Architecture Firms Adopting AI Design: What to Tell Your PI Insurer

Often yes, or at least you should ask. In many markets, insurance law expects you to tell your professional indemnity insurer about facts that would matter to its decision, whether or not the proposal form asks about AI. Raise material AI use in design or calculation at renewal, and ask your broker in writing how your policy treats it.

This is a practical checklist, not legal or insurance advice. Policy wording varies, and so do the rules on disclosure where you practise. What does not vary is the sequence that protects you: know what AI you use, write it down, read the proposal form honestly, and get your broker's answer in an email you keep.

Follow me on Instagram@sagnikteaches

Why insurers care in the first place

A professional indemnity policy pays out when a client says your professional work caused them a loss and you are legally liable. The insurer is pricing the chance of that claim and the likely size. When a practice changes how it produces design, three things matter to that price.

Connect on LinkedInSagnik Bhattacharya

• Error profile. A tool that generates plausible but wrong output (an invented clause reference, a miscalculated area, a drawing that fails a code check) can create errors that look right and pass casual review.
• Review quality. If AI speeds up production but nobody checks more carefully, the same number of reviewers is looking at more output, which pushes up the chance that something slips through.
• Contract and data exposure. Uploading drawings, client documents or confidential briefs to third-party tools can breach confidentiality terms, and rights in generated material can be unclear.

The insurer does not need to forbid AI to care about these. It needs to know whether you have controls. That is why the useful thing to bring to a broker is evidence of your controls, not a general statement that you are "careful".

Subscribe on YouTube@sagnikteaches

How the duty to disclose usually works

I am describing common principles in plain words. Check your own policy and take advice.

• At the start and at each renewal you complete a proposal form or equivalent. You are expected to answer questions truthfully and completely.
• Beyond the questions, in many markets business insureds are also expected to disclose facts that a prudent insurer would want to know when deciding whether to cover you and on what terms; elsewhere the duty centres on answering the questions asked accurately. Which applies to your policy is a question for your broker, and whether a given use of AI qualifies is a judgement, which is why a written question is worth the effort.
• During the policy year, some policies require you to notify significant changes to the business, so look for a change-of-risk condition in yours. Moving from no AI use to routine AI use on design deliverables is the kind of change you should at least raise with your broker.
• If you get it wrong, the typical consequence is a dispute at claim time over whether the insurer would have covered you, or on what terms, had it known. One architects' regulator's renewal advice puts it bluntly: a failure to provide accurate information may result in future claims being rejected or the policy being invalidated. Claim time is the worst moment to learn the answer.

Do not wait for a question to appear on the form. A broker that specialises in the design professions wrote in November 2025 that no insurer was yet putting AI questions on proposal forms, and expected that to change. The topics it listed as likely are a useful preparation list: AI governance and strategy, tool selection and vendor due diligence, data management and cyber security, human oversight and professional competency, and contractual risk management. The same commentary noted that one insurer had begun applying an AI exclusion to some liability policies. A form that is silent this year may not be next year, and your answers will read far better if the records below already exist.

The practical message is that silence is a choice with a downside, and a short written exchange with your broker costs nearly nothing.

Sort your AI use into three tiers first

Before you speak to anyone, list every use of AI in the practice and sort it. Insurers care most about the middle tier, and the list lets you answer precisely.

For tier-three uses, you should be able to say what the tool does, who checks it, and how the check is recorded. If you cannot, that is the gap to close before renewal, regardless of what the insurer asks.

A nine-person practice prepares for renewal

Illustration only. A nine-person architecture practice working on small residential extensions and a few light-commercial fit-outs has been using AI for about six months. Its fee income is about $1.4 million a year and renewal is in nine weeks. The partner builds the list:

• Tier one: two partners and the office manager use a chatbot for emails and meeting summaries on a business plan with training off. Four other staff use free consumer accounts, which nobody has recorded.
• Tier two: three designers generate concept images for client presentations; one uses AI to summarise planning documents and pre-application notes.
• Tier three: one technician tried an AI tool for door and window schedules and area calculations on a single job; another used a chatbot to draft parts of a specification.

What the partner does in the nine weeks:

1. Week 1. Surveys the staff, finds the free accounts, and moves everyone to the business plan or to no AI. Writes a one-page policy using the AI policy checklist for architects.
2. Weeks 2 to 3. Re-checks the one live job where the schedules were AI-assisted, line by line against the drawings, and records the result. Stops using AI for calculations until a review step is written.
3. Week 4. Reads the draft proposal form from the broker, highlights every AI-related question, and drafts honest answers.
4. Week 5. Emails the broker with the three-tier list, the policy and the questions below, and asks for a written reply.
5. Weeks 6 to 9. Keeps the log running so the renewal answers are backed by records, not recollection.

The effort is perhaps 10 to 12 hours across the nine weeks, mostly the survey and the re-check. The benefit is that the partner can answer any underwriter question with a specific, dated document.

Questions to put to your broker in writing

Email beats phone because you want the answer on record. Adapt this.

Subject: AI use in the practice and our PI policy

Hi [broker first name],

Ahead of renewal on [date], I want to check how our policy
treats the use of AI tools. In summary:
- Administrative use (email, notes): [yes/no, plan type]
- Design support (concept images, summaries): [yes/no]
- Design or calculation output: [none / limited, with review]

Could you confirm in writing:
1. Does our current policy wording exclude or restrict claims
   involving AI-generated or AI-assisted work?
2. Is our AI use a material fact we should disclose at renewal,
   or a change we should notify now?
3. Does the proposal form ask about AI, and how should
   we answer for the uses above?
4. Would the insurer like to see our AI policy or review log?
5. Does the policy's treatment differ for work on live projects
   versus concept stage?

I have attached our one-page AI policy.

Thanks,
[your name]

The five questions cover the points that matter: exclusions, materiality, the form, evidence and the concept-versus-live distinction. If the reply is vague, ask again with a specific scenario, such as "if an AI-assisted area schedule contained an error that led to a claim, would that be covered?".

Answering proposal-form questions honestly and briefly

If the form asks about AI, the answer should be accurate, proportionate and consistent with your policy and log. Here is an illustrative before and after for a question such as "Does the practice use artificial intelligence in the production of design work?".

WEAK
No.

(Not true if designers generate concept images, and risky
if discovered at claim time.)

WEAK
Yes, we use AI extensively across the business.

(Vague, alarming and impossible to evidence.)

BETTER
Yes, in limited ways. AI tools on business plans (training off)
are used for administrative drafting and for concept imagery.
No AI output is issued as design, calculation or specification
without review by a named senior member of staff, recorded in
our project log. A written AI policy is attached.

The better answer says what is used, where, what the controls are and offers evidence. If your real position is less tidy, say so and fix it before renewal, rather than writing something aspirational. An answer that overstates your controls is worse than one that admits a gap and describes your plan.

What to log so you have something to show

An insurer or broker, and later a loss adjuster, will want facts. A lightweight log gives you them without burdening the office.

• Tool register. Each tool, its plan, who may use it, the data-training setting, and the date of your last check of its terms.
• Use register. Which project stages and tasks use AI. One row per use, for example "concept images, stage 2, all projects".
• Review record. For tier two and three uses, who reviewed the output, when, and against what (drawings, code, specification, client brief).
• Incident record. Errors found in AI output, how they were caught, and what changed afterwards. A near-miss log shows insurers you notice problems.
• Client communication. Where clients were told about AI use, and any contracts that restrict it.

A first-quarter pilot structured around these records doubles as evidence; the first-quarter AI pilot plan for an architecture firm shows how to set one up.

A filled-in tool register

This is an invented register for the nine-person practice, after its clean-up. It is the sort of document that turns a broker conversation from vague to specific.

Tool           | Plan              | Users            | Training | Last terms check
Chatbot A      | Business (3 seats)| Partners, office | Off      | [date]
Image tool B   | Paid individual   | 3 designers      | n/a      | [date]
Chatbot C      | None approved     | Nobody           | n/a      | Removed from use
Schedules tool | Trial ended       | Nobody           | n/a      | Not renewed

Approved uses: emails, notes, planning-document summaries,
concept images (labelled indicative), specification wording drafts.
Not approved: calculations, areas, quantities, code compliance,
anything issued without a named reviewer.

Two lines matter more than the rest: "Removed from use" shows you acted on what you found, and "Not approved" shows you drew a line around the high-risk tier. A broker will understand both in seconds.

A second practice: a three-person firm with one enthusiast

A different picture. A three-person practice has no policy and no log, and one architect uses a chatbot daily for specification wording and for explaining planning policy to clients. Renewal is in five weeks. The sensible plan is smaller than the nine-person one, and the order is the same.

• One afternoon to write the tool register and the five-line review rule above.
• A check of the last six months of specification drafts the chatbot touched. Every clause reference gets checked against the standard it names. In a sample like this, finding one or two wrong references is realistic, and fixing them on live projects is the moment of real risk reduction.
• An email to the broker using the five questions, with the one-page rule attached.
• A note to clients in the next appointment letter that AI may be used for drafting, with the architect remaining responsible for all issued content.

The point is proportion. A small practice does not need a committee. It needs a short rule, a record that it is followed and a broker who has seen both.

Scoring each AI use to decide how loudly to raise it

If you are unsure whether your use is "material", a rough score helps you decide how formal the conversation should be. This is my own rule of thumb, not an insurer's test. Score each use: 1 point if output reaches a client, 1 if it reaches issued drawings or specifications, 1 if it includes calculations or code references, 1 if confidential client data is entered, 1 if no named person reviews it. Add them up per use.

• 0 to 1: administrative. Mention it in the tool register and answer any form question accurately.
• 2 to 3: design support. Raise it with the broker at renewal and attach the policy.
• 4 to 5: design or calculation output with weak control. Fix the control first, and ask the broker now, not at renewal.

Take a specification clause drafted by a chatbot: it reaches issued documents (1), includes code references (1), and is reviewed by nobody (1), a score of 3 before any confidential data is added. That is a clear prompt to introduce a review step this month.

Subconsultants, contracts and who carries the risk

Your exposure does not stop at your own staff. Structural engineers, services consultants and specialist designers may use AI in work you coordinate or rely on. Three habits help. First, add a line to your appointment and subconsultant terms asking them to tell you if AI is used in the work they supply and how they check it. Second, check how your contracts allocate responsibility for consultants' work, because that decides whether their errors become your claim. Third, when your own client appoints you under their form of contract, read it for clauses on technology, data and confidentiality before you accept. If a clause is unclear, ask your broker whether it affects cover, since onerous contract terms sometimes sit outside what a policy will pay for.

What an underwriter might ask, and a prepared reply

If an underwriter calls, short, factual replies work best. Here are three likely questions with illustrative replies from the nine-person practice.

Q: Do you rely on AI for design decisions?
A: No. We use it for concept imagery and drafting. Design decisions,
   calculations and compliance checks are made and signed off by
   named staff. Our policy and review log are attached.

Q: How do you stop confidential material leaking into public tools?
A: Approved tools on business plans only, training switched off,
   and a rule that client documents are not pasted into others.
   Staff acknowledged the policy on [date].

Q: Has AI caused an error in issued work?
A: One schedule discrepancy was found in review before issue on
   [project stage]. It was corrected and the use of that tool for
   schedules was paused pending a review procedure.

Notice the third answer. Admitting a caught error, and showing what changed, usually builds more confidence than claiming nothing has ever gone wrong. Only say it if it is true, and keep the specifics in your incident log.

Four claim scenarios an insurer would ask about

A concept image misleads a client

A generated concept shows a window arrangement that cannot be built within the planning constraints, and the client proceeds on the picture. The claim risk is misrepresentation of what is deliverable. Controls: label concept images as indicative, check them against the brief and constraints before they are shown, and keep the label in your standard presentation template. See AI design concept options for presentation habits.

A tool produces a wrong area or quantity

An AI-assisted area schedule is off by 6 per cent, which feeds a fee calculation or a cost plan. The error is an arithmetic one, but it came through an automated route. Control: re-derive key numbers independently and record the check. AI floor plan generation for architects explains why generated layouts fail real briefs.

Confidential client material is uploaded to a free tool

A junior pastes a client's confidential brief into a consumer chatbot. That may breach contract terms with the client and raise questions under your policy's cover for confidentiality or data breaches, which can sit in a separate cyber policy. Does cyber insurance cover AI incidents? sets out what cyber insurers tend to ask.

A client contract bans or restricts AI

A client in a regulated sector requires no AI on their project. Using it anyway is a breach regardless of the insurer. Add a contract check to your project start routine and record the answer.

Telling clients, not only insurers

Disclosure to the insurer sits next to disclosure to clients, and the two should match. If your client letter says AI is never used and your proposal form says it is used for concept imagery, one of them is wrong, and the mismatch is awkward in a dispute. Put one sentence in your standard appointment terms, for example: "We may use AI tools for drafting and early concept imagery. All issued drawings, calculations and specifications are reviewed and approved by a named member of staff, who remains responsible for them." Then keep the wording consistent across the proposal form, your policy and your website. Consistency is cheap to maintain and it is exactly what a claims handler reads for.

It also helps to decide who answers questions. Name one partner as the person who owns the AI register and speaks to the broker, so that staff do not each give their own account of what the practice does. A single owner also makes the quarterly review happen, which is the habit that keeps the log true.

Where the disclosure conversation goes wrong

• Disclosing only to the underwriter and not to the broker, or the reverse. Put your questions through the broker and keep the paper trail.
• Treating "we do not use AI" as true. Staff on free accounts are using it. Survey first.
• Waiting for the insurer to ask. A form that is silent on AI does not mean the matter is immaterial. Ask.
• Overselling controls. Claiming a review process that exists only on paper creates the exposure you were trying to avoid.
• Forgetting subconsultants. If your structural or services consultants use AI in work you rely on, ask how they control it, and check how your contracts allocate responsibility.

A short review rule you can adopt this week

AI review rule (draft)
1. No AI output is issued to a client, a consultant or a planning
   authority without review by a named senior member of staff.
2. Calculations, areas, quantities and code references produced
   or assisted by AI are re-derived independently and the check
   is initialled in the project log.
3. Concept images carry the label "indicative, not for construction".
4. Only approved tools on business plans are used. No confidential
   client material goes into any other tool.
5. Errors and near-misses are logged and reviewed each quarter.

Five lines are enough to show an insurer that the practice has thought about controls. The detail can grow with experience.

What to do at renewal, in order

1. Build the tier list and the registers.
2. Adopt a written policy and review rule.
3. Read the proposal form for AI-related questions, and draft honest answers.
4. Email your broker the five questions above and keep the reply.
5. If your use changes materially during the year, tell the broker then, not at the next renewal.
6. Re-read your client contracts for AI restrictions at each new appointment.

The professional indemnity renewal guide goes through the form itself, and the AI guide links related tutorials for practices. If you would like help mapping your practice's AI use and building the log, my AI implementation consultation covers that; start at the contact page. Take policy wording questions to your broker or a solicitor, because they can read your actual contract.

Questions architects ask about insurers and AI

Will using AI make my PI premium go up?

Nobody can say in general. Insurers price on your fee income, project types, claims history and controls. Some may ask more questions about AI. A written policy, a review step and a clean log are the things most likely to help the conversation.

Is a free chatbot for emails something to disclose?

Possibly not as a design risk, but read the proposal form wording. If the form asks about any use of AI in the business, answer it accurately and briefly. If it only asks about design or calculation, tell your broker how you read it and keep their answer.

What if my insurer has not asked about AI?

In many markets, disclosure duties turn on facts that would matter to the insurer's decision, not only on questions asked. Because that depends on your policy wording and the rules where you practise, ask your broker in writing whether AI use is a material fact for your policy.

Should I stop using AI until I have asked?

Not necessarily. Pause use on live design and calculation work if you have doubts, and keep to low-risk tasks such as drafting emails. Then ask your broker before the next renewal or sooner if your use changes materially.

Further reads

• Can AI Help a Small Architecture Practice? Where It Saves Time — Where AI saves a small practice time, and where it does not.
• AI Tools for Architects 2026: A Toolkit From Brief to Handover — Where AI fits at each project stage, for your risk map.
• How to Use Claude AI in Architecture: Specs and Planning Reports — Using Claude on specifications and planning reports with checks.
• AI for Architecture: Keep a Drawing Issue Register and Transmittal Log — Issue registers that also serve as review evidence.
• How Much Does AI Cost a Small Architecture Practice per Month? — What AI costs a small practice, for the business case.
• AI Consultant for Architects: What a First Engagement Costs — Seven consultant days priced deliverable by deliverable for a ten-person practice: audit, policy, one workflow, team session and a 30-day check.
• AI Training for Architects: What a Practice Workshop Covers and Costs — A half-day AI workshop agenda for an architecture practice, modules by role, and the cost per person once staff time is counted, set against self-taught routes.
• AI Automation for Architects: Enquiry to Fee Invoice, Step by Step — Six draft-only automation steps from enquiry form to overdue-invoice reminder, with costs, task counts and the architect approving every document.
• AI Software for Architectural Drawings: Notes, Schedules and Sheet Checks — How a small practice can check a drawing set for index, note and schedule errors with ChatGPT or Claude, and what the architect still signs.
• AI for Architects: Draft Design Schematics From a Client Brief — A brief-to-schematic workflow for small practices: stress-test the brief, draft an area schedule and options, then run the checks that stay with the architect.
• Architect Meeting Minutes: Record, Summarise and Issue Same Day With AI — A same-day workflow for architects: consent to record, a minutes template, a decisions log and an issue sheet, using Otter, Teams Copilot or Meet notes.
• Architect Site Visit Report: Notes and Photos to a Clear Report With AI — A worked notes-to-report example for architects: observation, instruction and action columns, photo captions, and a rule that keeps unseen items out.
• Architect Website: Project Descriptions and Service Pages Written With AI — A copy workflow for architects: project descriptions from verified facts, service pages by project type, honest area pages and an FAQ on fees and process.
• Architecture RFP Response: Reuse Past Proposals With AI, Without Copying — An answer library, win themes and a scoring checklist for architecture RFPs, with a confidentiality rule for reused client text.
• Can AI Write an Architect's Snagging List From Site Photos? — A snagging-list workflow for a small practice: photo and note capture, an AI draft by room, severity and responsibility, and a line-by-line check before issue.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: a design-sector insurance broker's November 2025 commentary on AI and professional indemnity insurance, an architects' regulator's PI renewal advice (both checked November 2026), and general principles on disclosure at proposal and renewal. This is not legal or insurance advice: check your own policy wording with your broker.