The Practical AI Series · Evergreen Guide

Artificial IntelligenceUK Construction

AI Prompts for Construction Companies: A Practical UK Guide

The useful prompt is not the clever one. It is the one that gives AI the same project context, boundaries and review standard you would give a competent member of your team.

By Andrew Davey•Evergreen Guide•13 min read

Most lists of AI prompts for construction companies have the same problem: they give you a sentence to copy, but not a system your estimator, contracts manager, site manager or office team can safely reuse on a real project.

A prompt such as “write an RFI” can produce something polished in seconds. It can also quietly invent the drawing reference, assume a specification requirement or turn an unresolved site question into a statement of fact.

For UK contractors, builders and specialist trades, the better approach is to treat prompting like a small work instruction. Give the AI the source material, tell it what job it is doing, define what it must not assume and state how the output will be checked.

The rule that makes this guide different: AI may help organise, compare, draft and question project information. It should not become the source of truth for safety, contractual entitlement, design compliance, programme commitments or final pricing. Those decisions remain with the appropriate competent people and project dutyholders.

Why Generic AI Prompts Fail in Construction

Construction information is unusually context-dependent. A sentence that is harmless in a marketing email can be expensive in a tender clarification, variation notice or subcontract order.

Documents matter
Drawing revisions, specifications, schedules, tender addenda and meeting records can change the answer.

Roles matter
A client update, subcontractor instruction and internal QS note may describe the same issue very differently.

Authority matters
A fluent answer is not evidence that the model had authority to make a technical, contractual or safety decision.

This is why “act as an expert quantity surveyor” is usually a weak prompt. Giving a model a professional title does not give it your contract, scope, rates, drawing register, employer’s requirements or commercial strategy.

General prompting guidance rightly emphasises clear instructions, context and iteration. Construction teams need one more discipline: tell the AI what evidence it may rely on and what it must flag rather than infer.

The SITE Method: A Construction-Specific Prompt Structure

For repeatable work, use the same four-part structure every time. We call it the SITE method.

Source material

Tell the AI exactly what information it may use: a tender pack, drawing register, site notes, quotation, programme, approved company template or the text pasted below.

Intended outcome and audience

State what the document is for and who will read it. An internal commercial review is not written like a client email, and a site briefing is not written like a board report.

Terms, constraints and truth boundaries

Define the format, tone, project terminology and—most importantly—what the model must not invent. Require unknowns, conflicts and missing evidence to be labelled explicitly.

Evaluation and escalation

Tell AI how the output will be judged and which items require review by a QS, project manager, designer, H&S lead, contracts manager or other competent person.

SOURCE Use only the project information I provide below. Do not rely on assumed industry practice where the documents are silent. INTENDED OUTCOME Create a concise internal briefing for our project manager and quantity surveyor. TERMS / BOUNDARIES – Use UK construction terminology. – Separate confirmed facts from assumptions. – Quote document references where available. – If two sources conflict, show both. Do not choose one. – Do not invent dates, values, drawing numbers, contractual rights or technical requirements. EVALUATION / ESCALATION Before finishing: 1. List missing information. 2. Flag anything requiring commercial, design, safety or contractual review. 3. Check every factual claim can be traced to the supplied material. PROJECT INFORMATION [Paste or attach documents here]

This framework is deliberately boring. That is a strength. A company-wide AI system should not depend on every employee remembering a clever prompt-writing trick.

Give AI Examples Before You Ask for Output

If you want AI to produce documents that sound like your business, show it two or three examples that your team already considers good.

Examples teach structure faster than adjectives. “Professional but friendly” is subjective. Two approved client updates show the model exactly how your business opens the message, describes progress, handles delays and closes with next actions.

Here are two APPROVED examples of our weekly client update format. EXAMPLE 1 [Paste approved example] EXAMPLE 2 [Paste approved example] Create this week’s update using the new site notes below. Match the examples for: – Structure – Heading pattern – Level of detail – Tone – Approximate length Do NOT copy project facts, dates, names, costs, delays or commitments from the examples. If the new notes do not contain information needed for a section, write: “Information required: [what is missing]” NEW SITE NOTES [Paste notes]

10 AI Prompts for UK Construction Companies That Are Actually Reusable

These are structured starting points. Replace the bracketed information and, where possible, attach the actual source documents.

1. Tender pack: build a compliance and action matrix

Review the attached UK construction tender pack. Create a tender compliance and action matrix with: Requirement | Source document | Page/clause | Required response/evidence | Owner | Deadline if stated | Status | Risk/observation Rules: – Do not infer compliance. – Do not invent deadlines, accreditations, insurance levels or project experience. – Flag conflicts between documents. – Separate mandatory requirements from desirable criteria where the documents do so. Finish with: – 10 highest-priority actions – Clarification questions – Missing evidence/documents

2. Tender revision: identify what changed

Compare VERSION A and VERSION B of the supplied tender information. Create a change register: Changed requirement | Old version | New version | Source reference | Impact category | Action required | Reviewer required Impact categories: Scope / Cost / Programme / Design / Procurement / H&S / Admin Do not calculate financial impact unless rates and quantities are supplied. Do not decide contractual entitlement. Prioritise changes that could alter our price, exclusions, programme or delivery obligations.

3. RFI: turn a site question into a clean request for information

Draft an RFI from the site issue below. Audience: [Architect / engineer / consultant / client team] Include: – RFI title – Location – Relevant drawing/spec references – Observed issue – Exact clarification required – Why the answer is needed – Programme impact only if explicitly supplied – Required response date only if supplied Do not invent a technical solution. If our notes suggest one, label it: “Contractor suggestion for design-team consideration.” SITE INFORMATION: [Paste notes]

4. Variation: organise evidence before QS review

Prepare an INTERNAL variation evidence summary for QS review. Sections: 1. What changed 2. Who communicated/instructed it, with date 3. Original scope evidence 4. Revised scope evidence 5. Cost evidence available 6. Programme evidence available 7. Missing records 8. Contradictions 9. Questions for the QS/contracts manager Do not conclude that we are entitled to payment or time. Do not interpret the contract unless the relevant clause is supplied. Do not invent causation or delay.

5. Subcontractor quotation comparison

Compare the attached subcontractor quotations against the supplied scope. Compare: – Price – VAT treatment if stated – Included scope – Exclusions – Qualifications – Provisional sums/allowances – Lead time/programme – Payment terms – Warranties – Design responsibility – Missing information – Apparent scope gaps Then create a NORMALISATION LIST: What must be clarified before these bids can be compared like-for-like? Do not recommend the cheapest bidder automatically. Do not infer an inclusion because it is common practice.

6. Site diary: turn rough notes into a structured record

7. Weekly client progress update

Create a weekly client update from the approved programme information and site notes. Format: 1. Executive summary 2. Progress this week 3. Planned work next week 4. Decisions/information required 5. Key procurement items 6. Risks/issues already recorded 7. Photos to include Tone: calm, factual and non-defensive. Do not: – promise dates not in the approved programme – assign blame – hide recorded delays – invent percentage complete – create a definitive contractual position

8. Meeting minutes: separate decisions from discussion

Turn the meeting transcript/notes into draft minutes. Separate: – Decisions made – Actions – Information required – Risks/issues discussed – Items discussed but NOT decided Action table: Action | Owner | Due date | Source wording Never invent an owner or due date. If unclear, write “Owner to confirm”. Do not turn suggestions into decisions. Flag anything that could create a contractual, design or safety commitment.

9. Construction phase plan or RAMS: use AI as a gap finder

Under CDM 2015, the responsible contractors and principal contractors retain duties to plan, manage and monitor construction work. AI can assist with document checking; it does not take on those duties.

Review the supplied [construction phase plan / RAMS / site procedure] as a DOCUMENT CHECK. Compare only against: 1. Our supplied company checklist/template 2. Supplied project information 3. Any guidance/regulatory excerpts I provide Output: Section | Information present | Information missing | Generic wording needing project detail | Contradictions | Outdated items | Question for competent reviewer Do NOT: – certify compliance – invent control measures – approve a method of work – replace the responsible competent person’s review End with: “Human review required for…”

10. Job-cost or project-data review

Analyse the supplied project cost data as an INTERNAL management review. First, describe the available fields and data limitations. Then calculate where supported: – Budget vs actual by cost code – Committed cost – Forecast variance – Largest changes – Unusual entries requiring checking – Missing data – Trends worth investigating Show the calculation/source fields behind every numerical conclusion. Do not invent missing values or assume the cause of an overrun.

Prompts by Role: What Should Each Team Member Use AI For?

Role	Good AI tasks	Keep human-led
Estimator / Bid Manager	Tender matrices, addendum comparisons, evidence lists, first-draft answers	Final price, exclusions, bid commitments
Quantity Surveyor	Quotation comparison, change chronology, cost-data analysis, missing-evidence checks	Entitlement, valuations, settlement positions
Project Manager	Progress reports, meeting actions, document summaries, decision logs	Programme commitments, contractual notices, project decisions
Site Manager	Diary formatting, RFIs, handover lists, note organisation	Safe systems of work, technical approval, site control decisions
Office / Admin	Call summaries, scheduling drafts, email formatting, document classification	Sensitive HR decisions, financial approval, out-of-policy promises
Marketing / BD	Case studies, award-entry drafts, project profiles, channel adaptation	Unverified claims, client permissions, invented outcomes

The Four Things Your Prompt Should Never Let AI Quietly Decide

1. Safety and method

Use AI to organise approved information or expose gaps—not to outsource competence or responsibility for safe systems of work.

2. Contractual entitlement

AI can build a chronology from emails, instructions and site records. It should not decide that an event creates entitlement to additional money or time.

3. Technical compliance

A model can compare supplied specifications, schedules and notes. It should not invent a design solution or certify that work complies with regulations, standards or designer intent.

4. Facts the project record does not contain

If the source does not say it, the model should not upgrade it into fact.

TRUTH CHECK Classify each material statement as: – CONFIRMED: directly supported by supplied information – INFERENCE: reasonable interpretation, but not explicitly stated – MISSING: information needed before a conclusion can be reached – CONFLICT: supplied sources disagree Do not hide inference inside normal prose.

Use a Two-Pass Workflow for Important Construction Documents

Pass 1

Extract and organise

Identify facts, source references, gaps, conflicts and questions. Do not draft the finished document yet.

Pass 2

Draft from approved facts

After a team member checks Pass 1, return the approved fact set to AI and request the final email, report, RFI or summary.

How to Build a Prompt Library Your Construction Company Will Actually Use

Start with recurring work

Tender review, client updates, meeting minutes, site records, quotation comparison or lead follow-up.

Attach approved examples

Use real templates and strong historical outputs, with project-specific information removed where necessary.

Add truth boundaries

For each workflow, define what AI may infer, what it must never infer and who signs off the output.

Version prompts like procedures

Give useful prompts an owner, version and review date.

Measure whether they save work

The goal is reduced rework, better consistency and earlier identification of missing information.

Do not build the library around one AI product. Store the process, inputs, examples, constraints and review rules separately from ChatGPT, Claude or whichever interface your team uses today.

What Information Should You Avoid Putting Into an Unapproved AI Tool?

Construction teams routinely handle commercially sensitive and personal information: tender pricing, subcontractor rates, client details, employee information, access arrangements, drawings and project correspondence.

Your company should decide which AI services and accounts are approved before staff upload this material. Access permissions, data handling, retention, contractual confidentiality and internal policies matter just as much as prompt quality.

If you would hesitate to email the document to an unknown external supplier, do not paste it into an AI tool until your company has approved that tool and use case.

One Master Prompt You Can Adapt to Most Construction Tasks

I work for a UK construction business. TASK [What you need done] AUDIENCE [Who will read/use the result] SOURCE OF TRUTH Use only: [List attachments, project documents, approved templates or data] PROJECT CONTEXT [Project, package, stage, role and relevant background] OUTPUT [Exact headings/table/format required] COMPANY STYLE [Add two approved examples if available] BOUNDARIES – Use UK construction terminology. – Do not invent facts, dates, costs, drawing references, programme commitments or approvals. – Do not infer contractual entitlement. – Do not invent technical or safety controls. – Separate confirmed facts from assumptions. – Highlight conflicting source information. – Mark missing information clearly. QUALITY CHECK 1. Trace material claims to the supplied source. 2. List assumptions. 3. List missing information. 4. Flag anything requiring QS, contracts, design, H&S, project-management or other competent-person review. 5. Remove statements that sound certain but are unsupported. Now complete the task.

How Prompting Fits Into a Wider Construction AI System

Prompts are useful, but prompting alone is not automation. The bigger opportunity comes when approved instructions are combined with controlled access to the right company information and workflows.

For practical use, see what actually works with ChatGPT for trades and the four-step method for making AI useful in construction.

Teams that want consistent company context can also read about bringing existing ChatGPT knowledge into Claude workflows and our Claude AI guide for construction companies.

For company-wide implementation, see our AI training for construction teams and AI solutions for construction and trades.

Frequently Asked Questions

What are the best AI prompts for construction companies?

The best prompts are tied to a repeatable business process and trusted source material. Useful examples include tender compliance matrices, RFI drafting, subcontractor quotation comparison, site diary formatting, meeting actions, client progress updates and variation evidence summaries.

Can UK builders use ChatGPT or Claude for construction admin?

Yes. Both can assist with administrative and information-processing tasks when the business supplies suitable context and reviews the output.

Can AI write RAMS or a construction phase plan?

AI can help organise approved project information, compare a document against a company checklist and identify missing sections. It should not be treated as the competent person responsible for deciding safe systems of work or certifying compliance.

Can AI review a JCT contract?

AI can help locate, summarise and compare clauses when the relevant contract text is supplied. It should not replace professional contractual or legal review.

How do I stop AI inventing construction information?

Define the permitted sources, prohibit unsupported assumptions, require missing information to be labelled and ask it to trace important claims back to supplied documents.

Should construction companies create a central prompt library?

Yes, if it is small and process-focused. Store approved examples, source requirements, constraints and sign-off rules alongside each prompt.

Are AI prompts confidential?

Do not assume information is appropriate to share simply because it can be pasted into an AI chat. Use approved tools and accounts, and consider confidentiality, personal data, permissions and retention.

Do I need a different prompt for every construction job?

No. Keep a stable structure and change the project-specific inputs. The SITE method is designed to work across different projects and AI platforms.

Sources and Further Reading

Turn good prompts into a company system

Construct Virtual can help construction teams turn recurring admin, tender, project and customer workflows into controlled AI processes—with examples, guardrails and human sign-off built in.

Explore AI Training

Practical systems for real construction work.