"AI" in construction covers several quite different technologies: language models that read and write text, computer vision that recognises things in images and scans, machine learning that predicts from historical data, and optimisation that searches design options. Each has real uses today, and each is surrounded by marketing. The useful question is never "should we use AI?" but "which task, with which data, checked by whom?"
At a glance
| Technology | What it does | Mature uses today |
|---|---|---|
| Language models (LLMs) | Read, summarise and write text; answer questions; call tools | Summarising specifications and contracts, drafting RFIs and reports, searching project documents, assistants that query models through APIs or MCP |
| Computer vision | Recognises objects, defects and progress in photos, video and scans | Progress tracking from 360° photos and scans, safety monitoring (PPE, exclusion zones), defect detection in inspections |
| Machine learning on project data | Predicts from patterns in past data | Risk of delay or cost overrun, equipment failure prediction from sensor data, estimating from historical costs |
| Optimisation and generative design | Searches many options against goals and rules | Site layouts, structural grids, room arrangements, MEP routing proposals |
Where it helps the BIM person
| Task | How AI helps | Check needed |
|---|---|---|
| Reading a 300-page EIR or specification | Summary, list of deliverables and deadlines | Compare with the source on anything contractual |
| Writing requirements and plans | First drafts of BEP sections, IDS rules from plain language | Expert review; requirements must be precise |
| Model queries | "Which rooms have no finish?" through an assistant connected to the model | Spot-check counts and totals |
| Classification | Suggesting classification codes for thousands of elements | Review low-confidence suggestions |
| Issue management | Grouping similar clashes or issues, drafting descriptions | Coordinator decides what is one issue |
| Handover data | Extracting product data from O&M manuals and datasheets | Check against the asset requirements |
Questions to ask before using an AI tool
- What data does it send, and where? Client contracts and security requirements may forbid sending models or documents to external services.
- Is it available today, or a demo or preview?
- How will results be checked, and by whom?
- What happens when it's wrong? Draft text is low risk; a changed model or a wrong quantity in a bid is not.
- Who owns the output, and can you explain how it was produced?
Worked example
A contractor receives tender documents with 2,400 pages. An assistant connected to the document set lists every information deliverable, its format and its milestone, with a reference to the page it came from. The information manager checks each reference (finding three it misread) and has the information delivery requirements drafted in a day instead of a week. The checking took half that day; it was the part that couldn't be skipped.
Common mistakes
- Using public AI tools with confidential project documents.
- Treating confident answers as correct answers.
- Buying a tool for "AI" before identifying the task and the data.
- Automating a bad process instead of fixing it.
Where to read more
National construction and digital bodies publish guidance on AI use and data protection; check your client's and company's AI policies. Vendor claims are best checked through pilots on your own data. The weekly "This week in BIM" roundup on the BIMpedia forum follows AI and automation news.