ChatGPT for architects: what it does well on a project, and what never to paste into it
Most architects who use ChatGPT use it for the half of the job that is words: the client update after a site visit, the plain-English version of a design decision, the summary of a sixty-page consultant report. This guide covers what the published cheat sheets and the AIA's own guidance say about that, prompts for client work, what to keep out of it, and the one record it does not hold: what each client said, approved and sent across a project.
This is the ChatGPT chapter of our AI for architects guide, which maps the three kinds of tool a practice meets (rendering and concept tools, assistants inside design and BIM software, general assistants) and what the AIA's AI Task Force has said about them. Here the question is narrower: you have ChatGPT, or you are weighing it, and you want to use it on real project work without regretting it.
Two disclosures first. This guide is published by Equerry, a personal assistant for the client side of a project, and Equerry does not design, draw, render or read a model. It comes in at the end, where it fits. And nothing here is professional or legal advice: what may go into any tool on a project is decided by your firm's policy, your client agreements and your licensing board, not by a vendor. Every statement about ChatGPT comes from OpenAI's own help pages, and every statement about the AIA from its own publications, all read 2026-10-27.
What architects use ChatGPT for, from the published cheat sheets
The guides that rank for this question agree on the shape of the work, and they are worth naming because each adds a caveat the others skip.
- Architizer's "Chat GPT Cheat Sheet for Architects and Designers" (by Architizer's Paul Keskeys; undated, read 2026-10-27) is an infographic of starting points, and its text sets the tone for the genre: "What Chat GPT provides will rarely be the finished product, so use it as a starting point." Its companion piece gives the formula the sheet is built on: "[Introduction or context] + [Specific question or instruction] + [Optional additional details or constraints]."
- AIA New Jersey's "How to Train and Use GPTs for Architectural Tasks" (Brian W. Penschow, AIA, and Architect GPT (ChatGPT), November 6, 2024) is about custom GPTs rather than the plain chat, but the tasks are the same: summarising briefs and reports ("GPTs can be prompted to summarize these documents, highlighting key objectives, constraints, and priorities"), drafting "professional, clear email responses," early research on zoning and materials. Its two warnings are the ones to remember: "Architects must be cautious not to input sensitive project information into GPTs, especially in open-access models," and "GPTs cannot replace official documentation or local code expertise."
- MyArchitectAI's "ChatGPT for Architects: Best Prompts & Use Cases (2026)" (undated beyond the title) runs through concepts, documentation, software help, client communication and admin. Its client-side examples are the practical ones: "Write a concise email update to a client summarizing today's site visit based on the attached notes," and a meeting transcript to "Summarize the key takeaways and turn them into a to-do list grouped by team member." Its caveat: "ChatGPT gets a lot right—but not everything."
- The AIA's own continuing-education course, "ChatGPT for Architects" (AIAU, instructor Nikita Morell), is narrower than its title: its objectives are about marketing and client engagement, "crafting effective ChatGPT prompts" and "firm-voice alignment," with "a downloadable ChatGPT Prompts for Architects guide with 100+ ready-to-use examples."
OpenAI's help center (article "Uploading files and audio to ChatGPT," read 2026-10-27) confirms the mechanics: ChatGPT "can work with documents, spreadsheets, presentations, images, and audio files you attach to a conversation"; PDF, DOCX and TXT are supported document formats, images are limited to 20 MB each, and audio uploads "are available with paid ChatGPT subscriptions and workspaces." Its examples include "Summarize a meeting recording, highlighting decisions, open questions, and next steps," with the caveat that "Transcripts may contain errors, and speaker identification may be unreliable."
| Job | What the published guides have it doing | What to check before it leaves your desk |
|---|---|---|
| Client correspondence | Site-visit updates, proposal text, "those awkward follow-ups," matched to your writing style | Every date, figure and commitment. It is fluent at changing a promise while tidying a sentence. |
| Plain-English summaries | A consultant report, a code section you pasted, a long email thread | That each point comes from the text you gave it, not from the model's general idea of what such a document says |
| Meeting notes and agendas | A recording or transcript into decisions, open questions and a to-do list | Who said what, and every figure in the recap, against your own notes. |
| Code and product research | Early orientation on zoning, materials and compliance | The source itself. The AIA New Jersey article: "GPTs cannot replace official documentation or local code expertise." |
Where it fails is the same place in every guide. Its answers sound equally sure whether right or wrong, and in architecture the costly wrong answers are a code clause, a product value or a dimension that it stated rather than read. MyArchitectAI's building-code FAQ says it plainly: ChatGPT is "not reliable enough to be a substitute for a licensed architect, engineer, or code official who has legal authority in your jurisdiction." And it checks nothing against the drawings: the same guide says it "doesn't have the necessary settings to produce house plans, elevations, or layouts that make any technical sense." For ChatGPT for architecture work, then, it is a drafting and summarising assistant for the words around the design, not a reader of the design.
ChatGPT prompts for architects' client work
Good prompts share three habits: a narrow job, no client or site identifiers, and output you can check in a minute. Fill every bracket with a placeholder, not the real detail.
- Site-visit note into a client update. "Here are my notes from today's site visit, names and address removed: [paste]. Draft a client update under 200 words. Do not change any date, quantity or commitment. After the draft, list every change you made to my wording."
- Explain a design decision. "Explain to a client with no technical background why the [stair] moved to the [north wall] in a [two-storey extension]. Under 150 words, no jargon. Flag anything that depends on a code or planning rule so I can confirm it."
- Summarise a consultant report you pasted in full. "Summarise the [structural engineer's] report below in ten points. For each point, quote the exact sentence it comes from. Do not add anything that is not in the text."
- Tighten a question to a consultant. "Rewrite this question to the [mechanical engineer] so it can be answered yes or no, and list the two pieces of information they will need from me: [paste]."
None of these asks for a code conclusion, a product specification, a dimension, or a document to send unread.
What the AIA's AI Task Force has said
The American Institute of Architects has published three things that bear on how you use ChatGPT, and the practical one is the least quoted. All read 2026-10-27; this page quotes them and does not conclude what the AIA permits. That is for your firm, your contracts and your board. On dates: both PDFs are linked, undated, from the AIA's AI in Architecture page; the Task Force's April article says they were "published in January 2026"; the Guidance PDF is marked "Last updated: 10/22/2025."
The Position Statement on AI is the short one: the AIA "encourages all members to thoughtfully explore and engage with AI as a powerful tool," and names complexities "including concerns around environmental impact, authorship, data privacy, and workforce transformation."
The Guidance for the Responsible Use of AI by Architecture and Design Firms is the practical one: three pages under nine headings, and the lines below are this page's "never paste" section in the profession's own words:
- Under "Maintain Professional Responsibility": "AI is a design support tool—not a replacement for professional judgment," and "All AI-generated outputs must be reviewed and validated by qualified professionals."
- Under "Protect Data, Privacy & Intellectual Property": "Do not upload confidential client or firm data into unvetted AI systems," and "Understand how external AI platforms store, reuse, or train on uploaded information."
- Under "Address Transparency, Bias & Equity": "Disclose to clients and stakeholders when and how AI is being used in project workflows."
- Under "Establish Firm-Level AI Policies and Oversight": internal policies addressing "Approved AI systems and use cases" and "Data privacy and confidentiality protection."
Its table of "Emerging AI Use Opportunities for Firms" lists "Specs drafting, code summaries" under Project Delivery and "Proposal creation, marketing personalization" under Operations & Business: the same words-side work the cheat sheets describe.
The Task Force's article "Architects and AI: Practical guidance for a changing profession" (April 6, 2026) is the narrative. On what firms are doing now: "Firms are streamlining code research and documentation checks, improving model coordination and clash detection, and automating repetitive drafting tasks, each freeing practitioners to spend more time on the work that requires their judgment." On where that leaves the architect: "as far as can be seen, the architect will remain the professional of record. Human oversight and professional accountability are the critical factors that make AI useful and less risky." Our practitioner's map quotes the same article on the wider landscape.
Rendering GPTs and design assistants are a different category
A search for ChatGPT for architects returns tools with "GPT" in the name that are not ChatGPT. ArchiGPT, for example, describes itself on its own site (read 2026-10-27) as "The World's First AI Design Assistant for Architectural Renders," with prompt starters such as "Generate render from elevation" and "Apply texture to floor plan," sold on its own plans from $19 a month billed monthly. It is a standalone rendering service; its page says nothing about any relationship to OpenAI, so this page draws no conclusion either. Custom GPTs built inside ChatGPT, the subject of the AIA New Jersey article above, are a third thing: ChatGPT with instructions and reference files attached by whoever built it, so that article's confidentiality warning applies to them too.
Rendering tools, and the assistants built into design and BIM software, are covered with examples in AI for architects; this page does not repeat the list.
What never to paste, and the settings that change what OpenAI keeps
This is a working default, not a rule of professional conduct, and your firm's policy overrides it. The test before you paste: could this identify a client, a site or an unreleased project?
Usually raises fewer questions:
- Public material you retrieved yourself: a code section, a planning policy, a published standard, a manufacturer's public datasheet.
- Your own drafts with every name, address, figure and distinguishing fact replaced by placeholders such as [Client], [Site], [Budget].
- Work with no client in it: a competition entry, a talk, a practice bio.
Keep out, unless your firm has decided otherwise for a specific tool and plan:
- A client's name together with the site address or the project.
- Budgets, fee proposals, contract terms and negotiating positions.
- Unreleased drawings, models, specifications and reports, including consultants' work you hold under their terms.
- Personal details of homeowners, tenants, neighbours and objectors.
The settings, in one sentence: on a personal plan, OpenAI's help center (read 2026-10-27) says you opt out of training by turning off Improve the model for everyone under Settings > Data controls, review or switch off memory under Settings > Personalization > Memory, and keep one conversation out of memory by starting a temporary chat and choosing Unpersonalized before the first message, while "By default, OpenAI does not use content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces to train its models." Which of those is enough for your firm is a different question; is ChatGPT confidential? sets out what OpenAI publishes plan by plan. One more input deserves its own decision: a connected mailbox. OpenAI's Memory article lists "content from connected apps, such as Gmail" among the sources Memory can draw on, and a project inbox is mostly other people's confidences.
What ChatGPT does not hold: what each client said and sent across a project
ChatGPT remembers, so the question is not whether it forgets but what reaches it, and how. On a personal plan it learns about your projects from what you type, paste or upload, and from any account you connect. The first route puts you in charge of every input and goes quiet the moment you are on site and the client's reply moves the budget; the second keeps up by taking the whole account, which is the decision the "never paste" list is about. And what it keeps is, in OpenAI's words, a summary that "does not necessarily include every detail or source that Memory can reference" (read 2026-10-27).
The record a project actually runs on sits in between: which cladding sample did the client choose at Tuesday's review, did the structural engineer answer the question about the transfer beam, which revision did I promise the contractor by Friday? It arrives across email threads, a call in the car and ten minutes on site, and the version that matters is the one that moved last.
Equerry is a personal assistant on your iPhone who helps you get things done, and for an architect that means that record, not the design. It does not read drawings, models or BIM files, and it connects to no design software. Nothing reaches it unless you send it.
| What you send | The one thread with the client or the engineer, by adding Equerry's own email address at mail.getequerry.app to the CC line or forwarding the single email that matters; replies that keep the address on the thread keep arriving. A voice note you record in Equerry walking off site: what was agreed, what the client pushed back on, what you owe. The engineer's PDF comments or a photo of the marked-up plan, from your iPhone's share sheet. A typed note after the review when that is quicker. |
|---|---|
| What you ask | "Which cladding sample did the client choose, and when?" "Has the structural engineer answered about the transfer beam?" "What did I promise the contractor after Thursday's site visit?" |
| What you get back | The answer, in plain words, from what you sent. When an answer draws on a note you saved, it carries a source chip: tap it to see the passage Equerry used and open the original. For a thread you CC'd or forwarded: What Equerry remembered from it is listed in Settings → Your memory, each line marked as from that email, and you can delete any line. If you allow calendar access and set up Meeting Prep, a short brief on where things stand with that client arrives before the review, built from what you have shared. |
That is the difference in kind. An account connection hands over the mailbox and leaves the assistant to decide what matters; CC'ing a thread hands over that thread, and what Equerry holds about a client is the set of threads, notes and files you decided to send.
The honest limits. Equerry does not open drawings or models, does not connect to design, BIM, project-management or accounting software, does not join or record your meetings, and does not send email on your behalf; it cannot find the thread you never forwarded. Anyone on a thread can see its address on the CC line; forward instead where that would be awkward. Sending something is still sharing it: what you send is processed by the outside providers named on our AI disclosure page, so your firm's policy, and the AIA line about unvetted systems, apply to Equerry exactly as they do to ChatGPT. What changes is the scope of the decision, from an account to the specific threads, photos and notes you chose. Equerry Pro is $39.99 a month or $329 a year.
So: ChatGPT for the words around the design, with the settings above and nothing identifying in the paste; Equerry for what each client said, approved and sent. For the note-taking habit underneath both, read how to take meeting notes; for the ten minutes before a review, meeting prep.
Sources, all read 2026-10-27. The American Institute of Architects: Architects and AI: Practical guidance for a changing profession (AI Task Force, April 6, 2026); Position Statement: Artificial Intelligence (AI) (PDF); Guidance for the Responsible Use of AI by Architecture and Design Firms (PDF, "Last updated: 10/22/2025"; both documents are linked from the AIA's AI in Architecture page); ChatGPT for Architects (AIAU course page). AIA New Jersey, How to Train and Use GPTs for Architectural Tasks: Basics and Beyond (Brian W. Penschow, AIA, November 6, 2024). Architizer, Chat GPT Cheat Sheet for Architects and Designers and Chat GPT for Architects and Designers: Prompt Formula and Working Examples (both undated). MyArchitectAI, ChatGPT for Architects: Best Prompts & Use Cases (2026). ArchiGPT, archigpt.ai (prices in US dollars as shown). OpenAI Help Center, read in a browser session (the help center refuses automated fetches): Memory in ChatGPT (sources, settings path, memory summary, Sources, temporary chat, training); How your data is used to improve model performance (the training setting and its path); Uploading files and audio to ChatGPT (file types, limits, audio availability, transcript caveat). Vendors and institutions change their products and pages often; where a current page disagrees with this one, that page is right. This guide is general information, not professional or legal advice.
Frequently asked questions
Can architects use ChatGPT?
Many do. What the AIA has published addresses how a firm uses these tools: its responsible-use guidance (read 2026-10-27) asks firms to set policies naming "Approved AI systems and use cases," says "All AI-generated outputs must be reviewed and validated by qualified professionals," and says "Do not upload confidential client or firm data into unvetted AI systems." Whether a given use is allowed on a given project is set by your firm's policy, your client agreements and your licensing board, not by a vendor or by this page. The AIA section below quotes the documents.
What does the AIA say about architects using AI?
Three things, all read 2026-10-27 and quoted in the AIA section below. The Position Statement on AI "encourages all members to thoughtfully explore and engage with AI as a powerful tool" and names "data privacy" among its concerns. The responsible-use guidance for firms has nine areas, including "Protect Data, Privacy & Intellectual Property." The AI Task Force's April 6, 2026 article adds: "Human oversight and professional accountability are the critical factors that make AI useful and less risky." This page quotes that material and does not conclude what the AIA permits.
Can ChatGPT read my drawings?
As pictures and documents, not as a model. OpenAI's help center (read 2026-10-27) says ChatGPT "can work with documents, spreadsheets, presentations, images, and audio files you attach to a conversation," PDF among them, so a drawing set or a photo of a marked-up plan can be described and its notes summarised. It does not open a BIM model or check a dimension against the set; one published guide says it "doesn't have the necessary settings to produce house plans, elevations, or layouts that make any technical sense." Verify any reading, and check your firm's policy first.
What should an architect never paste into ChatGPT?
As a working default: a client's name with their site or project; budgets, fee proposals and contract terms; unreleased drawings, models, specifications and consultants' reports; and personal details of homeowners, tenants or neighbours. The AIA's responsible-use guidance puts it in one line: "Do not upload confidential client or firm data into unvetted AI systems" (read 2026-10-27). Public material you pulled yourself, your own drafts with placeholders, and work with no client in it raise fewer questions. The full list is in the never-paste section below; your firm's policy wins over any list.
Does ChatGPT remember what each client approved?
It can remember things from your chats, but not the way a project file does. OpenAI's Memory article (read 2026-10-27) says that when Memory is enabled ChatGPT "can remember relevant preferences and details from your chats and other available sources," which depending on plan and region can include "past chats," "files in Library" and "content from connected apps, such as Gmail." It also says "Memory does not retain every detail from every conversation," that the memory summary "does not necessarily include every detail or source that Memory can reference," and that "Sensitive information can appear in Memory if you share it with ChatGPT." So a client's approval reaches it only if you typed, pasted or uploaded it, or connected the account it sits in, and what it keeps is a summary it manages, not a record you can audit line by line. For a per-client record of what was said, approved and sent, see the last section of this page.
What are good ChatGPT prompts for architects?
The ones that give it a narrow writing or summarising job, keep client and site identifiers out, and produce something you can check in a minute. Examples: turning anonymised site-visit notes into a client update that changes no date or figure; explaining a design decision in plain language for a client with no technical background; summarising a long consultant report you pasted, with the exact sentence quoted for each point; and tightening a question to a consultant so it can be answered yes or no. The full set, with the wording, is on this page. None of them asks for a code conclusion, a product specification or a document to send unread.
Last updated 2026-10-28.
All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.