Architecture Firm AI Policy: A One-Page Template for Small Studios

Three in four architecture practices now use AI on at least some projects, but only 19% of them have written that use down anywhere, according to RIBA's AI Report 2026, published July 22, 2026 (riba.org). That gap is one reason more clients now ask a studio, in plain terms, what its architecture firm AI policy actually says. You don't need a governance manual to answer well: a single page naming what AI is for, what data stays out of public tools, who checks the output and how it's labeled covers most of what a client, a new hire or a proposal reviewer wants to know. Below is that page, built from RIBA's 2026 numbers, the AIA's January 2026 guidance and a construction-law firm's contract advice, ready to adapt with your studio's own details.

Key takeaways

• RIBA's AI Report 2026 (July 22, 2026) found 74% of practices use AI on at least some projects, up from 59% a year earlier, yet only 19% have a documented AI policy.
• Beale & Co's August 2026 analysis of the same survey found early-stage visualization is the single most common use (56%), and 76% of respondents think AI raises imitation risk.
• The AIA published its Position Statement on AI and its Guidance for the Responsible Use of AI by Architecture and Design Firms in January 2026; both are free downloads.
• A usable architecture firm AI policy fits on one page and covers five things: approved uses, a client-data rule, a named reviewer, a disclosure line, and a way to protect junior staff training.
• Treat the page as something you revise as tools change, not a form you fill in once and file away.

The gap between using AI and writing down how

RIBA's 2026 survey, covered in the same July 22 release, shows adoption climbing fast: 74% of practices used AI on at least some projects this year, up from 59% in 2025. The productivity case looks real too — 75% of respondents reported improved productivity and 57% reported a positive return on investment. But only 17% agreed their designs are actually better because of AI, and 77% agreed AI can never replace human creativity. RIBA President Chris Williamson put the caution plainly: "Professional oversight of design remains essential if tomorrow's buildings are going to be safe and sustainable."

That oversight is uneven by firm size. Beale & Co's August 2026 summary of the same data found adoption reaches 90% at practices with 50 or more staff but only 63% at smaller ones (beale-law.com). Documentation follows the same pattern: 56% of large practices already have a written AI policy, against 19% of practices overall, with another 17% developing one and 15% planning to start. A small US studio sits on the wrong side of both numbers — using AI daily, with no page to show for it if a client, an insurer or a new hire asks.

What's pushing this from "someday" to "this quarter"

The same report gives two reasons a policy is no longer optional-feeling. Beale & Co found 56% of AI users rely on it most for early-stage visualization — renderings, massing studies, concept video — which is exactly the category most small studios already touch. And 76% of respondents agreed AI increases the risk of imitation, "raising familiar issues around intellectual property, authorship and originality," while 55% believe AI will let people without sufficient professional qualifications produce designs, a competence and accountability concern the report ties directly to building safety.

Beale & Co's practical advice, aimed at firm principals rather than policy writers, is to put AI terms directly into appointment and sub-consultancy agreements: "clear controls around approved tools, confidentiality and data handling, checking and verification of outputs, intellectual property, record keeping and responsibility for final decisions." A one-page internal policy is the source document for that contract language — write it once and you can pull from it every time a client agreement needs an AI clause. None of this is legal advice; a firm with real contract exposure should still have counsel review the specific wording.

There's a second, less contractual pressure: 61% of RIBA respondents agreed AI makes it harder for early-career professionals to acquire essential skills. That's a staffing risk as much as a legal one, and it belongs in the policy, not just in a hallway conversation about who gets to use which tool.

The five things a one-page policy needs to cover

A policy this short only works if it answers five specific questions rather than gesturing at "responsible use." Each maps to a concern in the data above:

• What AI is approved for. Name the categories — visualization, marketing video, meeting notes, drafting non-technical text — and name what requires a licensed professional's independent sign-off regardless of what a tool produced, such as code compliance or structural calculations.
• What never goes into a public AI tool. This is the line most firms haven't drawn yet. Unreleased client renderings, site security details, personnel information and unexecuted contract terms need an explicit "no" for any tool without an enterprise data agreement.
• Who reviews outputs before they leave the building. Name a person or role, not a committee, so there's no ambiguity about who catches an invented detail, a wrong dimension or a passage that resembles another architect's published work.
• How AI-assisted work is labeled. Client deliverables and public posts should say when AI played a substantive role, and unbuilt work shown as a concept study should say so plainly. See our guide to AI rendering disclosure for wording you can copy for renderings and video specifically.
• How training is protected. Given RIBA's 61% figure on early-career skill-building, name a floor: a task type, an experience threshold, or a review step that keeps AI from quietly replacing the reps a junior staff member needs.

Copy this one-page template

Adapt the five lines below on your own letterhead. They're deliberately short enough to fit on one page with room for your logo and a signature line.

Approved uses. AI tools may be used at [Studio Name] for early-stage massing studies, marketing renderings and video, meeting notes, and drafting non-technical text such as proposals and social captions. AI tools may not be used for code compliance checks, structural calculations or life-safety analysis without a licensed professional's independent verification.

Data rule. No unreleased client renderings, drawings, site security details, personnel records or unexecuted contract terms go into a public AI tool. [Name of approved enterprise tool, if any] is the only tool cleared for that category of data.

Review. [Name or role] reviews every AI-assisted deliverable before it reaches a client or goes public, checking for invented details, incorrect dimensions, and resemblance to another firm's built or published work.

Disclosure. AI-generated or AI-assisted renderings and video are labeled as such in client deliverables and public posts. Unbuilt work shown after AI generation or alteration is identified as a concept study, not a finished design.

Training. Staff with fewer than [X] years of experience complete at least one manual pass on each project task type before using AI on that task, so AI use doesn't substitute for building core skills.

Where the AIA's guidance fits

The AIA published two documents in January 2026 that back up this kind of page with organization-level authority: the Position Statement: Artificial Intelligence (AI), and the longer Guidance for the Responsible Use of AI by Architecture and Design Firms, both linked from the AIA's AI in architecture page. The AIA also runs a separate AI Firm Toolkit for firms wanting a slower, guided drafting process. Point to both documents if a client or an AIA-member architect on staff wants to see the page against a recognized professional source.

Applying the review and disclosure lines to AI video specifically

If your studio uses Arch2Video to turn renderings or finished-project photos into a short film, the review and disclosure lines above apply directly, not as an afterthought. Arch2Video's AI director plans the story across six connected scenes and generates the motion; that generated motion interprets your source images and can change details, so a person should watch every film before it goes out, exactly what the "review" line requires. Every film also includes a small original fictional cast — people in your source photos aren't animated, they're replaced — worth a line in your own disclosure wording. Say the source images show a hillside house that hasn't broken ground: the disclosure line above should say so plainly, not let the finished look of the film imply the project is built. Our own example films are labeled as concept studies made from published project photography for the same reason.

Two related checks belong in the same review pass. First, confirm you own or hold a license covering AI-assisted use for any image you feed into a public tool — see our photo licensing guide if a photographer shot the originals. Second, know that some delivered AI video files now carry embedded content credentials that platforms read automatically, which is a separate layer from the caption wording your policy asks staff to write.

Keeping the page current

RIBA's 2026 report already compares this year's adoption figure against 2025's, so a future edition should show whether the policy gap closed or widened. Treat your one-page document the same way: put a review date on it, six to twelve months out, and revisit it whenever your studio adopts a new AI tool. A policy that's a year out of date on approved tools is worse than no policy, because it gives staff a false sense that someone already thought through the current one.

The page doesn't need a committee to approve changes. Whoever owns the document — a principal, an office manager — should be able to add a line the day a new tool shows up, then tell the team.

Frequently asked questions

Do small studios really need a written AI policy, not just a verbal understanding?

Yes, for two practical reasons beyond RIBA's numbers: a written page is what you hand a client or insurer who asks, and it's the only version everyone on staff is actually working from. A verbal understanding drifts as tools change and staff turn over; a one-page document doesn't, and it takes an afternoon to write using the template above.

What should never go into a public AI tool?

Unreleased client renderings, drawings, site security or access details, personnel records, and terms from a contract that hasn't been executed. Treat any tool without a signed enterprise data agreement as public by default. If a tool's terms are unclear, don't upload the material until someone confirms them in writing.

Does labeling AI-generated video satisfy the disclosure line in this policy?

It covers the public-facing half. The disclosure line should also apply inside client deliverables, not just social posts, since the AIA's guidance ties disclosure to any deliverable a client will rely on. See our AI rendering disclosure guide for wording that covers both built and unbuilt work.

How often should we update the AI policy once it's written?

Put a review date six to twelve months out, and revisit sooner whenever the studio adopts a new AI tool or a new law or platform rule changes what disclosure requires. RIBA has now published back-to-back yearly figures, so treat each new edition as a prompt to check your page against current adoption and risk data, not just your own tool list.

Is this template legal advice for our specific contracts?

No. It's a starting point for an internal policy and for the kind of AI language Beale & Co recommends adding to appointment and sub-consultancy agreements — approved tools, data handling, output verification, IP and responsibility for final decisions. A firm with real contract or liability exposure should still have counsel review the exact wording before it goes into a signed agreement.