How Architects Use AI for Fee Proposals, Briefs and Paperwork

Architects use AI to turn first-meeting notes into a written client brief, draft the scope, assumptions and exclusions sections of fee proposals from their own past proposals, and produce minutes, site reports and letters. Work out the fee yourself from a resourcing sheet; AI writes the words around your numbers and shouldn't choose them.

The brief is where AI helps most and gets least credit. A vague brief produces scope creep that no fee proposal can fix, and small practices often skip writing one properly because it takes an evening. The other thing that makes AI useful here is your own paperwork. An assistant that has read three of your approved proposals, your standard exclusions and your appointment terms writes in your voice and your terms. One that hasn't writes generic proposals any client could get from anyone.

Follow me on Instagram@sagnikteaches

Set up a practice project with your own documents first

Create a shared project in Claude Team or ChatGPT Business, or keep a folder of source files in SharePoint for Copilot in Word to draw on. Put in:

Connect on LinkedInSagnik Bhattacharya

• Three past proposals you were happy with, for different project types (say, a rear extension, a loft conversion and a small commercial fit-out).
• Your standard terms of appointment and your description of each work stage.
• Your standard exclusions and assumptions list.
• A one-page house style note: how you address clients, words you avoid, how long sections should be.
• A brief template with the headings you want every brief to have.

Business plans cost $25 a seat billed monthly or $20 annually for ChatGPT Business or Claude Team standard seats (two-seat minimum). Microsoft 365 Copilot Business is $21 a user a month on annual billing, or $23.50 bundled with Business Standard. Setting up the project takes an afternoon; it's the step that makes every later prompt shorter.

Subscribe on YouTube@codingliquids

The house style note is the document practices skip, and it changes the output more than any other. A filled-in version for a small residential practice:

House style: proposals and client letters
- First names after the first meeting; sign off with the
  architect's first name
- Say "we", never "the practice" or "our firm"
- Avoid: bespoke, stunning, vision, journey, passionate,
  world-class, light-filled
- Use our stage names from the terms, not generic ones
- "Understanding of the project": 150 words at most
- Scope: bullet points under each stage
- Money: always "fee" for our charges; "build cost" for the
  contractor's; fees exclude expenses unless stated
- Tone: plain and confident; never apologise for a fee

Each line corrects something assistants do by default. Without the banned words, the first draft of an extension proposal calls the kitchen "a stunning, light-filled space" and promises a "design journey". Without the fee-versus-build-cost rule, drafts drift between the two, and clients come away unsure what they're being asked to pay you. The stage names matter because a client comparing your proposal with your terms should find the same words in both.

For drafting straight from your own files in Word, see Copilot in Word: draft proposals from your own files.

From first meeting to written brief

Record or type notes during the first meeting, add the client's answers to your pre-meeting questionnaire, and ask for a brief under your template's headings:

Using our brief template, write a client brief from the meeting
notes and questionnaire below.
- Separate "requirements" (the client clearly wants this) from
  "aspirations" (mentioned as maybe, one day, if budget allows).
- Quote the client's words for anything about how they want the
  space to feel.
- List open questions we must answer before the fee proposal.
- Budget: record only what the client said, word for word.
Do not add requirements that weren't mentioned.

Notes and questionnaire: [paste]

An illustrative extract for a couple planning a rear extension to a 1930s house:

Requirements
- Open-plan kitchen and family area, seating for 8
- Utility room with a side door to the garden
- Retain the existing front reception room as a quiet room
Aspirations
- Rooflights over the dining area ("we want it to feel bright
  even in winter")
Budget
- "Somewhere around 150, but we'd stretch for the right design"
Open questions
- Home office: required now, or future?
- Is the rear boundary wall shared, and in what condition?

What needed fixing: the first version of this output listed "home office" under requirements. The client had actually said "maybe a home office one day, if the kids ever leave". The prompt's requirements-versus-aspirations rule caught it on the second run, and it moved to open questions. That distinction matters later: if a home office is in the brief as a requirement and not in the fee proposal's scope, you've set up an argument.

Send the brief to the client to confirm before quoting. A brief the client has agreed in writing is the most useful document you'll have if the scope is ever disputed.

Watch the budget line at the next step too. When the proposal's "understanding of the project" section is drafted from the brief, assistants like to tidy, and "somewhere around 150, but we'd stretch" becomes "a confirmed construction budget of $150,000". The client said neither "confirmed" nor what the figure includes, and a proposal that restates it as fixed invites the question, months later, of why the tender came in at 180. Keep the client's words in quotation marks, and add in your own voice what the figure does and doesn't cover once you've discussed it.

Fee proposals: what AI drafts and what it doesn't

The relevant-experience row earns its warning. Asked to "add a short relevant-experience section", an assistant with the practice's three sample proposals in its project produced, in one illustrative run: "We have extensive experience of sustainable residential design, including several certified low-energy extensions." The practice had designed one extension to a high insulation standard and none that were certified. Nobody had asked for sustainability credentials; the model inferred the client would value them from a line in the brief about lower heating bills. Give it a list instead (project type, year, one line on what was relevant) and a rule that it may use only projects on the list. Any claim you can't trace to the list comes out.

Programmes go wrong more quietly. An illustrative AI draft: "Planning approval will be obtained within eight weeks of submission, with works starting on site in month five." The practice can't promise what a planning authority decides or when, and "month five" assumes a contractor is free. After the architect's edit: "We'll aim to submit the planning application within six weeks of your approving the concept design. The planning authority's decision period starts once the application is validated; we'll confirm the expected date then and tell you if it's extended. The start on site depends on the tender and your chosen contractor's availability." Less exciting, and far less likely to be quoted back at you in month six.

The fee comes from hours, not from a language model. A simple resourcing sheet for the illustrative extension, at a blended rate of $95 an hour:

Once the proposal is drafted, the most valuable AI step is a cross-check between the proposal and the agreed brief:

Compare this fee proposal with the agreed brief. List:
1. Anything in the brief (requirements, aspirations or open
   questions) not covered by the scope or the exclusions.
2. Anything in the scope not asked for in the brief.
3. Any number in the proposal text that doesn't match the fee
   table.

An illustrative result: "Brief aspiration 'rooflights over dining area' is in scope. Brief open question 'rear boundary wall' is not addressed in scope or exclusions. Proposal text says 'five site visits'; fee table says six." Two catches in a minute, both of which would otherwise have surfaced during the job, when they cost money. For the wider habit of checking AI-drafted numbers, see catching made-up figures in AI-drafted proposals.

Checking your quotes against what jobs actually took

The resourcing sheet is only as good as the hours in it, and most small practices estimate from memory. If you record time by stage, export the timesheets for your last few completed jobs of one type and ask AI to compare quoted hours with actual hours:

Attached: quoted hours by stage and actual timesheet hours by
stage for our last six residential extensions. For each stage,
give quoted total, actual total, and the difference as a
percentage. Then list any stage that overran on four or more
of the six jobs. Show your working in a table so I can check
the sums.

An illustrative finding: technical design overran on five of six jobs, by an average of about a third, while planning drawings came in slightly under. That's a pattern worth acting on: raise the technical-design hours in the resourcing template, and look at why. Often it's late client changes that were absorbed rather than charged, which the variation letter below is designed to stop. Check the table's sums in a spreadsheet before changing any template; the prompt asks for working precisely because arithmetic is where language models slip. This is also the honest answer to "are we charging enough?": your own completed jobs are better evidence than anything a chatbot can offer about other practices.

Exclusions that prevent fee disputes

Most fee disagreements come from something neither side wrote down. Ask AI to tailor your standard exclusions list to the project, adding anything the brief makes likely. For the extension, a tailored list might include:

This fee excludes:
- Structural engineer's design and calculations (we'll
  coordinate with your appointed engineer)
- Measured survey: we'll work from the survey you supply
- More than two rounds of changes at concept stage; further
  rounds charged at $95/hour
- Planning application fees and any specialist reports the
  planning authority requests (for example, trees or drainage)
- Work to the rear boundary wall, pending the open question in
  the brief
- Interior design, furniture and loose fittings
- Additional site visits beyond six, at $95/hour plus travel

Check each line against what you actually intend to do. AI will happily add plausible exclusions you don't normally use, and a long list of exclusions you don't enforce reads as defensive. Shorter and specific beats long and generic. If you also want to say how your practice uses AI in client contracts, mentioning AI use in contracts and proposals has suggested wording.

Variation requests: the email that decides whether you get paid

Clients rarely say "I'd like to commission additional services". They say something like this:

"Could you just quickly redo the kitchen layout with the island on the other side? And while you're at it, could you have a look at adding a garage at the side? Shouldn't be much work."

Paste the request, the agreed scope and the job's record of changes so far, and ask AI to classify each part and draft a reply. An illustrative draft after the architect's edit:

"Happy to look at both. Moving the island is within the concept-stage changes in our agreement; you have one round of changes left, so we'll include it in that. The garage is a new element that isn't in the current brief or scope. It affects the planning drawings and the structure, so I'll send a short additional-services note with the extra fee and time before we start on it. Would you like the island change done now, or held so both can be reviewed together?"

The architect checked one thing before sending: whether a concept-stage round really was left. The job record showed it was. Getting that wrong in the client's favour is a lost fee; getting it wrong against them is a lost client.

The additional-services note then needs a number, and it comes from the same resourcing sheet. For the garage, illustratively: 6 hours of concept design for the new element, 10 hours revising the planning drawings, 12 hours of technical design and one extra site visit at 5 hours, 33 hours in all, or $3,135 at the practice's $95 blended rate. AI can write the note around that figure and list what it covers. The hours are the architect's estimate, checked against how long similar additions took on past jobs.

Minutes, site reports and the rest of the paperwork

• Site visit reports. Dictate a voice memo walking round the site, then ask AI to structure it into your report headings: attendees, progress by area, issues with photo references, instructions given, actions with owners. Check every instruction; a site report is an instruction record, and a misheard "don't" changes everything.
• Design team minutes. From a transcript or rough notes, with an action table. Circulate within 24 hours.
• Consultee and planning comments. Paste lengthy comments and ask for a table of each point, whether it needs a design change, and a draft response. The architect decides the responses.
• Client update emails. A weekly two-paragraph update during construction, drafted from the week's site notes.

A site report shows why the instruction check matters. Say the architect's voice memo includes: "Contractor to hold off boarding the utility ceiling until the electrician's first fix has been inspected, and the rooflight upstand needs re-levelling before the roofer comes back." An illustrative structured report listed "Contractor: board utility ceiling" and "Electrician: first-fix inspection" as two separate actions, and dropped "before the roofer comes back" entirely. The pieces of each instruction survived; the order between them, which was the whole point, didn't. Add a line to the site-report prompt: "Keep any condition or sequence (until, before, after, not until) inside the same action, word for word." Then read the actions column against the memo before the report goes out.

Specifications deserve caution. AI can tidy the format of your own specification clauses, but generating technical clauses from scratch is where confident errors live: wrong standards, superseded product references, clauses that sound right and aren't. Treat any AI-generated technical content as unchecked until an architect has verified it against current regulations and manufacturers' data.

What it adds up to for a three-person practice

For a three-person practice doing mainly residential extensions and small commercial fit-outs, these are illustrative figures for a typical month: six fee proposals, four briefs, ten site visits and fortnightly design team meetings on two jobs. Before AI, a proposal takes about three and a half hours, a brief two hours, a site report forty minutes and a set of minutes an hour. With the practice project set up and the review steps kept, realistic figures are about ninety minutes per proposal, forty minutes per brief, fifteen minutes per site report and twenty minutes per set of minutes.

That's roughly 12 hours saved on proposals, 5 on briefs, 4 on site reports and 2.5 on minutes: about 23 hours a month, more than three working days. On cost, the Business Standard plus Copilot bundle for three people is $70.50 a month on annual billing, and it replaces a Business Standard subscription the practice would probably pay for anyway. The saving holds only if two things stay manual: the fee calculation and the final read of anything with an instruction or a number in it. For where else AI fits in a small practice, where AI saves time in a small architecture practice takes the wider view, and copyright and client data risks covers what to keep out of AI tools.

Questions practices ask about AI paperwork

Should we tell clients we use AI to draft proposals and briefs?

It's sensible to mention it in your terms or proposal, briefly: AI assists with drafting, and an architect reviews and takes responsibility for everything issued. Some clients, particularly commercial ones, have policies on suppliers' AI use and will ask. Saying so up front avoids an awkward question later and costs nothing.

Can AI help with fee benchmarking against other practices?

Not reliably. General assistants don't have trustworthy data on what other practices charge for a given project type, and anything they produce will be a plausible-sounding guess. Benchmark from your own completed projects: hours actually spent per stage against what you quoted. That comparison is where AI can help, by summarising your timesheet exports.

Is it safe to put clients' house plans into an AI tool?

Only into a business-plan tool your practice has approved, where content isn't used for training and you control retention. Drawings of private homes can reveal security details, such as alarm positions or entry points, so treat them as confidential. Don't upload them to free consumer accounts, and remove them from projects when a job closes.

Further reads

• How Much Does AI Cost a Small Architecture Practice per Month? — What this set-up costs a small practice each month.
• Best AI Rendering Tools for Small Architecture Practices (2026) — The visual side of AI in a small practice.
• How to Set Up an Approval Step for AI-Written Quotes — An approval step before any AI-drafted proposal goes out.
• Claude vs ChatGPT for Business Writing, Proposals and Reports — Which assistant writes better proposals and reports.
• How to Turn Meeting Notes Into Tasks Automatically With AI — Turning minutes into tracked actions automatically.
• AI Renders vs Outsourced Visualisation: What Practices Save — When to render in-house and when to outsource.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business, Claude Team and Microsoft 365 Copilot Business pricing and plan terms, checked September 2026.