RIBA 2026
74% vs 19%
Respondents reporting practice use on some projects, compared with reported documented AI policy.
Survey analysis
What 53 surveys say about AI adoption.
Five surveys that repeat each year show AI use rising, with the limits explained below. Evidence of routine, governed use and measured impact is less consistent.
More use, year on year
Five surveys, each compared with its own earlier editions.
| Survey | What it asks | Earliest | Latest |
|---|---|---|---|
| RIBA (mainly UK) | Respondents reporting that their practice uses AI on at least some projects | 41% (2024) | 74% (2026) |
| Deltek (US and Canada) | Architecture and engineering firms using or adopting AI | 38% (2024) | 70% (2026) |
| Deloitte / Autodesk (Asia Pacific) | Construction and engineering businesses using AI or machine learning | 30% (2024) | 46% (2026) |
| Statistics Canada | Construction establishments using AI to produce goods or services in the last 12 months | 1.9% (2024) | 9.2% (2026) |
| RICS (global) | Construction respondents using AI regularly in specific processes | 12% (2025) | 19% (2026) |
The figures differ because the questions differ. RIBA's broad test and Statistics Canada's narrow one can both be right, so the surveys are never averaged into one rate. Deltek changed its wording and who it asked between years. Deloitte/Autodesk’s 2026 sample replaces Malaysia with Vietnam; it is not clear how the regions are weighted. So the figures show a reported change, not a like-for-like comparison of the same people.
Trial is not routine
Reported use is higher than reported governance and impact.
Deltek 2026
70% vs 38%
Using or adopting AI, against those reporting a measurable positive impact.
RICS construction 2026
39% vs 19%
In a pilot phase, against regular use in specific processes.
These are different questions and may have different response bases. The policy and impact figures are shares of all respondents, not just of those using AI. Respondents also report benefits, such as productivity gains, but these are self-reports with no common baseline, and not independently verified.
What this means
Use the surveys to choose where to look.
Named cases describe repeated use in asset operations, inspection and site progress, as well as emerging design and advisory applications. Surveys alone cannot rank these workflows by maturity or value. A reported time saving, lower energy use and fewer incidents answer different questions.
Surveys show where AI may be worth testing. They cannot show what it will be worth to your organisation, which depends on your work, your controls and the effort of checking the output.
Method and limits
How the review was done.
38thMove catalogued 53 AI-related survey programmes and survey-based publications for architecture, engineering, construction and property, published or current between 24 September 2025 and 24 September 2026. Five recurring series supported comparison over time, subject to changes in wording, response bases and sample composition. Results were kept separate where the population, definition of use or denominator differed.
This is not a representative global survey. Evidence is stronger for Europe, North America and parts of Asia Pacific than elsewhere, and self-selection and sponsor interests remain limitations. Responses and reported benefits have not been independently verified by 38thMove. Reviewed 24 September 2026; updated 30 September 2026.