
If you make flat roofing systems, you probably assume that when a specifier asks an AI tool for options, your brand has a fair chance of appearing. We wanted to test that assumption. So we asked ChatGPT, Claude and Gemini the same 40 questions a UK specifier might ask about a flat roof system, and compared which brands each one named.
The short version: the tools often disagree. Which shortlist a specifier sees depends on which tool they open.
Between 20 and 23 October 2026, we put 40 UK specifier questions to ChatGPT, Claude and Gemini. We asked each question three times per tool, with web search on. That gave us 360 answers.
The questions covered the ground specifiers actually cover: build-ups, fire performance, warranties, insulation, waterproofing membranes and certification. We then logged which brands were named, which sources were cited, and what advice each tool gave.
The full results are in our AI Visibility Index for flat roofing systems. Treat the figures here as approximate. This is a snapshot of a few days, not a permanent ranking, and these tools change often.
On average, only about a third of the brands named for a question were named by all three tools. Two-thirds of the names appeared in some answers and not others.
On 14 of the 40 questions, the tools agreed on fewer than 30% of the brands they named. For those questions, the three tools were close to building three separate shortlists.
Think about what that means in practice. A specifier who uses one tool may see your brand in the first answer. A colleague using another tool may never see it. Neither knows the other's list was different.
This is not about one tool being better. It is about there being no single "AI answer" to optimise for. Being named by one tool tells you very little about the other two.
The disagreement goes beyond brand names. Each tool has its own habits, and those habits shape what a specifier reads.
Advice to check the certificate. Claude told users to check the certificate in 79 of its 120 answers. ChatGPT did so in 6 and Gemini in 2. A specifier using Claude is repeatedly nudged towards third-party certification. One using Gemini rarely is.
Number of sources. Gemini cited about 19 sources per answer. Claude cited about 9 and ChatGPT about 6. More sources means more places your information can be picked up from. It also means more places where it can be inconsistent.
What gets cited. Manufacturer websites made up 49% of all citations. NBS Source was cited in 151 of the 360 answers.
That last pair of figures is the most useful for a marketing team. Almost half of what the tools relied on was manufacturer websites. Your own pages are doing a lot of the work, whether or not you designed them with AI tools in mind.
It is not all variation. On rules and standards, the answers were consistent and correct.
Take a question such as whether changing the insulation in a build-up affects a BROOF(t4) classification. The tools gave consistent answers, and they were right. Regulatory and standards content appears to be well covered and well understood.
So the variation sits elsewhere. It is in the open-ended question of who to consider, and which products to put on a shortlist. That is the part of the process where a manufacturer wants to be present, and the part where the tools are least aligned.
You cannot control which tool a specifier uses, or how it weighs one source against another. You cannot assume you are on the shortlist because you appeared in one test.
What you can control is the quality and consistency of the information these tools are drawing from. Based on what we saw being cited, these are the places to start.
None of this is new advice for a good specification-led website. What has changed is that more of your audience may now meet your information first through a summary written by a tool, built from your pages and others'.
Pick five questions a specifier would ask about your flat roof system. Ask them in all three tools. Note which brands come up, which sources are cited, and whether the details about your products match your own documentation.
You may find gaps you did not know about. You may also find your information is being used well. Either way, you will have a clearer picture than a single search would give you.
If you would like help making your product pages and specification content clearer, get in touch. We work with manufacturers across the construction supply chain, and we are happy to talk through what you are seeing.