The AI visibility audit for construction subcontractors: what ChatGPT says about you and how to change it

Open ChatGPT, Perplexity or Google AI Mode and ask which groundworks subcontractors in your region have worked for a Tier 1 contractor. It will name three or four firms, with a sentence about each. Read the names. If yours is not there, the quantity surveyor who asked the same question last week did not see you either, and the invitation to tender went to someone it could describe.

That is AI visibility: whether the tools buyers now use to shortlist suppliers can find your firm, say what it does, and verify what it claims. This guide shows you how to audit yours in an hour, what the answers mean and what to fix first.

Why buyers are asking AI tools, and what they ask

A procurement team assembling a supply chain for a bid has a trade, a region and a value in mind and not much time. Five years ago they searched Google and opened ten websites. Now a growing share type the question into an AI assistant and get a shortlist in one answer. The questions are specific: which steel fabricators in Yorkshire hold EN 1090 execution class 3, which M&E contractors in the Midlands are SSIP accredited and have delivered NHS work, which demolition contractors in London are NFDC members.

The assistant does not know who is best. It knows who it can verify. It reads websites, Companies House, the Constructionline and CHAS registers, Google Business Profile and anything else public, and it names the firms where those sources agree and say something concrete. A site that shows accreditations as logos, has one page for every trade and no case studies with values gives it nothing to work with.

The hour-long audit

1. Write down ten questions a buyer would ask about your trade and region. Use their language: “subcontractor”, the trade, the region or nearest city, and a qualifier such as “SSIP accredited”, “worked for a Tier 1” or “delivered education projects”. Two or three should name a sector you want more of.
2. Ask each question in ChatGPT, Perplexity and Google AI Mode. Use a private window so your history does not shape the answer. Record who is named and what is said about them.
3. Ask the three tools directly about your firm by name. Note what they get right, what they get wrong and what they leave out. Wrong addresses, old directors and lapsed accreditations are common.
4. Open the pages the tools cite for the firms they named. Look at how those pages present accreditations, case studies and contact details. That is your benchmark.
5. Score yourself. Named in how many of the thirty answers? Described accurately? Cited from your own site, or from a directory? Most subcontractors we audit are named in none and described from Companies House alone.

What the answers tell you

If the tools name competitors and describe them in detail, those competitors have written their credentials out as text, kept their business details consistent and published case studies with substance. If the tools name nobody useful, the opportunity in your trade and region is wide open. If they name you but get the facts wrong, your details disagree between sources and the tool picked the wrong one.

The six fixes, in order

1. Write every accreditation out. CHAS, Constructionline, SafeContractor or SMAS with the level and expiry, ISO certificates with the certification body, insurance with the limits, each linked to its public register. On one page, in text, not in a logo strip.
2. Match your business details everywhere. Name, registered number, address and phone identical on your website, Companies House, Constructionline, CHAS and Google Business Profile. Tools compare them and downgrade firms that disagree with themselves.
3. Give each trade its own page. A groundworks page, a drainage page, an RC frames page, each with scope, typical values and sectors. A tool asked for an RC frame contractor cannot cite a page about everything.
4. Publish case studies with substance. Client, main contractor, sector, value band, programme and scope. The tools quote these, and so do bid writers.
5. Add structured data. Schema markup that states your organisation, accreditations, services and address in a form machines read directly, and an llms.txt file that summarises the site for AI assistants. Almost no subcontractor sites have either.
6. Earn reviews that say what you did. A Google review from a contracts manager that mentions the trade, the project and the sector is evidence. “Great lads” is not.

How long it takes to show

Structural fixes such as accreditations in text and structured data, and a corrected Google Business Profile, tend to appear in AI answers within four to eight weeks. Being named consistently for competitive questions builds over three to six months as pages and case studies accumulate and the tools re-read your site. It is worth tracking monthly: the same ten questions, the same three tools, a record of who is named.

What to do this week

• Run the ten-question audit and keep the results. It is the baseline you will measure against.
• Run the free Tender Check on your website. Its Verified stage scores exactly the things the AI tools look for first.
• Fix the accreditations page. It is the cheapest change and the one the tools weigh most.

How many buyers are actually doing this

This is no longer a prediction. Forrester’s 2026 Buyers’ Journey Survey of around 18,000 business buyers, published in January 2026, found 94% had used AI during their most recent purchase and that twice as many named it their most meaningful research source as named any other. Gartner’s 2026 survey is more conservative at 45% using generative AI in a recent purchase, alongside about seven other sources. G2’s March 2026 research found half of buyers now start supplier research in a chatbot rather than a search box, and 69% chose a different supplier from the one they had in mind because of what the chatbot said.

Those are cross-sector figures and construction procurement is slower to change than software buying. But the people doing the shortlisting in a Tier 1 bid team are the same people using these tools at home, and the behaviour arrives with them. The practical test is the one in this guide: ask the questions yourself and see who is named. The Tender Check scores the same things in its Found and Verified stages, and the website management plan re-runs the ten questions every month so you can see the answer move.