Architecture & Design · Fundamentals
Assemble a precedent research brief
The team has collected twenty precedents for a new community center: magazine articles, award citations, a few architects' project pages, photos from a site visit and some half-remembered favorites. It lives in a shared folder and three people's heads. Before the design charrette, someone needs to turn it into a brief the whole team can read in half an hour. This guide does that consistently, without letting the AI quietly fill in architects, dates and facts it doesn't actually know.
What AI is for here, and what it isn't
Precedent research is how a design team learns from buildings that have tackled similar problems. A useful precedent brief puts each project on the same footing (what it is, what it does, how it does it, and why it matters for this project), so the team can compare them quickly and argue about the right things.
AI is good at the normalizing. Give it a messy pile of articles, captions and notes, and it will produce consistent entries with the same fields, the same length and the same structure. It can also point out gaps: no precedents in a cold climate, nothing at this budget, nothing that combines these two uses.
It is a poor source of facts about buildings. Asked about a project, it may produce a confident architect, completion year, area or cost that is partly or entirely wrong. So the AI works only from the material you've collected, every fact carries its source, and anything the material doesn't say is marked unknown. Which precedents are relevant, and what the team takes from them, is a design judgment, and it's yours.
Collect material with sources
For each precedent, gather what you have and note where it came from: the architect's project page, a published article, an award citation, a book, a site visit. Write down the date you accessed each online source.
Add the project brief: program, site conditions, climate, budget level, and the two or three design questions the team is most uncertain about. Those questions decide what makes a precedent relevant.
Normalize each precedent
Run the prompt once per precedent, with that precedent's material labelled by source. One at a time keeps sources from leaking between entries, and makes it easy to see which facts the material supports.
Create a precedent entry from the material below. Use only the
material provided. Do not use your own knowledge of this building.
SOURCES (labelled [S1], [S2]…): [paste text, captions, notes]
FIELDS (each with source tags):
- Name, location, architect, completion year: as stated in sources;
write UNKNOWN for anything not stated.
- Program: uses and approximate sizes, if stated.
- Site and climate: as stated.
- Key strategy: the main design moves (2–4), as the sources describe
them.
- Structure and materials: as stated.
- What the sources claim about performance or use (energy, daylight,
community use, cost), labelled as the source's claim.
- Relevance to our brief: leave blank; I'll write it.
RULES: Never fill a field from memory. If two sources disagree, show
both. Keep to about 150 words plus fields.
Leaving relevance blank is deliberate. Relevance is the team's judgment, and writing it yourself is what makes you read each entry properly.
Look at the set as a whole
When every entry is done, look for gaps and patterns:
Here are [n] precedent entries and our brief. Without judging quality:
1. Group them by the main strategy they illustrate.
2. Which of our design questions does each group speak to?
3. GAPS: which of our questions, conditions (climate, budget, scale,
program combinations) or strategies have no precedent in the set?
4. IMBALANCE: is the set dominated by one region, scale, era or type?
5. UNKNOWNS: fields marked UNKNOWN that matter for our questions.
The gaps list tells you what to look for next. The imbalance list keeps the set from simply reflecting where the team happened to look.
Write relevance and assemble the brief
For each precedent, the team writes the relevance line: what we're taking from it, and what we're not. "Relevant for the way the hall opens to the park on market days; not for the structure, which depended on a much bigger budget." Then assemble the brief: the design questions at the top, the precedents grouped under them, and the gaps at the end.
Check the facts you'll actually rely on (sizes, dates, costs) against the sources. If one matters for a decision, confirm it with a second source.
Worked example: the community center
Fictional example; the precedents are described generically. A team designing a community center on a tight urban site has 18 precedents from articles, project pages and a site visit. Their design questions: how to stack a gym with community rooms, how to make the entrance welcoming on a busy street, and how to keep it affordable.
The entries come back consistent, with source tags on every field and UNKNOWN where the material was silent. One entry is the exception. For a small library the team had only photographed on a visit, with no article, the entry lists an architect and a completion year.
The set review explains it: neither appears in the supplied material, which was only the team's photo captions. The AI had supplied them from its own knowledge. When checked, the architect was wrong and the year was out by six years. Both fields go back to UNKNOWN until someone finds a reliable source.
The set review also shows:
- Gaps: no precedent with a gym stacked over community rooms, and nothing at the project's budget level. The team adds three precedents found specifically for these.
- Imbalance: 12 of 18 precedents from one country's award lists. The team adds two from closer climates.
The brief that goes to the charrette is 24 entries in three groups, with relevance lines written by the team. The first discussion starts with the gap list rather than the photos, which is where the useful argument was.
Record the brief
[Project] — Precedent brief v[number], [date]
Design questions: 1. [ ] 2. [ ] 3. [ ]
PRECEDENT: [name] — [location]
Architect: [ ] [S#] Completed: [ ] [S#]
Program: [ ] [S#] Site/climate: [ ] [S#]
Key strategy: [ ] [S#]
Structure/materials: [ ] [S#]
Source's claims: [ ] [S#]
Relevance (team): [what we take; what we don't]
Images: [file] — credit [photographer/publisher] — permission/license [ ]
GAPS AND NEXT SEARCHES
[ ]
SOURCES
[S#] [title, publisher/author, date, URL, accessed date]
AI assistance: entries normalized with [tool] from collected material;
facts checked and relevance written by [team]
What stays with you: which precedents are relevant, and what you take from them.