AI in architecture: what 15,000 German architects reveal about AI in practice

Why a German survey matters to architects elsewhere

In Germany, anyone who uses the professional title “Architekt” must be registered with a regional chamber of architects. That makes the chambers’ survey unusually broad: it reaches sole practitioners, employees and owners of large firms alike, not only the firms that are already interested in technology.

German project work is organised along the nine work stages of the HOAI, the statutory fee scale – from basic evaluation and preliminary design through permit planning, detailed design and tendering to site supervision. They play a similar role to the stages of the RIBA Plan of Work in the UK. When this article speaks of “project stages”, it means these phases of a real building project.

How many architects use AI

Across all respondents, 29 % use AI-based tools in their daily work and 71 % do not. Use rises with firm size – in firms with 25 or more employees, more than half of the owners use AI – and falls with age: 44 % of architects under 40, but 17 % of those aged 60 and over.

The UK numbers are higher. In the RIBA AI Report 2025, 59 % of practices said they use AI, up from 41 % a year earlier. The two surveys are not directly comparable – the RIBA asked practices, the German chambers asked individual members – but the gap points in a clear direction: German practice is earlier in the curve.

What architects actually use AI for

Among German architects who use AI, text dominates: 58 % use it very often or often for writing, such as minutes and public relations. About a third use it for legal questions on contracts and fees, about a quarter for documentation, and about a fifth each for checking building-regulation requirements, visualisation and practice information material. Scheduling and cost planning are named by 13 %.

In other words, AI has arrived where it works without preparation: in a general chat window, for a single text. It has barely reached the stage-by-stage project work where most of an architect’s hours go.

Why AI stalls in architecture firms

Nine in ten German AI users report difficulties. The three most common are practical, not technical:

• no reliable, complete data to work with (54 %)
• no experience with AI tools (52 %)
• tools that do not work with the software already in use (41 %)

The risks architects see

95 % of all respondents see risks. The biggest is the uncritical adoption of AI output (71 %). About half also name dependence on software companies, unresolved legal questions including data protection, and data misuse and security.

At the same time, 77 % want training on AI – and 91 % of those who already use it. The profession is cautious, aware of the limits, and eager to learn: a good starting point, as long as firms do not confuse trying a chatbot with using AI in their projects.

How architecture firms can start with AI

The survey does not describe a technology problem but an adoption problem. Five principles follow from it, and they apply to firms in any country.

• Start with a task, not a tool: pick one routine-heavy task in one project stage – checking areas against the brief, the completeness of a planning application, a bid comparison, the site diary – and test it on a live project.
• Put your own data in order first: drawing lists, room data sheets, minutes, templates and office standards. AI can only use what it can find – and tidy data pays off even without AI.
• Replace open questions with checking steps: break the task into checks against the fee scale, standards and building regulations, make every result cite its source, and compare several models on critical questions. Design, decisions and liability stay with the architect.
• Use AI where the firm already works: connected to CAD exports, tendering software, project files and templates, set up with the access rights your IT defines – not in a separate window.
• Train the team and set clear rules: which data may go into which tool, how results are checked and who signs off.

The EU AI Act and data protection

For firms working in the EU, training is not optional. Since 2 February 2025, Article 4 of the EU AI Act requires organisations that use AI systems professionally to take measures, to their best extent, so that their staff have a sufficient level of AI literacy. It does not prescribe a format.

Client drawings, contracts and project files are confidential and often contain personal data. Before a tool is used, a firm should know where data is processed, whether the provider trains its models on it, and whether a data processing agreement under Article 28 GDPR is in place. As of October 2026; not legal advice.

What this means for your practice

The German numbers describe a profession that is curious but has not yet moved AI from the chat window into its projects. The next step is not another tool, but one well-chosen task on a live project – with data in order, checking steps and a connection to the software the team already uses.

Kernmoment brings AI into architecture practices along the project stages, alongside the practice’s own team and IT.