
In short
AI in construction, for a UK SME, mostly means using general assistants such as ChatGPT, Claude, Copilot and Gemini to take the writing, searching and admin out of the office. It saves time today on tender answers, first-draft RAMS, estimate checks, site paperwork, follow-up emails and researching prospects. It can't sign off safety, replace a quantity surveyor's judgement or set your strategy. PKB Civils, a UK concrete yards contractor, rolled Claude out across its business in 2026 and is now building its own tools with it. The safety rule: keep client, personal and pricing data out of free accounts, and have a competent person check everything.
On this page
- How can I use AI in my construction business?
- What a UK civils contractor actually did with AI
- Practical ways to use ChatGPT and Claude, with prompts you can copy
- Which AI tools are worth paying for?
- AI in civil engineering and for quantity surveyors
- The other side of AI: how buyers now find contractors
- Is AI safe to use with client and site data?
- What AI won’t do for you
- What it costs to get started
- Where to start this month
AI in construction, for a UK SME, mostly means using general AI assistants such as ChatGPT, Claude, Microsoft Copilot and Gemini, plus a few specialist tools, to take the writing, searching and admin out of bids, estimates, site paperwork and follow-up. People stay in charge of pricing, safety sign-off and client relationships.
I’ve spent twenty years in marketing, and I run an AI-first business that works only with trades and construction firms. Some of what you’ll read about AI in construction is overhyped, and I’ll say which bits. This guide covers where AI earns its keep in a contractor’s office, what one UK civils firm has done with it, prompts to copy, which tools to pay for, what’s safe, what it costs and where to start.
What is AI in construction? It’s software that writes, summarises, searches and spots patterns in text, images and data. Construction firms use it for bids, estimates, documents, site records, marketing and finding leads. For most UK contractors, the first use isn’t site robotics or drones. It’s a general assistant such as ChatGPT or Claude, working from the firm’s own documents.
How can I use AI in my construction business?
Start with the jobs your office repeats every week. AI is good at first drafts, summaries and searching through documents, so the best early wins are the tasks where someone rewrites the same thing again and again. The table below is our rough guide to where it earns its keep, and what has to stay with a person.
| Job | What AI does | Which tool | Rough setup | Risk | What stays human |
|---|---|---|---|---|---|
| Tender and PQQ answers | Drafts answers from your past bids and policies | ChatGPT, Claude or Copilot, with a project of past bids | An afternoon to gather the bid library | Medium | The final answer, every claim checked, the price |
| RAMS and method statements | Writes a first draft from your standard RAMS and the site details | Any general assistant | A morning to load your standard documents | High | Review and sign-off by a competent person |
| Estimating checks | Compares a new schedule with past won and lost jobs, flags gaps | Claude or ChatGPT with your past data, or estimating software | Days to weeks to tidy past job data | Medium | The rates, the margin, the decision to bid |
| Site diaries and daily reports | Turns voice notes and photos into a tidy written record | Any assistant on a phone or tablet | Same day | Low | Checking facts, times and names |
| Client and supplier emails | Drafts replies, chasers and summaries of long threads | Copilot in Outlook, or any assistant | Same day | Low | Tone, commitments and anything contractual |
| Case studies and project pages | Writes a project story from job notes, photos and the client’s comments | Any assistant | An hour per project | Low | Accuracy, client permission, the photos |
| Finding and researching prospects | Summarises what a company is doing before you call, watches for trigger signals | An assistant with web search, or a signals tool | An hour per prospect, less with practice | Low | The call, the relationship, the judgement on fit |
| Answering enquiries fast | Drafts a reply or a first response to a website enquiry | AI receptionist or enquiry tools, or an assistant | Days | Medium | Anything about price, programme or safety |
| “Ask the business” knowledge | Answers staff questions from your own policies and documents | A shared Claude or ChatGPT project, or Copilot over SharePoint | Weeks, then ongoing | Medium | Owning it and keeping it up to date |
If you’re not sure where to begin, pick the low-risk rows that happen most often: site diaries, emails and case studies. They teach your team how to brief an AI tool well, and a mistake there is cheap to catch.
What a UK civils contractor actually did with AI
PKB Civils builds heavy-duty concrete yards for depot operators and facilities managers across the UK. When I started working with them in October 2024, new work came from the two founders’ relationships and their black book of contacts. That’s a familiar picture in civils: the people who should be leading delivery are also chasing the next job and answering every question in the office.
The order matters, because AI came last. First came a strategy workshop with the directors in October 2024 and a plan the marketing team could act on. A rebuilt website went live in January 2025, paid search started in February 2025, and there was a filming day in January 2026. Only then did AI arrive in earnest: a new AI lead generation tool in June 2026, Claude Teams rolled out across the business in July 2026, and a new AI-powered website signed off in October 2026, with the next build already under way.
I think that order is the first lesson. AI made PKB faster at things it already knew how to do. It didn’t decide what those things should be.
Claude across the office, so the answers aren’t stuck in one person’s head
PKB adopted Claude Teams in July 2026 and rolled it out across the business, so that experienced, well-trained people can get more done. That’s the case for starting with a general assistant: the paperwork in a civils business is relentless, and most of it is a variation on something you’ve written before.
The bigger piece of work is a company brain. PKB is building one with Claude, and it’s still in progress. The idea is simple. In a lot of growing contractors, the founders are the only people who know where everything is and what the answer is, so every question goes through them. A company brain puts the policies, accreditations, standard documents and past answers in one secure place that anyone on the team can ask, so the business relies less on the founders for the routine stuff.
If you want to try the same thing, these are the first ten documents I’d put into a shared AI project:
- Your company facts: what you do, where, for whom, and your size, written plainly.
- Every accreditation, with its level and expiry date (CHAS, Constructionline, SMAS, ISO).
- Your health and safety, quality and environmental policies.
- Your standard RAMS for the tasks you do most.
- Your best past tender answers.
- Your capability statement.
- Your insurances and their limits.
- The assumptions behind your prices, not the prices themselves.
- The questions clients and staff ask most, with the agreed answers.
- Your project list, with the client, scope and outcome for each job.
The honest limit is that a brain has to be fed. Someone has to own it, or it fills up with old versions and junk. Every new accreditation, changed policy and updated answer has to go in, so budget time to keep it current.
Building their own tools with AI
PKB is also building its own tools with AI to make the business run more efficiently. Until recently that was out of reach for a firm this size: your own internal software meant a development team and a long project. Now a small team can prototype a working tool in a chat, test it on real jobs and improve it as they use it.
If you want to do the same, here’s the order that keeps it safe:
- Prototype in a chat. Describe the job and the steps to Claude or ChatGPT and let it build a rough working version you can click through.
- Connect it to where your files already live. For a lot of contractors that’s SharePoint, so the tool reads and saves documents there rather than in yet another system.
- Harden it before real use. Proper sign-in, user roles, backups and someone who checks it does what it should.
- Name an owner. Every tool you build needs a person who fixes it when it breaks and decides what changes.
A home-made tool still needs looking after, and security matters more than speed. If it holds client documents or prices, sign-in and backups come before anything clever.
An AI-built website the team can update
PKB’s new AI-powered website was signed off in October 2026. For a contractor, the point is speed. On an old-style website, adding a project page can be enough of a chore that it simply doesn’t happen. A site the team can update with AI makes it far quicker to turn job notes and photos into a page, so more of your work gets published.
That matters because, as our research below shows, AI tools favour firms whose pages say plainly what they do and where, and Google’s AI Overviews cite pages that answer a search plainly, such as a project with its scope and value. The other side of the coin is that speed isn’t taste. Someone with an eye still needs to look at every page and say “that’s not quite us”.
Finding leads with trigger signals
PKB added an AI lead generation tool in June 2026. Tools like this, including our own Lead Engine, work on trigger signals: they watch for signs that a business is likely to be in the market, match the company to the right person to contact, and draft an opening message with a real reason to get in touch. You still send it yourself, in your own words.
You don’t need special software to start using signals. Write down three real-world events that mean a business might need your work soon. For a yards or groundworks contractor that could be:
- a planning application for a new depot, warehouse or yard extension
- a recent contract win that means a client needs more space
- an announcement that a company is expanding or opening a new site
Then, before you pick up the phone, ask an AI assistant with web search to research that one company (the prompt is in the next section), so your first call is about their world, not your brochure.
Practical ways to use ChatGPT and Claude, with prompts you can copy
Each workflow has a prompt to copy, what to give the tool first and what to check before anything leaves the building. Use a business or team account for anything involving client or company information.
Tender question answers from your bid library
Give it your past tender answers, your policies and the question, word for word. (Our guide to construction tender marketing covers what buyers look for.)
Prompt: “You are a bid writer for a UK civils contractor. Using only the past tender answers and policies I’ve attached, draft an answer to the question below in no more than 400 words. Use the buyer’s scoring criteria as headings. Where you need a fact that isn’t in my documents, write [CHECK] instead of guessing. Question: [paste the question]. Scoring criteria: [paste them].”
Check: every [CHECK], every number, every named project and every accreditation date. The tool will sound confident even when it’s wrong.
A first-draft RAMS or method statement
Give it your standard RAMS for the task, the site details and the hazards you already know about.
Prompt: “Using my standard RAMS attached as the template, draft a site-specific risk assessment and method statement for [task] at [site]. Keep my format and headings. List any hazards from my site notes that the template doesn’t cover. Don’t remove any control measure from the template. Mark anything you’re unsure about as [REVIEW].”
A competent person reviews and signs every RAMS. The duties under CDM 2015 stay with the dutyholders. Northern Ireland has its own CDM regulations (NI) 2016.
A quote covering letter, and a sanity check on a schedule
Give it the schedule, the scope, your exclusions and your standard terms. Never let it set your prices.
Prompt: “Here is my priced schedule and scope for [job]. First, list anything in the scope that doesn’t appear in the schedule, any quantities that look inconsistent, and any exclusions I haven’t stated. Don’t change any rates. Then draft a one-page covering letter in plain English that explains what’s included, what isn’t and how long the price is valid.”
Check: the gaps it flags against the drawings yourself. It can spot a missing line. It can’t tell you whether your rate is right.
A site diary or daily report from a voice note and photos
Record a short voice note at the end of the day and add a few photos.
Prompt: “Turn this voice note and these photos into a daily site report with these headings: date, weather, labour on site, plant on site, work completed, deliveries, delays and reasons, health and safety observations, actions for tomorrow. Don’t add anything I didn’t say. Flag anything that sounds like a variation or a delay so I can record it properly.”
Check: names, times and anything that might matter in a dispute later.
A project case study from job notes
This is the job most contractors put off. One page per project, with the client, scope, value, programme and outcome, is step six of the ten in our report (“turn projects into answers”), and in our audits firms with more project pages scored much higher on tender readiness.
Prompt: “Write a 300-word project case study for my website from these notes. Use these headings: the client and what they needed, the scope, the programme, how we did it, the outcome. Title it the way a buyer would search for this kind of job, for example ‘heavy-duty concrete yard for a logistics depot in Kent’. Don’t invent any figures or quotes.”
Check: the client is happy to be named, and the photos are yours to use.
Client and supplier emails, including chasers
Prompt: “Draft a polite but firm chaser to [supplier] about [late delivery or overdue payment]. Reference the order number and the date we agreed. Keep it under 120 words, and suggest two dates for a call. Don’t promise anything on my behalf.”
Check: anything that commits you to a date, a cost or a change in scope. That’s contract territory.
Prequalification and capability statement drafts
In our research, only 2 of the 42 contractors we checked offered a capability statement a buyer could download from their website. That’s an easy gap to close.
Prompt: “Using my company facts, accreditations, policies and project list attached, draft a two-page capability statement for a procurement team. Include who we are, what we do and where, our accreditations with expiry dates, our insurances, five relevant projects and a named contact. Flag anything that’s missing.”
Check: every accreditation is current and every project is one you’re allowed to name.
Researching a prospect before a call
Prompt: “Research [company name] in [town] using the web. Tell me what they do, where they operate, any recent news, planning applications, expansions or contract wins, who leads operations or facilities, and one question I could ask that shows I understand their situation. List your sources so I can check them.”
Check: the sources. AI research is a starting point for a conversation, not a fact file. For the wider picture, see where contractors’ best work comes from.
Which AI tools are worth paying for?
One good general assistant on a business plan, used well, beats five specialist subscriptions nobody opens. Here’s how the main ones differ.
- ChatGPT (OpenAI) is the one most people have tried: a good all-rounder for writing, research and files. OpenAI says its ChatGPT Business plan never uses your data to train its models (OpenAI).
- Claude (Anthropic) is strong on long documents and careful writing, and its Team plan shares projects across the team. Anthropic lists no model training on your content by default on Team and Enterprise, and an opt-out on individual plans (Anthropic).
- Microsoft Copilot is built into Microsoft 365, so it suits firms that already run on Outlook, Teams and SharePoint.
- Gemini is Google’s assistant, and the natural fit if your business runs on Google Workspace.
Whichever you choose, use a business or team plan and check the data settings yourself.
Then there are the specialist tools. Estimating and takeoff software with AI built in can measure from drawings and compare rates; look at it once you know where your estimating time goes. AI receptionists answer calls and website enquiries when the team is on site. And many job management tools now have AI built in, so check what yours already does before buying anything new.
The tool matters less than having a named owner and a clear job for it. Buying licences for everyone and waiting for something to happen is the expensive way to learn that.
AI in civil engineering and for quantity surveyors
In civil engineering, AI helps most with the reading and writing around the work, not the engineering. It’s useful for reviewing long specifications and pulling out the clauses that matter, checking a bill of quantities for missing items or inconsistent units, comparing subcontractor tender returns side by side, turning employer’s requirements into a one-page brief for the site team, and drafting programme narratives and progress reports.
For quantity surveyors, it’s a fast reader and a decent first checker. Give it two subcontractor quotes and the scope, and it’ll tell you where they differ.
Where it doesn’t help matters just as much. It can’t make measurement judgements, it doesn’t carry commercial risk, and it shouldn’t be your interpreter of NEC or JCT clauses when money is at stake. It will also state wrong facts with total confidence. Use it to get to the question faster, then let the person with the experience answer it.
The other side of AI: how buyers now find contractors
AI isn’t only a tool for your office. It’s also how your next client finds you, or doesn’t.
In our Construction Visibility Report 2026, we audited 50 UK construction, civils and trades businesses between 20 October and 4 November 2026, put 2,717 buyer questions to six AI tools and ran 408 Google searches. Here’s what a buyer sees:
- AI tools named the business in 30% of answers to buyer questions about its own work and area (809 of 2,717). ChatGPT named firms most often, in 39% of its answers, and Google AI Mode least, in 20%.
- Google showed an AI Overview on 71% of contractor searches (220 of 312), but cited the contractor’s own website in just 4% (14 of 312).
- 35% of AI descriptions of a contractor got something wrong or left a gap. Only 8 of the 42 contractors were described accurately by all six tools.
- The ten most-named contractors had a median of 1,486 websites linking to them, against 103 for the ten least-named.
So a buyer who asks ChatGPT for a groundworks contractor in your area may get a shortlist without you on it, or a wrong description of you. The report is a snapshot: we checked each business once, and the comparisons show what goes together, not what causes what.
Fixing that is a different job from everything above. Our guide on how to get your firm named by ChatGPT covers it step by step, and the report ends with ten steps in order.
Is AI safe to use with client and site data?
Yes, if you set it up properly and keep a few rules.
The NCSC advises against putting sensitive information into queries to public AI tools, or anything that would cause problems if it were made public. It points out that the provider can see what you type, and that these tools can get things wrong and “hallucinate” incorrect facts. If you’re putting personal data into an AI tool, data protection law still applies, and the ICO’s guidance on AI and data protection is the place to check what that means for you. The ICO notes the guidance is under review after the Data (Use and Access) Act.
Here are the rules I’d put in place before anyone on the team uses AI for work:
- No free personal accounts for company work. Use a business or team plan, and check the data and training settings yourself.
- Keep personal data out unless you’re sure. That covers employee details, clients’ personal information and anything from a medical or HR file.
- Be careful with drawings and pricing. Client drawings and commercially sensitive tender prices only go into tools and plans you’ve checked.
- A person checks everything that leaves the building. Anything going to a client, a regulator or a site is read by someone who knows the job.
- AI drafts, a competent person signs. RAMS, design, safety documents and anything contractual.
- Write it down. A one-page AI use policy that says which tools, which accounts, what’s off limits and who to ask.
None of this needs a lawyer to get started. It needs someone to own it.
What AI won’t do for you
AI is good at first drafts. It’s poor at the parts that make your business yours.
Strategy. Ask an AI tool for a marketing or growth strategy and you’ll get something that reads well and could belong to any firm. It doesn’t know your clients, your margins or why your best customers chose you over the competition, and good strategy is as much about what you choose not to do. That still needs people who know the business.
Pricing judgement. It can check a schedule for gaps. It can’t tell you whether a job is worth winning at that price.
Relationships. Civils work is still won through trust built over years. AI can help you prepare for a conversation. It can’t have it for you.
Personality. Humour, opinion and the way your team actually talks don’t come out of a prompt.
Accuracy. It will state wrong facts with confidence. In our report, the most common mistake AI tools made about a contractor was its size, wrong or missing for 25 of the 42 firms. Whatever AI tells you, or tells a buyer about you, check it. And any company brain goes stale if nobody feeds it.
Will AI replace quantity surveyors or tradespeople? On what we can see today, no. It takes away some of the reading, writing and admin, which changes the job, but the judgement, the responsibility and the work on site stay with people.
What it costs to get started
Here’s an honest picture, with no invented savings.
- Free accounts. Fine for testing what the tools can do with made-up examples. Not for client or company data.
- Business and team plans. Priced per user per month. When we checked on 16 November 2026, Claude listed its Team plan at US$20 per standard seat a month billed annually, or US$25 billed monthly, before tax (Anthropic). ChatGPT Business and Microsoft Copilot are also priced per user, and prices change often, so check each vendor’s own page on the day you buy.
- Your team’s time. The real cost is the hours it takes to gather your documents, set up shared projects and learn to brief the tools well. Hands-on training shortens that.
- Custom tools and automation. Anything built for your business, from an enquiry handler to your own estimating tool, is usually priced per project.
I’m not going to give you an hours-saved figure, because I don’t have a measured one I’d stand behind. Start small, on one repetitive job, and build from there.
Where to start this month
Here’s a 30-day plan for a contractor starting from scratch.
- Pick one repetitive job. Site diaries, chasers or case studies are good first choices.
- Set up a business account on one general assistant for the people who’ll use it first.
- Build your shared document set. Start with the ten documents listed in the PKB section above.
- Write a one-page AI use policy using the six rules in the safety section.
- Test three buyer questions about your own firm in ChatGPT and Claude, for example “who are the best groundworks contractors in [your region]?”. Write down who gets named and what they say about you.
- Review at the end of the month. Keep what worked, drop what didn’t, and pick the next job.
If step five worries you, start with our free tender-ready audit. It gives you two scores out of 100, checked on Google and six AI tools, so you can see how buyers and AI tools see your firm today. If you’d rather talk it through, book a growth call: 30 minutes with me, and I’ll tell you plainly where I’d start in your business.
Frequently asked questions
How is AI used in the construction industry in the UK?
Most UK contractors start with general AI assistants rather than robots or drones. They use ChatGPT, Claude or Microsoft Copilot to draft tender answers from past bids, write first drafts of RAMS and method statements, tidy site diaries, write client and supplier emails, check estimates against past jobs and research prospects. The firms getting the most from it give the tools their own documents to work from and keep a person checking every output.
What is the best AI for construction companies?
There isn't one best tool. For most firms the choice is between ChatGPT, Claude, Microsoft Copilot and Gemini, and the best one is usually the one that fits the systems you already use, on a business plan that keeps your data out of model training. Specialist estimating, takeoff and enquiry tools come later, once you know which job you're trying to speed up.
Can ChatGPT write a method statement or RAMS?
It can write a useful first draft if you give it your standard RAMS, the site details and the task. It can't make the document right. A competent person has to review it, correct it and sign it, and the duties under CDM 2015 (or the Northern Ireland CDM regulations) stay with the people who hold them. Treat the draft as a starting point, never as the finished document.
Will AI replace quantity surveyors?
Not on the evidence so far. AI tools are quick at reading documents, comparing tender returns and checking a bill of quantities for gaps, but they get facts wrong with confidence and they don't carry commercial risk. Measurement judgement, contract interpretation and negotiation still need a person. The job is more likely to change than disappear, and the same goes for tradespeople on site.
Is it safe to put company information into ChatGPT or Claude?
It depends on the account and the information. The NCSC advises against putting sensitive information into public AI tools. Business and team plans from OpenAI and Anthropic say they don't train their models on your content by default, but you should still check the settings, keep personal data and commercially sensitive pricing out unless you're sure, and write a short policy so your team knows the rules.
How much does it cost to start using AI in a construction business?
You can test the main tools on free accounts, but don't use those for client or company data. Business and team plans are priced per user per month: when we checked on 16 November 2026, Claude's Team plan was listed at US$20 per seat a month billed annually, before tax. Add the time to set up your documents and train your team. Custom tools and automation are priced per project.


