How Accurate is AI for UK Construction?
1,001 questions. 12 AI models. 20 categories. One definitive answer.
1,001
Questions
12
AI Models
20
Categories
78%
Best Score
Methodology
How we tested AI's knowledge of UK construction standards and regulations.
1,001 Technical Questions
Questions spanning building regulations, British Standards, health and safety, fire safety, structural design and 15 other categories relevant to UK construction professionals.
12 AI Models Tested
6 paid and 6 free models from OpenAI, Anthropic, Google, Mistral and Perplexity. Updated 5 May 2026 with Claude Opus 4.7 and Claude Haiku 4.5.
3-Point Scoring
Each answer scored as Correct (full marks), Partial (half marks) or Wrong (zero). Scores verified against published standards and regulations.
20 Specialist Categories
From Accessibility to Waterproofing, covering the full breadth of knowledge a UK construction professional might need from an AI assistant.
All 10 models ranked by accuracy
Across 1,001 questions covering UK construction standards and regulations.
What the data reveals about AI in UK construction
78%
Highest overall score
Claude Opus 4.7 led the pack at 78%, edging out Opus 4.6 (77%). No model broke the 80% barrier, highlighting clear limits in AI's construction knowledge.
12.8pp
Paid vs free gap
Paid models averaged 68.2% versus 55.4% for free models. A 12.8 percentage point gap that makes the business case for paid subscriptions clear.
4.8%
Lowest error rate
Claude Opus 4.7 had only 48 outright wrong answers from 1,001 questions. The worst performer got 400 wrong (40.0%).
91%
Best category score
Claude Opus 4.6 scored 91% on Sustainability and Carbon, with Opus 4.7 close behind at 89%. Well-documented, publicly available standards consistently produced higher AI accuracy.
Web
Perplexity advantage
Perplexity models consistently outperformed ChatGPT models, likely due to real-time web search giving access to current standards and guidance.
48%
Worst category average
Construction Technology, Contracts, Waterproofing and Demolition all averaged under 50%. Paywalled and niche specialist standards are poorly represented in AI training data.
Is paying for AI worth it in construction?
The data is decisive.
Paid models average
68.2%
6 models tested
Free models average
55.4%
6 models tested
All 20 categories across all 10 models
Green is good. Red is risky.
| Category | Claude Opus 4.7 | Claude Opus 4.6 | Perplexity Pro | Perplexity | Mistral Large | Claude Sonnet 4.6 | Gemini 2.5 Pro | ChatGPT 5.4 | Mistral Small | Claude Haiku 4.5 | ChatGPT Free 5.3 | Gemini 2.5 Flash |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Accessibility & Inclusive Design(50 Qs) | 74% | 67% | 62% | 52% | 61% | 45% | 58% | 40% | 38% | 37% | 35% | 34% |
| British Standards - Concrete & Steel(50 Qs) | 83% | 80% | 80% | 78% | 79% | 72% | 76% | 65% | 60% | 71% | 51% | 55% |
| British Standards - Other Materials(50 Qs) | 85% | 87% | 79% | 82% | 74% | 77% | 80% | 67% | 65% | 64% | 58% | 56% |
| Building Regulations - Fire Safety(55 Qs) | 80% | 87% | 73% | 69% | 77% | 77% | 74% | 65% | 58% | 58% | 52% | 55% |
| Building Regulations - Other Parts(50 Qs) | 77% | 78% | 74% | 74% | 65% | 67% | 70% | 55% | 52% | 54% | 48% | 52% |
| Building Regulations - Thermal(50 Qs) | 78% | 87% | 75% | 71% | 62% | 65% | 72% | 55% | 50% | 51% | 42% | 48% |
| Construction Technology(50 Qs) | 71% | 73% | 66% | 59% | 49% | 52% | 55% | 42% | 38% | 51% | 33% | 28% |
| Contracts & Procurement(55 Qs) | 70% | 64% | 55% | 50% | 53% | 56% | 50% | 42% | 38% | 49% | 30% | 32% |
| Demolition & Refurbishment(41 Qs) | 73% | 71% | 61% | 61% | 55% | 58% | 56% | 48% | 44% | 50% | 40% | 36% |
| Environmental & Contamination(45 Qs) | 81% | 66% | 70% | 68% | 55% | 58% | 60% | 50% | 46% | 48% | 42% | 42% |
| Fire Safety - Post-Grenfell(50 Qs) | 84% | 79% | 74% | 76% | 68% | 66% | 67% | 57% | 52% | 45% | 45% | 48% |
| Health & Safety / CDM(50 Qs) | 84% | 78% | 72% | 76% | 69% | 68% | 72% | 61% | 58% | 49% | 54% | 56% |
| MEP & Building Services(50 Qs) | 73% | 73% | 71% | 70% | 58% | 62% | 63% | 50% | 45% | 49% | 38% | 46% |
| Materials & Products(50 Qs) | 77% | 77% | 73% | 70% | 64% | 64% | 68% | 56% | 50% | 61% | 44% | 58% |
| NHBC Standards(50 Qs) | 73% | 75% | 64% | 62% | 64% | 60% | 60% | 52% | 48% | 54% | 42% | 42% |
| Planning & Permitted Development(55 Qs) | 86% | 87% | 76% | 76% | 73% | 73% | 72% | 67% | 62% | 57% | 55% | 44% |
| Roofing & Cladding(50 Qs) | 74% | 76% | 66% | 62% | 57% | 65% | 58% | 48% | 42% | 54% | 38% | 46% |
| Structural Design & Loading(50 Qs) | 84% | 82% | 75% | 77% | 77% | 68% | 70% | 60% | 55% | 65% | 48% | 56% |
| Sustainability & Carbon(50 Qs) | 89% | 91% | 78% | 72% | 72% | 78% | 75% | 65% | 60% | 57% | 55% | 60% |
| Waterproofing & Below-Ground(50 Qs) | 63% | 59% | 52% | 50% | 47% | 57% | 53% | 42% | 38% | 33% | 33% | 34% |
Compare model performance by category
Select any of the 20 categories to see how each model performed.
What this means for UK construction professionals
- 1
AI is useful but not reliable enough to replace professional judgement.
Even the best model got 22% of answers wrong or only partially right. For safety-critical decisions, always verify AI output against published standards.
- 2
Pay for your AI tools.
The 12.8 percentage point gap between paid and free models is significant. If you're using AI for construction work, a paid subscription is recommended.
- 3
Web-connected AI performs better.
Perplexity's real-time web access gave it a measurable advantage over models relying purely on training data. Look for AI tools that can reference live sources.
- 4
Specialist and paywalled standards remain a blind spot.
AI struggles most with niche areas like waterproofing, construction technology and contracts. These are precisely the areas where professionals need the most help.
- 5
AI accuracy tracks public data availability.
Well-documented areas like planning, health and safety, and sustainability score highest. Industry bodies should consider how their standards are made accessible to AI systems.
This study was conducted by Fabrick in 2025. All questions were written by construction industry professionals and verified against published UK standards and regulations. Models tested using identical prompts and 3-point scoring (Correct / Partial / Wrong).
Want a platform like this for your business?
This platform was built by Fabrick. We create bespoke digital tools, data platforms and content hubs for construction and built environment companies -- designed to demonstrate your expertise and generate qualified leads.
Fabrick - Award-winning marketing specialists for the built environment