Architectural project description with ChatGPT: method and checks
Architectural project description with ChatGPT: the content required by Article R.431-8, the PCMI4 label and the 3D-render confusion.
By Educasium
Install the ChatGPT skill: Architectural project description →
An architect is preparing the building-permit file for a single-family home extension and asks ChatGPT to draft the project description from the drawings, a 3D render and a few site photos. In the render, the new facade appears beige; however, no material or colour code has yet been agreed with the client at this stage. The most common reflex is to describe that render as if it already were the finalized project — a mistake that can prove costly if the permit officer later compares the description with a site that never adopted that shade.
This article details what a project description must contain under the French Urban Code, why not every description is a "PCMI4," the method for drafting this document with ChatGPT without turning a render's intention into a confirmed characteristic, and the situations where this pack is not enough.
Summary
- What the Urban Code actually requires in a project description
- Not every description is a PCMI4: a nomenclature not to confuse
- The most common confusion: a 3D render mistaken for a confirmed material
- Drafting the description with ChatGPT, step by step
- When this pack is not enough
- Training to make project descriptions more reliable
- Frequently asked questions
What the Urban Code actually requires in a project description
The content of the architectural project description is set out in Article R.431-8 of the French Urban Code: it describes the initial condition of the land and its surroundings, indicating where applicable existing buildings, vegetation and landscape features, then presents the design choices made to ensure the project's integration into its surroundings — siting, organization, composition and volume of buildings, treatment of buildings and boundary fencing, materials and colours, treatment of open spaces, and organization of access.
A description is not a free-form document: it answers a precise text
Unlike a presentation booklet meant for the client, the description filed with the permit answers a list set by law, not a formatting preference. A description that develops the architectural intent at length but omits the land's initial condition, or that describes materials without ever mentioning how access is organized, misses part of the content Article R.431-8 requires, regardless of how well it is written.
Not every description is a PCMI4: a nomenclature not to confuse
The building-permit form for a single-family home (PCMI) labels this description "PCMI4" in its document checklist. That reference is specific to the PCMI file; it does not apply as-is to a standard building permit (PC), a prior declaration, or a subdivision permit, whose checklists carry different references, even though the expected content of the description stays close from one file to another.
| File type | Description reference | What actually changes |
|---|---|---|
| Building permit for a single-family home (PCMI) | PCMI4 | Checklist and Cerfa form specific to single-family homes |
| Standard building permit (PC) | Reference specific to the PC checklist | Content of Article R.431-8 identical in substance, different form |
| Prior declaration (DP) | Reference specific to the DP checklist | Generally lighter documents depending on the nature of the works |
Why this confusion costs time during review
Systematically labelling "PCMI4" a description meant for a file that is not a PCMI does not necessarily invalidate the file, but signals to the permit officer a less rigorous preparation of the document checklist — a detail that can slow down the review if other documents carry similar errors. Checking the actual file type before naming each document avoids this confusion, particularly when ChatGPT reuses an earlier file as a structural template.
The most common confusion: a 3D render mistaken for a confirmed material
A 3D render helps visualize a project, but it never proves on its own that a material, a shade or a texture has been agreed with the client or definitively chosen by the design team.
Separating what is documented from what remains an intention
A rigorous description explicitly distinguishes characteristics confirmed by a written document (a quote, a technical data sheet, client sign-off) from intentions still open and only illustrated by a rendered image. Writing "the facade will be beige render" with no source locks in a choice that may never have been settled; writing "a light-coloured render is being considered for the facade, shade to be confirmed" reflects the project's actual state at this stage.
Drafting the description with ChatGPT, step by step
Step 1: Gather site plans, elevations, sections, photos and the project brief. Without a documented initial condition of the land, ChatGPT can only describe the project, never the existing situation Article R.431-8 also requires.
Step 2: Identify the actual file type before naming a single document. PCMI, standard PC or prior declaration do not carry the same description reference; this check comes before any drafting.
Step 3: Separate existing conditions and the proposal into two distinct sections. This separation, required by the very structure of Article R.431-8, prevents a description of the proposal from silently replacing the actual existing conditions.
Step 4: Link every area figure mentioned to its type and its source. An area taken from a dimensioned drawing does not carry the same reliability as an area estimated from a render; the document must state which of these two origins applies to each figure cited.
Step 5: Isolate materials, shades and finishes not yet confirmed in a separate list. This list must stay visible in the final document, never silently absorbed into the body text as if it described a settled choice.
Fictional example for testing the method: an elevation appears beige in a render, but no material or colour code is confirmed. The correct request to ChatGPT is to describe the documented volume from the drawings and put finish and colour in the list of items to confirm, never present them as settled.
When this pack is not enough
A well-drafted description does not cover every situation encountered when assembling a planning file.
Whether the project complies with the local urban plan's rules — maximum height, ground coverage, distances from boundary lines — is a regulatory analysis specific to each municipality, which ChatGPT can never establish from the project description alone: that check must be carried out separately, using the planning document applicable to the plot. A project located in a protected sector, near a listed historic monument, or in a designated site, requires additional documents and approvals beyond the description's standard content. A description prepared with AI assistance, ChatGPT like any other tool, remains in every case a draft: it must be reviewed and approved by the architect responsible for the file before any submission.
Training to make project descriptions more reliable
A well-designed pack is not enough on its own: it also needs adapting to every file type, without confusing nomenclatures from one document to another. For an architect practising as a self-employed professional (NAF code 7111Z in France), the FIFPL fund covers part of the cost of Qualiopi-certified training under 2026 criteria set at €300 per day and €900 per year, with e-learning capped at 50% of the daily rate — funding that must be arranged before training starts, not after.
Frequently asked questions
Is every architectural project description called a PCMI4?
No: PCMI4 specifically labels the description in the document checklist of a building permit for a single-family home. A standard building permit or a prior declaration use other checklist references, even though the expected content of the description stays close to what Article R.431-8 of the Urban Code sets out. Before naming the file or submitting it, it is therefore necessary to check the checklist reference actually applicable to the current file, rather than defaulting to the PCMI4 reference.
Can a 3D render be used to describe materials in the file?
A render can illustrate an intention, but it never proves that a material, a shade or a texture has been definitively chosen. Until a written document confirms that choice, it must stay in the list of items to confirm, never presented as a settled characteristic of the project. That list must be shared with the client before the file is submitted, so choices still open are settled before the review process begins.
Can ChatGPT check that the project complies with the local urban plan?
No: that check depends on the planning document applicable to the specific plot, which is set by each municipality, and must be carried out separately by the design team. ChatGPT can structure the description from the supplied material, but can never establish regulatory compliance on its own. That check must be carried out before the description is finalized, so any incompatibility is not discovered only after the file has been submitted.
What should be done if the land's initial condition is not documented?
The description should then explicitly flag that gap rather than filling it with a plausible-sounding description. Topography or existing vegetation not checked on site must never be described as an established fact in the final document. A site visit or a supplementary document supplied by the client remains the only reliable way to resolve that gap before the file is finally submitted.
A reliable architectural project description rests on three elements that do not change from one file to the next: content that follows Article R.431-8 of the Urban Code, a document reference matching the actual file type, and a strict separation between documented characteristics and intentions still open.
Once the description is stable, our ChatGPT method for CCTP by trade details how to translate these same intentions into technical requirements by trade, and our ChatGPT method for DCE consistency checking revisits the cross-check of the tender file's documents.
Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To structure your planning files with AI as part of our AI for architectural management training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal. This pack is also available on its dedicated ChatGPT skill page.