Architectural critique with Claude: ground each finding in the project
Architectural project critique with Claude: six reading axes, a fictional example, and its limits before you use it.
By Educasium
Review a project with Design Critic →
A jury, a client or a colleague may already have handed you feedback like "the project lacks coherence" or "the whole thing needs more work." That kind of comment, however sincere, does not say where to look or what actually drove the judgment. You are left guessing which piece of the file fell short, with no certainty you are fixing the right problem before the next submission.
Design Critic offers a different way to get this kind of feedback from Claude. You attach a plan, a section, an elevation or a render, along with the brief and the design intent, and the pack ties each observation to something visible in the drawing rather than a general impression. It keeps two to four supported findings, organized around six reading axes, and deliberately stops short of any code, structural or cost conclusion.
This article details those six axes, shows on a fictional example where the line sits between a finding and an assumption, and states plainly what the pack never does — particularly on feasibility, which stays a question to investigate rather than a constructability check.
Summary
- Why a critique finding must point to a location in the project
- The six reading axes of Design Critic
- A reading example to check
- Preparing your request step by step
- What Design Critic does not do
- Training to make project critique more objective
- Frequently asked questions
Why a critique finding must point to a location in the project
An architectural critique becomes usable only once it ties an observation to something visible in the drawing: a room, a circulation path, a facade, a scale relationship between two volumes. "The project lacks coherence" points to nothing specific to fix; "the entrance described as calm is crossed by the service route marked on the ground-floor plan" points to an exact location, evidence in the document and a consequence worth studying. Design Critic builds every finding around that requirement, rather than stacking general impressions about the project as a whole.
A critique that cites nothing cannot be checked
A finding that does not point back to any area of the plan or image cannot be rechecked after the project is revised: nothing tells you whether the problem disappeared or simply moved elsewhere in the drawing. A finding that cites a specific room, a level or a facade, by contrast, lets you return to the exact same spot after a change, to check whether the observation still holds. That requirement for traceability is what separates a useful critique from a judgment that merely sounds plausible.
What changes once the critique rests on the brief
An exterior view alone cannot support conclusions about a project's entire internal circulation, just as a plan alone says nothing about the intent the brief is trying to reach. Supplying the brief and the design intent alongside the drawings gives Design Critic a point of comparison: does the project answer what it claims to be pursuing, rather than an external standard nobody actually stated? Without that declared intent, the pack can only describe what it sees, with no way to say whether that matches what the project was aiming for.
The six reading axes of Design Critic
Design Critic organizes its reading around six fixed axes rather than producing one general impression: concept consistency, scale and proportion, circulation and program, materiality and detail, relationship to context, and ambition versus feasibility. Every finding it keeps ties back to one of these axes, which lets you immediately see what kind of question it is answering.
| Axis | Question asked | What the axis does not cover |
|---|---|---|
| Concept consistency | Does the stated design approach show up in the plan and volume choices? | Does not judge the aesthetic value of the concept itself |
| Scale and proportion | Do the perceived dimensions match the brief and the intended use? | Measures no precise dimension or code tolerance |
| Circulation and program | Do the routes serve the spaces without conflict between different flows? | Does not calculate any code-mandated route width |
| Materiality and detail | Are the stated materials consistent with the visible detailing? | Does not check material availability or actual cost |
| Relationship to context | Does the project respond to its surroundings — street, neighboring plot, orientation? | Does not replace a sunlight or overlooking study |
| Ambition versus feasibility | Does the stated ambition remain plausible given what is drawn? | Does not price or validate the project's actual constructability |
Why the pack limits findings to two or four per reading
Scoring every axis on a scale every time would create an impression of rigor that nothing supports: a project may deserve no remark at all on three axes and two solid findings on a single one. Design Critic therefore prioritizes two to four genuinely supported findings over an artificial score across all six axes. Silence on a given axis never means that axis was approved — it only means no visible element, among the documents supplied, prompted a remark on it.
The feasibility axis is not a constructability check
The "ambition versus feasibility" axis questions how plausible the project's stated ambition remains given what appears in the supplied drawing — not the actual feasibility of construction. Design Critic does not price anything, does not verify a load path, and replaces neither a structural engineering firm nor a construction economist. A remark on this axis is a question to investigate further, not a conclusion about what can actually be built. For a genuine code or structural check, resources such as the architects' professional body remain the necessary next step.
A reading example to check
Fictional example, with no connection to any real practice or building: imagine a residential extension we will attribute here to an imaginary firm, Atelier Cardon. The brief passed to Claude states a simple intent; the attached ground-floor plan shows a detail that partly contradicts it.
Fictional supplied brief: "a calm entrance, protected from street noise and everyday coming and going." Fictional finding on the plan: a service route, leading to the bike storage and the bins, crosses that same entrance space.
The useful finding concerns whether these two elements remain compatible during the hours when that service route sees the most traffic — deliveries, bikes going out, waste collection. It goes no further: it does not set a code-mandated width for that route and does not propose moving a partition, since both actions call for a study Design Critic never performs.
What this example lets you check
In your own file, the exercise is to locate this same type of crossing: a space described with a specific quality in the brief, crossed on the plan by a use that puts that quality under strain. Then check that the brief genuinely claims that quality — a conversation with the client can confirm or qualify an intent the written document only half expresses.
What it does not let you conclude
Without an actually supplied plan, this example remains a method illustration and says nothing about a real project in progress. It also does not license inventing a code-mandated route width absent from the drawing, nor deciding on your own that a partition must move: both decisions belong to a follow-up study, not a plan reading.
Preparing your request step by step
Five steps, followed in this order, avoid the most common situation: a critique that feels right at the moment but becomes impossible to check once the project has been revised.
Step 1: Attach the documents with their revision index and design stage. A plan with no revision index and no stated stage — sketch, preliminary design, full project — invites a critique built on an already-outdated version. State both pieces of information every time before asking for a reading.
Step 2: Describe the brief and the design intent. State what the project is trying to achieve — what relationship between spaces, what atmosphere, what starting constraints. Without that description, Design Critic can only compare the drawing against itself, never against what it was meant to reach.
Step 3: Choose the design stage and the tone of the critique. A sketch-stage project can take more open-ended remarks than a permit-submission file. Also state the tone you want — mentor, direct or encouraging — knowing that tone changes the wording, never the level of evidence required behind each finding.
Step 4: Ask for a summary and two to four findings, not a score per axis. Asking for a systematic score across all six axes pushes the pack toward an evaluation that looks complete without necessarily being so. Prefer a summary followed by a limited number of genuinely supported findings.
Step 5: Locate the evidence for every finding before acting on it. Before folding a remark into your next iteration, locate the element of the plan or image that justifies it. A finding with no locatable evidence should be discarded, even if it sounds right on first reading.
What Design Critic does not do
Design Critic validates no code compliance, performs no structural calculation and prices nothing — those checks belong to a control body, a structural engineering firm or a construction economist, not to a plan reading done by an AI. It also does not redesign anything automatically from its own findings: the decision to change a plan remains entirely yours, once the identified lead has been validated by the necessary studies.
It does not replace a jury's or a client's judgment either: it is a preparatory read, useful before a presentation, not a substitute for that presentation itself. In our exchanges with practices testing this method, the difficulty is almost never understanding the six proposed axes — it is accepting that a finding with no visible evidence must be dropped, even when it matches an intuition the whole team already shares.
Training to make project critique more objective
Knowing how to tell a supported finding apart from an impression, and recognizing when a feasibility remark calls for a follow-up study rather than an immediate conclusion, is learned through practice. Educasium's Master Claude training devotes modules to using Claude in design professions, for people who have never used AI in their professional practice before. For a practice owner or a self-employed architect under the relevant scheme, the FIFPL covers part of the cost of a Qualiopi-certified training under the 2026 criteria set at €300 per day and €900 per year, with e-learning capped at 50% of the daily rate.
Frequently asked questions
Can Design Critic validate a project's code compliance?
No. The pack reads what is visible in the supplied documents and compares it against the stated brief, but it performs no code check, no structural verification and no cost calculation. A remark on the feasibility axis flags a question to pursue with a structural engineering firm or a control body, never a validation or invalidation of the project's compliance. Only a qualified professional — an approved control body, a structural engineering firm or the relevant permitting authority — can issue that validation, independently of any reading an AI produces.
What happens if I only attach an image with no plan or section?
The reading then stays limited to what that image actually shows: an exterior elevation or render cannot support conclusions about internal circulation or the layout of spaces, for instance. Design Critic generally flags this limitation rather than guessing the content of the missing documents; for a more complete critique, attaching a plan and a section remains preferable to a single isolated image. The findings you get in that case stay confined to the axes the image actually lets you observe — materiality, context, the apparent coherence of the concept — never circulation or the interior brief.
Can the pack score every axis out of ten to compare several projects?
That is not the recommended use, even though nothing technically prevents asking for it. A systematic score across all six axes creates an impression of precision the method does not guarantee: it prioritizes two to four supported findings over a complete numeric evaluation. To compare several projects rigorously, a structured decision method is better suited than a score derived from a design critique.
Do I need particular CAD software to use this skill?
No: Design Critic works from whatever you attach — an image, a PDF export or a screenshot of a plan — without requiring a specific format or software. The quality of the reading depends mainly on how legible the supplied document is and how precise the communicated brief is, not on the software used to produce the drawing. An overly compressed export or a blurry screenshot can hide useful detail, though, which reduces the precision of the findings you get back by the same margin.
A project critique becomes useful once it points to a precise location, cites evidence and proposes a question to investigate rather than a final verdict. That is exactly what the Design Critic architectural critique pack structures, stopping short of any code or cost conclusion that falls outside its method.
Once the critique request is properly framed, turning a still-vague intent into a precise instruction for Claude follows a related method: our guide to structuring a prompt from a brief details how to apply the same evidence and boundary requirements before handing any other task to the AI.
Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To make your project critiques more objective with AI as part of our Master Claude training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal.