AI for Architecture Site Analysis: Research Before Design

Use AI for architecture site analysis to organize sourced information about a site, identify missing evidence, and prepare questions for the design team. Check every constraint against the relevant map, document, or professional finding before using it to guide design. A generated site narrative is useful only when you can trace its statements back to evidence.

Tools like fenbs keep source references, assumptions, and decisions together as the research develops. That gives the next assistant a record of what the team actually checked. This guide covers research before design; the broader selection of AI architecture tools is a separate question.

How do you define a useful site analysis task?

1. Identify the parcel, jurisdiction, proposed use, and design question. Confirm boundaries from appropriate project records rather than letting the AI infer them from an address.
2. Collect the authoritative documents for that question and note the issuer, date, location covered, and source link.
3. Ask an approved AI tool to organize the supplied evidence into observations, possible implications, and unanswered questions.
4. Check the underlying sources with the relevant professionals, then record which design responses people approve and what remains provisional.

Begin with one decision the research should inform. For example, a hypothetical studio might ask which access and drainage questions need answering before it studies entry locations. That is a clearer assignment than asking for a complete “best design” based on an aerial photograph.

Set the data boundary too. Confirm which surveys, client documents, and consultant reports may enter the chosen AI system. The AIA’s responsible AI guidance↗ (opens in a new tab) calls for protection of confidential information and qualified review of generated outputs.

Which sources should the AI summary point to?

Build a small source register before generating a narrative. Record the parcel or area each source covers and the question it can answer. A regional climate dataset, a property survey, and a zoning map describe different things; they should not become interchangeable evidence because they mention the same city.

For a US starting point, NOAA’s U.S. Climate Normals↗ (opens in a new tab) provide information about typical conditions. They are not a site-specific future-weather simulation. The USGS guide to flood maps↗ (opens in a new tab) distinguishes FEMA flood-map sources from forecast and other flood products.

Use the actual jurisdiction’s sources for zoning. For example, New York City’s Zoning Resolution↗ (opens in a new tab) contains text and maps governing land use and development. A summary of another city’s rules cannot establish what applies to your parcel.

How should you ask AI to summarize site constraints?

Request a source reference for every statement and keep interpretation in a separate field. If a document is unavailable, the assistant should mark the question unanswered. A plausible setback or soil description is still unsupported when the evidence is missing.

Use only the supplied site evidence.
For each finding return:
1. Observation and exact source location
2. Area covered, date, and units where relevant
3. Possible design implication, labeled as a proposal
4. Missing information or conflicting evidence
5. Question for the responsible reviewer
Do not invent site measurements or declare approvals.
Keep surveyed facts separate from regional data and assumptions.

Test the format with a few findings you already understand. Check whether the assistant preserves a document’s limits and distinguishes a proposal from a requirement. The same approach applies to other AI tasks at work: a defined input and acceptance check make the output easier to assess.

How do you verify maps, dates, and measurements?

Open each cited source yourself. Confirm the correct parcel, map layer, publication status, and date, then inspect the specific passage or location behind the statement. A citation to a homepage is not enough to reproduce a parcel-level finding.

Check units before comparing numbers. Keep elevations tied to their stated reference system, areas tied to their measurement basis, and climate observations tied to the selected location and period. Ask the appropriate consultant to resolve incompatible evidence instead of asking the assistant to average away the difference.

Keep the review result beside the source in fenbs Links and listings. Note what was checked and what the source cannot establish, so the next assistant has a useful reference rather than a bare URL.

How do you turn evidence into design questions?

Use three distinct statements: what the evidence says, what you think it might mean, and what a person has approved. For example, a documented afternoon exposure might prompt a shading study. It does not prove that a particular facade treatment meets the project’s performance goals.

Create follow-up work for the unresolved step: verify the obstruction model, ask the civil engineer about drainage, or confirm the relevant zoning provision. Write the expected response and the evidence needed to close the question. Keep a design response provisional until the responsible person accepts it.

When new information changes a constraint, record the replacement decision and identify affected tasks. Preserve the earlier reasoning so a later reviewer can understand why the team changed direction. This is one practical use for an AI brain that stores work outside a single chat.

How to do this with fenbs

1. Create a free account and add a project for the site research.
2. Add source references in Links and listings, including the covered area, document date, and relevant location. Keep original surveys and reports in their approved systems.
3. Write context notes that distinguish checked observations from assumptions. Add tasks for missing evidence, with review steps and the responsible contact in each plan.
4. Record the decisions people make and save Where we left off with the current evidence gaps and next step. Ask the next connected assistant to read that context before continuing.

fenbs does not provide geographic mapping, site simulation, or zoning determinations. Use specialist tools and professional review for those jobs. It fits when the sources and follow-up work need to stay understandable across people and assistants; a single research note may be enough for an isolated exercise.

What belongs in the final research handoff?

Deliver a short statement of the project question, the reviewed sources, the confirmed findings, and the unresolved work. Name the assumptions behind each proposed design response. Someone taking over should be able to reproduce the important checks without rereading every AI conversation.

Start the next session by checking whether source documents or the project scope changed. Carry forward the evidence that still applies and reopen affected questions. The AI at work hub collects other workflows built around that same habit of checking generated work against its source.