AI Tools for UK Architects: A Practical Evaluation Guide (2026)

For a site-research tool, select a location with a known constraint and one with incomplete records. Check whether the output identifies its sources, dates and geographic coverage and makes missing information visible.

Atlasly supports early site analysis and access through compatible assistants. Read the methodology and limits, inspect material findings in their primary sources, and check the current account allowances. Availability depends on the place, data layer and workflow; this is not a guarantee of complete local records or compliance.

For another vendor, ask for the same evidence. Do not assume that a planning-focused assistant has or lacks maps, exports or current local data without checking its present product documentation.

Autodesk describes Forma Site Design as supporting early design with contextual data and environmental analysis. Test the particular analysis and exchange workflow you need rather than treating a product-family name as a feature guarantee.

When evaluating concept imagery, keep the design inputs and any altered geometry visible. A generated image does not establish dimensional accuracy, buildability or environmental performance. For spatial handovers, use the 3D context export checklist to check scale, location and missing elements before continuing design work.

For a drafting assistant, provide a bounded set of project notes and approved sources. Ask for a draft that distinguishes project facts from assumptions and flags missing information. A reviewer should be able to trace each material planning statement to its actual source, document version and relevant passage.

Do not treat a plausible reference as a verified citation. Check it in the original source. Review the service's current data-handling terms and your practice's confidentiality requirements before supplying project material. Keep professional judgement and sign-off with the appointed team.

Use the same brief and record these results for each candidate:

1. Task completion: did the output address the specified question?
2. Evidence: which material statements could be verified, and which could not?
3. Coverage: what was missing, stale or outside the supported scope?
4. Correction effort: what had to be redrawn, rewritten or researched separately?
5. Handover: could another person open and understand the output and its assumptions?
6. Commercial fit: what are the actual plan, seat, data and export limits?

Record setup and review time separately from generation time. Repeat on more than one representative task before making a practice-wide purchasing decision. This is a proposed evaluation method, not a claim that we have performed the tests.