The short version
- 15 of 50 answers recommended the company, all under its parent company's name.
- 0 of 50 used the name of the new commercial brand, and no answer cited the new website.
- 14 of those 15 came from siding questions. On a flat-roof question it was recommended 0 times in 10.
- 9 of 9 pages we checked on the new site carried a setting that keeps a page out of Google.
- Google AI Overviews named the company most often: 7 of 10 answers.
The question
The contractor works for developers, general contractors and property owners. It had split its commercial work from its residential work and given the commercial side its own brand and website. We ran the study for its marketing adviser, to see whether AI tools had picked up the new brand.
What we asked
Five buying decisions, agreed in advance and worded the way a buyer would type them. None named the company.
- An out-of-town developer choosing a commercial siding contractor for a multifamily project.
- The manager of an occupied apartment community replacing siding without disrupting residents.
- A general contractor choosing a siding and exterior partner for a mixed-use project.
- The manager of a light-industrial building with a leaking flat roof, deciding between repair and replacement.
- A portfolio manager deciding where to spend on roofs and siding across several properties.
Each question went to ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews, and then again about a day later. That gave 50 recorded answers.
What the answers said
| Buying decision | Answers that recommended the company |
|---|---|
| 1. Developer, commercial siding | 6 of 10 |
| 2. Occupied apartments, siding replacement | 4 of 10 |
| 3. General contractor, mixed-use exterior | 4 of 10 |
| 4. Light-industrial flat roof | 0 of 10 |
| 5. Portfolio roof and siding assessment | 1 of 10 |
| All five | 15 of 50 |
| AI tool | Answers that recommended the company |
|---|---|
| Google AI Overviews | 7 of 10 |
| ChatGPT | 6 of 10 |
| Gemini | 2 of 10 |
| Claude | 0 of 10 |
| Perplexity | 0 of 10 |
Finding 1: AI knew the parent, not the brand
All 15 recommendations used the parent company's name or a short form of it. None used the full name of the commercial brand. No answer cited the new commercial website as a source.
The parent's reputation was doing the work. That's worth keeping. But a buyer who followed the answer would have landed on the parent's site, not on the commercial pages built for them.
Finding 2: known for siding, missing for roofing
Fourteen of the 15 recommendations came from the three siding questions. On the flat-roof question the company wasn't recommended once, although it offers that service and has a relevant project to show.
The roof specialists that did appear had pages explaining how they assess a roof, how they decide between repair and replacement, and how they work around a building that stays open. AI had clear material to describe them with.
Finding 3: one setting was hiding the new site
We checked nine pages on the commercial site, including the home page, the service pages and two project pages. All nine carried a noindex tag.
That tag tells Google to keep a page out of its search results. Google says a page marked noindex is dropped from Search, and that a page has to be indexed to appear as a supporting link in its AI features.
So the first job wasn't new content. It was asking whoever maintains the site to check that setting.
Finding 4: the answers moved, the pattern held
We asked every question twice. Whether the company appeared was the same in both passes for 24 of the 25 question and tool pairs. The order of the shortlist changed far more often, and two tools showed almost entirely different firms the second time on one question.
That's the reason to count across many answers instead of trusting one.
What we recommended
- Fix access first. Remove the unintended noindex setting and confirm in Google Search Console that the pages can be indexed.
- Connect the brand to the parent. Keep the separate commercial site, and explain the relationship clearly on both sites so AI can link the two names.
- Put proof beside each service. Link the siding page to its multifamily project and the flat-roof page to its roof project, and link each project back.
- Make the assessment tangible. Show an anonymised extract of a real assessment report.
- Then ask the same five questions again.
We also advised against a few things: a weekly blog before access was fixed, dozens of near-identical location pages, and any schema or AI file sold as a guaranteed shortcut.
What this study doesn't show
- Fifty answers are repeated observations of five questions. They aren't a sample of every buyer in the market.
- The page check happened after the answers were recorded. It shows the setting was there that day, not how long it had been there.
- The study doesn't prove why any tool left the company out. Fixing the setting is a first step, not a promise of a result.
More on how we count and what we don't claim in the method. For the wider picture in this industry, see AI visibility for commercial contractors.