Every architect knows the sequence. An Instagram DM or a contact form arrives: "hi, how much for a design?" You reply and ask for details. The person sends a three-minute voice memo, disappears for four days, then comes back asking for "just a ballpark." By the end of the month there were 14 conversations like that, three became meetings and one signed. The other eleven ate about six hours of your time.
That is the point where using ChatGPT for architects stops being a toy and starts making sense as a system. Not to "talk to clients for you" with a cheerful bot, but to do three boring things: collect the right information, organize it into a program you can read in two minutes and draft the fee proposal from your own rules. You only step in to decide and send.
The full flow, from first contact to proposal
- Entry. Someone reaches out by Instagram, the website or email. An auto-reply (Instagram's saved replies, an email autoresponder or your website form's confirmation, no AI involved) sends a link to a 12-question questionnaire and sets the expectation: "you will get a proposal within two business days."
- Questionnaire. Tally or Google Forms. It asks property type, square footage, city, owner or renter, condo or single-family, what they want to change, timeline, construction budget range, who makes the decisions, plus a field to upload plans and photos. An optional voice-memo field helps people who hate typing.
- AI organizes it. When the form comes in, Make, Zapier or n8n sends the answers to the OpenAI or Anthropic API with a fixed prompt. The AI returns the structured program, the contradictions, the missing questions and a lead grade.
- Proposal draft. Using your pricing rules (per square foot, per phase, firm minimum), the flow fills a Google Docs template: scope, phases, dates, fee and terms.
- You review. You get an email or Slack message with the program and a link to the draft. You adjust the fee if needed and send it yourself.
Notice what is not in the flow: the AI does not chat freely with the client and never sends a proposal on its own. You could set that up, but for an architecture practice I would not. A design client is buying trust in you, and one wrong answer to "can we take this wall out?" costs more than the hours you saved.
Real input: what the questionnaire received
This is a case assembled from typical inquiries, with details changed. Answers as they arrived, typos included:
Property: condo 1050 sqft, 2 bed 2 bath, building from 1974, Chicago, Lakeview
Own or rent: own, closed last month
What do you want to change: open kitchen to living room, primary bath is awful, same floor everywhere, built ins in bedrooms. my husband wants a spot for a home office
Timeline: want to be done by june
Construction budget: around 165k all in
Who decides: me and my husband
Voice memo (transcribed): "...we have a 14 month old so it has to be practical, no sharp corners... and I cook a lot but I don't want the smell in the living room, you know... the building manager said there's some paperwork for renovations..."
Attachments: original floor plan from the listing (PDF), 11 photos
The prompt that runs inside the flow
It is a fixed prompt, always the same. The rules section is what keeps the AI from making things up.
You organize client questionnaires for a small architecture firm that does residential renovations and interiors in the United States. You receive form answers (sometimes with a transcribed voice memo) and return ONLY the format below. RULES - Do not invent information. If it is not in the answers, write "not provided". - Give no technical opinion on structure, demolition, code or zoning. - Point out contradictions between requests, budget and timeline, with numbers. - Grade the lead: A (clear scope, budget in range, feasible timeline), B (missing information), C (budget or timeline out of range). FORMAT 1. Summary in 3 lines 2. Program (list by room) 3. Constraints (property, building or HOA, household, timeline, budget) 4. Contradictions or risks (with the reason) 5. Questions for the first meeting (max 6, most important first) 6. Lead grade and why, in 1 sentence 7. Suggested scope for the proposal (service phases, no fee) FORM ANSWERS: [paste here]
Real output: the program that lands in your inbox
I ran the prompt above on the case. Output, unedited:
1. Summary: Couple with a 14-month-old bought a 1,050 sq ft condo (2 bed, 2 bath, built 1974) in Chicago and wants a full renovation: open kitchen, new primary bath, continuous flooring and built-ins. Stated all-in budget of US$ 165,000 and a June completion target. Decision makers are the couple.
2. Program:
Kitchen: open to the living room; client cooks often; odor control needed.
Living room: connected to kitchen; continuous flooring.
Primary bath: full gut and rebuild.
Bedrooms (2): built-ins; home office in one of them or elsewhere.
General: same flooring throughout; child safety (no sharp corners).3. Constraints: 1974 building (plumbing and electrical possibly original); condo association has a renovation approval process; whether the family will live in the unit during construction: not provided; completion target June.
4. Contradictions and risks:
Open kitchen vs. "no smell in the living room": requires an exhaust solution or partial separation.
Timeline: with design, association approval, permit and full construction, June is tight unless design starts now.
Budget: US$ 165,000 for flooring, kitchen, bath and built-ins in 1,050 sq ft is workable at a mid-range finish level but tight if built-ins are extensive. Unclear whether it includes design fees and appliances.
Kitchen wall: unknown whether it is load-bearing or carries building risers.5. Questions for the first meeting:
1. Will you live in the unit during construction?
2. Does the US$ 165,000 include appliances, built-ins and design fees?
3. Does the association allow a new exterior vent for a range hood?
4. Is the home office for daily use or occasional?
5. Does the building manager have the original drawings or riser diagrams?
6. Is June a hard date (lease ending, a baby due) or a preference?6. Lead grade: A, because scope and budget are clear and the timeline is possible with an immediate start.
7. Suggested scope: existing conditions survey and measured drawings; schematic design with one revision round; design development; construction documents with millwork details; permit and association submittal support; construction administration (to be defined).
My read: questions 1 and 6 are the ones I would have forgotten. If the couple plans to live through a gut remodel with a toddler, the whole schedule changes. And if June is the end of a lease, that becomes a clause in the proposal. The A grade is right. The AI also kept its hands off the wall, as the rule said, and listed it only as a risk.
From lead grade to fee proposal
The proposal is where many architects want AI to "calculate the price." Do not do that. The fee comes from a rule you write once, and the AI only applies it. Build the rule from your own data: your hourly cost, the hours your last five similar projects actually took per phase, and what your local market accepts. AIA surveys on firm compensation can give you context, but your timesheets are the real reference. An example of the rule you hand to the flow:
- Residential renovation design, fixed fee: US$ 16 per sq ft, firm minimum US$ 9,000.
- Millwork details included; construction administration billed separately at US$ 185 per hour, site visits capped at 2 hours each.
- Payment: 25% at signing, 35% at end of schematic design, 40% at issue of construction documents.
- Timeline: schematic design in 15 business days after the survey; CDs in 25 business days after schematic approval.
For the case: 1,050 sq ft × US$ 16 = US$ 16,800 in design fees. The AI fills the template with scope, exclusions (structural engineering, MEP engineering if required, permit fees, purchasing), phases, calculated dates and one paragraph on approvals: who prepares the permit drawings, that the association's alteration agreement and the contractor's insurance certificate must be in place before work starts, and that you will respond to plan review comments. That paragraph alone shows the client they are talking to someone who knows the process.
If you use an AIA agreement such as B105 for small residential work, the flow can generate the proposal letter and an exhibit with the scope in plain English, while the agreement itself stays the licensed AIA document. The numbers above are an example. Every firm has its own, and those are what the flow should use.
Tools and monthly cost
| Piece | What I would use | Approximate cost |
|---|---|---|
| Auto-reply | Instagram saved replies, an email autoresponder or the website form confirmation | Free |
| Questionnaire | Tally (file uploads on the free plan) | Free; paid plan, check the site |
| Automation | Make (easiest), Zapier (most integrations) or n8n (most flexible, can run on your own server) | Make has a free plan and paid plans from about US$ 9/month; Zapier from about US$ 20/month; n8n Cloud from about €24/month; check the sites |
| AI | OpenAI or Anthropic API | A few cents per program; 30 a month stays under US$ 2 |
| Proposal | Google Docs template with merge fields | Free with a Google account |
Realistic total: from zero to about US$ 40 a month, depending on volume and which automation tool you pick. Setup takes an afternoon if you have used Make before, a weekend if you have not.
Make, Zapier or n8n? For a one- or two-person firm I would go with Make. The interface is clear and errors show up on screen, and it costs less than Zapier for the same flow. Zapier makes sense if you already pay for it. n8n is worth it when you have someone technical nearby or want client data on your own server.
Why I would not put an AI chatbot in your DMs
You can connect an agent that chats on its own through Instagram, a website widget or SMS. For architecture, I would avoid it for three reasons:
- Volume is low. A firm gets dozens of inquiries a month, not thousands. The questionnaire handles it.
- The risk of a wrong technical answer is high, and it lands under your name and your license.
- Prospects who reach an architect often want to feel heard by a person. A bot that answers "great question!" to a structural concern loses the job.
If you grow into a team with real volume, that conversation changes.
When this automation is not worth building
- You get fewer than five inquiries a month. Do it by hand with the prompt above, pasting the answers into ChatGPT.
- Your projects are large, few and always start with an in-person meeting. Programming happens at that meeting.
- Your clients come through a builder who already hands you a defined scope.
Data and licensing cautions
The questionnaire collects names, phone numbers, addresses and floor plans. Add one line to the form explaining what the data is used for and how long you keep it, and send the AI only what it needs: the prompt works fine without the name, phone and unit number. Over the API, the major providers state that they do not train on submitted data by default; check the current policy before you switch it on. If you have clients in states with consumer privacy laws, such as California, a short privacy notice on the form is a good habit anyway.
As for your license: the proposal is yours, the fee is yours and the seal on the drawings is yours. The automation only makes all of it faster and more organized. Your state board will not care which tool drafted the email, only whether what you sent was accurate.
Common setup mistakes
- A questionnaire with 30 questions. Nobody finishes it. Twelve is the limit, and only the ones that change the fee should be required.
- A prompt with no "do not invent" rule. The AI fills gaps with guesses and you do not notice.
- Sending the proposal straight out, unreviewed. The first time a wrong scope goes out, you lose the client and your trust in the flow.
- Not measuring. Write down how many proposals went out and how many signed, before and after. If the rate did not move, the problem was somewhere else.
For the rest of the practice (renderings, specs, marketing), see the full map in AI for architects and the standalone prompts in AI prompts for architects. And if you would rather not build any of this, the ready-made agents run this whole flow in your format.

Want to skip this part?
Ready-made AI agents for architects: nothing to build, no prompts to write, they work for you from the first inquiry to the spec.
- Takes the inquiry and returns an organized program, with the questions still missing
- Drafts the fee proposal using your own pricing and timeline rules
- Writes the outline spec from your finish schedule and tags what you need to verify
Step by step
- Set up an auto-reply with the questionnaire link and the proposal turnaround time.
- Build the questionnaire in Tally with 12 questions and upload fields for plans and photos.
- In Make, connect the form to an OpenAI or Anthropic module with the fixed intake prompt.
- Write your fee and timeline rules once and use them to fill a proposal template in Google Docs.
- Get the notification with the program and draft, review the fee and send it yourself.
- After 30 days, compare proposals sent and signed against the month before.
Frequently asked questions
Is there a ready-made AI agent for architects?
There are standalone tools for renderings and text, but an agent that handles intake and proposals in your firm's format is usually built with a form, Make or n8n and an AI API. The ready-made agents from Onne Company are on a waitlist.
Can ChatGPT calculate my design fee?
It should apply a rule you define (per square foot, per phase, firm minimum) based on your own hourly cost and timesheets. Letting AI invent the fee gives you generic ranges scraped from the internet.
Do I need a chatbot on my website or Instagram?
Not for this flow. An auto-reply with the questionnaire link is enough. A chatbot that talks to prospects on its own adds the risk of wrong technical answers under your name.
How long does it take to set up?
An afternoon if you have used Make before, a weekend if you have not. What takes the most time is writing your pricing rules and the proposal template.
Is it safe to send a client's floor plan to AI?
Send only what is needed, without name, phone or unit number, and say on the form what the data is used for. Over the API, check the provider's data use policy before going live.
