Back to Insights

Specified By AI: Building Products & AI Search.

Specifiers use AI tools for the early, time-consuming part of product research: narrowing a category down to a handful of credible options. The manufacturers that get named are the ones whose technical information AI tools can find, read and trust.

There is growing evidence that more and more specifiers are using AI tools during their research. The shortlist that comes back is formed before any manufacturer knows the project exists.

This is moving a comparison point against alternative products and competitors earlier. Before any CPD session or sales call, the first contest happens inside the AI tool: whether it names the product at all, and whether what it says about it is accurate.

Across 495 prompts we tracked for construction and property businesses, only 4.3% of the 1.3 million citations in AI answers pointed to the businesses’ own websites. The other 96% went to third parties.

What specifiers ask

Specifiers use AI tools for the early, time-consuming part of product research: narrowing a category down to a handful of credible options. Their questions are technical and specific:

  • “Which external wall insulation systems have been tested to BS 8414 for use on residential buildings?”
  • “What are the A1 or A2-s1,d0 rated cladding options for a residential building over 18 metres?”
  • “Which acoustic floor systems achieve the Approved Document E performance standard for separating floors in timber frame?”
  • “What are the alternatives to [product] with similar fire performance?”

Each of these needs an answer built from performance data, test certificates and compliance information. The manufacturers that get named are the ones whose technical information AI tools can find, read and trust.

Where product information falls down

Most manufacturers hold the information specifiers need. The problem is how it’s published.

The biggest issue is that technical data sits in PDFs. Datasheets, test reports and installation guides are often available only as downloads, which AI tools read inconsistently or not at all, while the product page itself carries a paragraph of marketing copy and a download button. Where those downloads are gated behind an email form, AI crawlers can’t see them at all.

Our own tracking reflects this. PDFs and downloadable documents made up under 1% of the citations to these businesses’ own sites, while product and service pages took around two-thirds. The one exception was ChatGPT, which regularly cited a single business’s well-organised PDF library. Documents can be cited, but only when they’re accessible and structured properly.

Consistency is the next problem. The website, NBS listing, BIM objects and distributor sites can each show different figures for the same product, usually because one was updated and the others weren’t. An AI tool has no way of telling which is current, and specification platforms carry real weight: NBS and the External Works Index were both among the four most cited third-party domains in our data.

Then there’s the content itself. Specifiers want to know how one system compares with another, which applications it suits and where it shouldn’t be used, and few product pages say. Case studies are often a photograph and two lines of text. One that names the building type, height, performance requirement and the reason the product was chosen gives an AI tool something it can actually cite.

The compliance risk

Outdated product information in an AI answer is a risk for any manufacturer. In construction, the consequences can extend to safety and legal liability.

Since the Building Safety Act, product information carries far more weight, particularly on higher-risk buildings where the golden thread requires accurate records of what was specified and installed. If an AI tool cites a superseded datasheet, a withdrawn test certificate or a fire classification that applied to an earlier version of a product, a specifier may carry that error into their work.

Regulation shapes these answers directly. Government and regulatory sites took at least 8.5% of citations in our tracking, almost twice the share going to the businesses’ own websites.

The Construction Products Association’s Code for Construction Product Information (CCPI) sets out what clear, accurate and up-to-date product information should look like. Manufacturers verified against it already have the discipline AI search rewards. The same standard needs to extend to every place the product appears online, including old PDFs still sitting on the website, discontinued product pages and third-party listings.

“Or equivalent” just got easier

Getting specified has never guaranteed the order. Once a project reaches a contractor, a QS looking to protect margin will often look for a cheaper “similar or approved” alternative, and a manufacturer’s differentiators (safety, quality, durability) are hard to defend at that stage.

AI tools make that search much faster. A contractor can now ask for alternatives to a specified product and get a list of apparently equivalent options in seconds, with a summary of how they compare. If the specified manufacturer’s advantages aren’t documented in specific, verifiable terms, the AI tool will treat the alternatives as interchangeable. Competitors are already well represented: in our data, rival suppliers’ websites took about 10.5% of citations, more than twice the share of the businesses’ own sites.

Saying a product is “high quality” or “built to last” doesn’t survive that comparison. Test results, certified performance figures, design life, warranty terms and installation evidence do, because they give the AI tool concrete differences to report.

What manufacturers can do

Most of the fixes are straightforward.

  • Publish key performance figures, classifications, test standards and certificate references as text on each product page, with the PDF as a supporting download.
  • Keep datasheets and certificates ungated, and save lead capture for design support and sample requests.
  • Give one person responsibility for keeping the website, NBS, BIM objects and distributor listings aligned, and for taking superseded documents offline.
  • Write application guides and honest comparisons with alternative systems, including where a product shouldn’t be used.
  • Make every case study name the building type, the performance requirement, the specification decision and the outcome.
  • Get the technical team on record in articles, video and podcasts explaining how and why products should be specified, because that’s the expertise AI tools recognise and cite.

The starting point is the same as in any other sector: run the questions specifiers ask through the main AI platforms, record which manufacturers are named and which sources are cited, and fix the gaps that show up. That audit has to cover more than ChatGPT. In our tracking, Google’s AI platforms cited businesses’ own sites far more often (7.85% of citations on Gemini and 6.7% on AI Overviews) than ChatGPT did (1.24%).

About the data

The figures in this article come from Profound, an AI visibility platform. 

Related insights.