Creating an Adjacency Diagram With AI

Adjacency studies used to take hours. Sorting through a program, weighing sound and vibration constraints, and grouping rooms by workflow is tedious even for experienced designers. An AI assistant with the right prompt can produce a solid starting point quickly, and with a few refinements it looks like something an architect would want to present to a client.

Upgrade to a Prompt Sandbox

Before writing the prompt, open a separate document to act as a prompt sandbox. Google Docs or Microsoft Word both work. A dedicated document is better than typing directly into the chat window, which tends to be squished, prone to accidental submissions, and hard to reshape when the prompt gets longer. Name the sandbox something obvious, like PromptSandbox, and use it to draft, edit, and refine every prompt before pasting it into the chat.

This small workflow change pays off immediately. You can see the full prompt at once, fix spacing, copy and paste sections, and keep earlier versions for reuse. It also makes it easier to share the prompt with teammates, which is how best practices spread inside an office.

Start with Role and Task

The first prompt uses the role, task, format structure. The role is senior architectural designer, which signals to the model that the output should feel architectural rather than purely analytical. The task is to convert the synthesized building program into a functional adjacency matrix. Writing the task explicitly matters, because an adjacency matrix has a specific format and the model needs to know that is what you want.

Define Logic and Symbols

The logic and symbols section is where the diagram starts to look like something a designer would actually use. Pick symbols that convey visual weight, because a matrix is not just data, it is a piece of communication. A clean set of symbols looks like this:

• A pound sign represents direct adjacency, for spaces that require a shared wall for workflow or infrastructure.
• A plus sign represents a desirable adjacency, which is convenient but not critical.
• A period, or a middle dot if you prefer a lighter mark, represents a neutral relationship where adjacency does not matter.
• Two spaces represent an undesirable adjacency, for rooms that should be separated due to noise, vibration, or privacy.

The middle dot is an optional refinement that makes the diagram cleaner. Hold ALT and type 0183 on a Windows keyboard, press Option Shift 9 on a Mac, or long-press the period on a mobile keyboard to produce the character. It is a small detail, but small details are what make an AI-generated diagram feel like architectural work.

Ask for a Square Markdown Table

The format section ties everything together. Ask for a square markdown table, which means all columns and rows are equally spaced. A square grid looks deliberate on screen and reads like a real adjacency matrix. Also ask for room IDs as both row and column headers, and for a simple legend that explains the symbols. A request for a simple legend produces a short note under the table that keeps the whole thing self-explanatory.

Once the prompt is ready in the sandbox, copy the full text and paste it into the chat. Submit it, and the model returns a complete adjacency matrix with room IDs along the top and the left side, an X or a dash running down the main diagonal for cells where the row and column match, and the chosen symbols filling the rest of the cells. The legend appears below the table for reference.

Resort the Matrix to Cluster Adjacencies

The first pass usually sorts rooms in numerical order, which is fine but not especially architectural. A second prompt can cluster the matrix to show how rooms group together, which starts to resemble a real plan. Go back to the sandbox, skip the role section since it has not changed, and jump straight to the task. The new task is to resort the rows and columns of the adjacency matrix to maximize direct adjacency clusters along the main diagonal, using the word direct in quotes to tie it back to the legend.

Add a logic line that groups rooms with the most pound connections, then reaffirm the format with instructions to maintain the same visual weight and produce a square markdown table. If you want to control the diagonal symbol where the row and column are the same room, include that in the prompt, though the model may ignore it and choose its own glyph. Either way, the second matrix tells a more architectural story. It is a cluster diagram that shows which spaces naturally belong together and which sit at the edges as outliers.

Compare the Two Diagrams

Running both prompts produces two different but useful views of the same program. The numerical-order matrix is easy to read against the program document and useful for verification. The clustered matrix is closer to the kind of schematic a designer would sketch by hand during a programming workshop. Scrolling around the output, the legends and notes may end up in slightly different places depending on the chatbot, but everything you need for a polished handout is there.

From here, the matrix can be dropped into a presentation, a programming report, or a client workshop. It gives the team something concrete to react to early in the process, and it captures the adjacency reasoning in a format that is easy to revisit as the design evolves.

Draft your prompts in a sandbox document, use role, task, format for the first pass, and include logic and symbols that have real visual weight. Ask for a square markdown table with a simple legend, then use a second prompt to cluster the matrix along the diagonal for a more architectural output. The whole exercise turns adjacency analysis from a grinding manual task into a fast, client-ready diagram that shows how the program wants to organize itself.