few:

• Compliant: demonstrated by the cited evidence, within its own scope only.
• Gap: a difference established on a valid comparison, with the same definition.
• Unverifiable: a necessary piece of data or capability is missing.
• To be clarified: ambiguity, contradiction between documents or uncertainty crossing the threshold.
• Not applicable: a justified, sourced exclusion.

A note that accompanies the matrix

The note summarises the documents actually read, provisional decisions, contradictions, assumptions, actions ranked by their impact, sources and limits. The default format is plain text or Markdown, with a CSV where useful; Excel or Word only if your environment allows it, with the resulting file checked. The plugin suggests corrections but never modifies the drawings or the model.

Written rules, graphic rules and measurements: the context of the check

The regulations of a PLU (plan local d’urbanisme, the French local zoning plan) include a written part, organised by zone, and graphic documents: the zoning map, graphic prescriptions and sometimes OAPs (orientations d’aménagement et de programmation, area development guidelines) with their own terms of application. A plot can be covered by several zones or by a graphic prescription that alters the written rule. The plugin therefore does not assume a PLU rather than a PLUi (inter-municipal plan), a carte communale or the RNU (règlement national d’urbanisme, the national default rules): it looks for the applicable regime and its version, and flags what it could not establish.

A definition before every figure

Emprise au sol (building footprint) is not surface de plancher (the French regulatory floor area), nor habitable area. A relative height is not an NGF level (French official altitude datum), and a height is measured from a reference ground level that the regulations specify. A roof terrace level is not the top of the building. So for each rule the plugin records definition, exceptions and units, and for each measurement the document, page and datum used. Values come from explicit dimensions on drawings or schedules: a scan without a reliable scale allows no measurement.

Using the plugin step by step

Step 1: Frame the mission. State the project, the scenario (current project or an older file to be read as of a given date) and the drawing revision. The plugin only asks the questions that change the check.

Step 2: Provide the requirements, or have them looked up. Ideally attach the written and graphic regulations for the zone. Otherwise, the plugin looks for the applicable document through the official sources accessible in your environment, such as the Géoportail de l’urbanisme (the French official planning documents portal); without up-to-date access, it presents a conditional comparison "based on the extract provided".

Step 3: Provide the measurements. Dimensioned PDF drawings, site plan, sections, elevations, area schedule. An address alone is enough to prepare the checks, not to measure the building.

Step 4: Run the comparison. The plugin documents threshold, value, unit and precision, reproduces decisive calculations with an available calculation tool and checks the condition of a conditional rule before concluding.

Step 5: Review and decide. The work moves through successive reviews (documents, measurements, comparisons, deliverable). The judgement calls remain yours.

With the fictional Maison Durand example and an equally fictional UB zone extract, the matrix might look like this:

The thresholds in this table are invented for the exercise: they are not regulatory values.

What changes compared with a plain ChatGPT prompt

A one-off prompt quickly produces a neat-looking table. The difference lies in the rules applied to every row.

• No threshold inferred from a zone label. "UB zone" says nothing about a height or a setback: without a text actually read, the rule stays open.
• Missing data is not a gap. It is unverifiable, and the missing document is listed.
• Uncertainty counts. A measurement of 9.95 m plus or minus 0.10 m against a 10 m maximum gives a range of 9.85 to 10.05 m: the check is to be clarified, not compliant.
• Documentary contradictions and regulatory gaps are kept apart. 100 m² existing, plus 20 created, minus 5 removed, declared as 120 m² instead of 115: that is a 5 m² contradiction between documents, not a planning conclusion.
• Contradictory values are kept, without picking the more favourable one.
• No overall compliance. A specific result is never extended to the whole project.

Installing the plugin in Codex

The plugin runs in Codex, OpenAI’s agent environment; its download is reserved for participants in Educasium training courses. From the plugin page, download the ZIP archive and unzip it into a stable folder you will not move. In a terminal, add that folder as a plugin source with codex plugin marketplace add "path-to-folder", then install the plugin with codex plugin add archi-controle-projet@educasium-controle-projet. Finally, open a new Codex session and ask for example: "Compare the rules provided with my project’s drawings and data. Produce a control matrix with evidence, gaps and unverifiable points."

When this plugin is not enough

The plugin is a method tool, not a planning officer. It contains no geometric engine: it reads dimensions and values, it does not compute a footprint from a drawing. It has no IFC reader. Reading a Revit model is optional, on request and with a connector available; BIM is never required. Reading scans, up-to-date access to official sources and Word or Excel exports depend on the tools in your environment.

Its measurements are never more reliable than the drawings provided: a wrong dimension gives a wrong check. Finally, the matrix is not an official ruling: it replaces neither the processing of the permit application, nor a planning certificate, nor the architect’s own reading of the regulations. The plugin does not submit, send or sign anything.

Training to check your projects with AI

A plugin does not replace the ability to frame a mission, prepare a clean set of documents and review a matrix with a critical eye. That is the purpose of our Mastering AI for Architecture training, which works through these uses on practice-based cases and gives access to Educasium’s ChatGPT plugins.

Frequently asked questions

Can the plugin tell me whether my permit will be granted?

No. It produces a matrix of scoped checks, each linked to a cited rule and measurement, but it gives no opinion on the administrative decision. A compliant check is only compliant within its scope and according to the version of the regulations actually read. The planning authority may interpret a definition, an exception or a graphic prescription differently. The matrix is there to identify gaps and open points before submission, so you can resolve them in design or document them, not to guarantee the outcome of the application.

Do I need a BIM model to use the plugin?

No. The plugin is designed to work without BIM, from the regulations, dimensioned PDF drawings and area schedules. A Revit model can optionally be read, read-only, only if you ask for it and if a connector is available in your environment. IFC files are only used if a reader is genuinely available. In every case, if a dimensioned drawing and the model contradict each other, both values are kept and the plugin asks you which version is the reference.

What happens if I do not provide the PLU regulations?

The plugin tries to identify the applicable planning document from the address or plot references, using the official sources accessible in your environment. An empty response from an online service does not prove that the RNU applies: the regime is then treated as unknown. Without up-to-date access, the plugin presents a conditional comparison and says so. The most reliable option is still to attach the written and graphic regulations for the zone yourself, with their approval or latest amendment date.

Why are there so many "unverifiable" rows in my matrix?

Because missing data is turned neither into a gap nor into compliance. If natural ground level does not appear on the elevations, the regulatory height cannot be checked; if parking is not drawn, the rule stays open. These rows are useful: they state precisely which documents or dimensions must be added to close the check. A matrix with many unverifiable points usually signals an incomplete set of drawings rather than a problematic project, and gives you the exact list of what is missing.

Checking a project against the PLU complements two other reviews we cover elsewhere: our ChatGPT method for DCE consistency checking revisits the cross-check of a file’s documents, and our ChatGPT method for the architectural description shows how to describe the project from established facts only.

Funding may be available depending on your eligibility (OPCO, FIF PL). Qualiopi-certified training organisation. To learn how to check your projects against planning rules with AI as part of our Mastering AI for Architecture training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal. You will find the download and installation tutorial on the dedicated ChatGPT plugin page.