An image can look resolved before the design behind it has been tested.

AI can help a practice explore visual ideas, organise a brief or prepare a first draft. Each output needs a clear purpose, a suitable review and an explanation of what the recipient can rely on.

The RIBA AI Report 2026 explores adoption across design and practice operations, alongside questions about people, ethics, intellectual property and the environment. It is useful reading alongside the ARB Architects Code, which sets the professional standards expected of registered architects.

The practical opportunity is to improve a defined part of the work while making the basis and status of its outputs clear.

The curved glass facade of One Blackfriars in London photographed from street level
One Blackfriars, London. Photo: Nick Fewings / Unsplash. Editorial illustration.

01 / THE PROFESSIONAL CONTEXT

Understand the two roles

RIBA / RESEARCH AND INSIGHT

Understand the opportunities

The report combines survey findings with expert contributions. External contributors’ views are their own and do not represent RIBA’s position. It is not a mandatory AI practice standard.

ARB / PROFESSIONAL STANDARDS

Apply the standards of practice

The 2025 code’s six standards cover honesty, public interest, competence, professional practice, communication and collaboration, and respect. Registered architects must meet the standards; supporting examples explain ways to do so.

Read ARB’s full code. The practical recommendations below are Go Left’s interpretation for AI-assisted workflows, rather than additional rules issued by either body.

02 / OUTPUT STATUS

Give each output a clear purpose

Before generating anything, decide what the output is for and who will see it. A visual used to discuss a mood or material palette needs a different assessment from information issued to support a project decision.

ILLUSTRATIVE OUTPUT BOUNDARIES / GO LEFTWhat is the recipient being shown?
Design exploration
Ideas for discussion. Identify generated or altered content and what has not yet been checked.
Project development
Work in progress. Check the relationship to the brief, current model and agreed design decisions.
Information for reliance
A defined issued purpose. Complete the relevant checking and authorisation, with its status and limitations clear.

Use wording, document status and communication that suit the recipient. If a concept visual is forwarded without its explanation, consider whether it could be mistaken for a developed proposal.

A label helps communicate the boundary. The way the work is reviewed and issued needs to support it.

03 / CHECKING

Check what the output implies

For a generated image, review more than appearance. Has it changed the relationship between spaces, introduced an opening or suggested a material detail that has not been designed?

A convincing visual cannot establish dimensions, buildability, performance or compliance. Identify the underlying design work needed before those matters can be represented as resolved.

For a brief summary or requirements register, check each important item against the source. Preserve client qualifications, competing requirements and points that remain undecided.

Name the person who will carry out the review, give them the necessary sources and allow time for correction. Keep records proportionate to the purpose and consequences of the output.

04 / PROJECT INFORMATION

Protect briefs and design material

A practice may handle confidential development proposals, client records and design material belonging to other parties. Establish which information can be used in the selected AI service.

Check input rights, licences, client terms and the service’s provisions for access, storage, retention and model training. Consider what happens when images or files are shared through plugins and connected applications.

Set a clear route for requests that fall outside the approved use. A team member should be able to find out whether a file can be uploaded without having to interpret a software contract during a deadline.

Our guide to AI data handling sets out the provider questions. The ICO’s resources support the assessment where personal information is involved.

05 / PEOPLE AND LEARNING

Keep learning in the workflow

Consider how a changed process affects the way less experienced colleagues learn. If AI prepares a first draft, give the team a structured opportunity to compare it with the source and discuss the decisions behind the final output.

For example, a junior colleague could identify where a draft brief summary omitted a qualification, then review the correction with a supervisor. That builds understanding of both the project and the tool’s limitations.

Ask users what improves their work and what creates extra checking. Include the people reviewing outputs, maintaining information and explaining the results to clients.

Choose an owner who can respond when the process, software or project requirements change.

06 / A FOCUSED TRIAL

Test a focused improvement

Choose one workflow with a visible cost or delay. Preparing a draft requirements register from an agreed set of briefing documents could be a suitable task to investigate.

  • Define the documents, users, output and reviewer.
  • Agree how omissions, contradictions and uncertain statements should be flagged.
  • Measure preparation, review and correction effort together.
  • Check whether the final output supports the intended design conversation or decision.

Agree the evidence needed to continue and the conditions that would stop the trial. Use the results to decide whether to refine the workflow, extend it or choose a different approach.

This is the role of BUILD™: connect a business need with a controlled test and develop the solution as the evidence grows.