First edition · Ontario, Canada
The AI Visibility Index.
By Josh Redecop · Published
Across 37 Ontario building-product manufacturers scanned between July and September 2026, not one scored above 74 out of 100 for AI visibility. The median was 55 and the range ran from 27 to 74. No company in the set has established itself as the answer an AI assistant reaches for in this category — which means the position is unoccupied rather than defended. The ceiling held when the sample was widened by a third, which is the result that makes it worth reporting.
- Companies scanned
- 37
- Median score
- 55
- Highest score
- 74
- Lowest score
- 27
Scanned 2026-07-28 to 2026-09-13. Every figure on this page is our own measurement of our own dataset. No company is named — all 37 are real businesses, and a published ranking of companies we may later approach would turn a measurement into a pressure tactic.
What the scans found
There is no leader
The highest score in the set was 74 out of 100, and only seven companies of 37 reached the Competitive band at all. In most categories a clear leader has already emerged in traditional search — but not here, and not yet. Nobody has been displaced from the top of this category, because nobody occupies it.
The weakest link is the one they control most
Site readability averaged under half its weight, and it is the only dimension a manufacturer can fix entirely alone, without anyone else publishing anything. The specification data almost always exists — it is in the brochure, the installation guide, the warranty document — published in a form an engine cannot parse.
Identity is not the problem here
Entity consistency scored 75.1% of its weight, comfortably the strongest dimension. These are long-established businesses with stable names and addresses. That is worth stating because it inverts the usual advice: for this category, work on entity signals has the least headroom, and work on readable content has the most.
Being named regionally is not being named
The most common pattern in the set is a company named first on a narrow query about its own town or its own branded product line, and entirely absent from the broad category question. Thirteen of 37 sat in exactly that position. The narrow query is reassuring and carries little volume; the broad one decides the shortlist.
Metal roofing is furthest behind
At a median of 35, metal roofing and steel scored lowest of the five categories — while having arguably the most numeric buying question, where pitch, span, gauge and finish narrow the answer to a specific product. The category where an assistant could be most decisive is the one giving it least to work with.
Some are blocking the engines outright
A separate sweep of 65 sites in the same category found five whose robots.txt actively disallows AI crawlers — ChatGPT, Claude and Gemini among them. One of those five is currently named in AI answers anyway, and publishes an llms.txt, while telling three of the four engines not to read the site. That is a configuration accident rather than a strategy, and it is the cheapest thing on this list to fix.
Reviews are the structural gap
Review signal was the weakest dimension overall, with 21 of 37 under half. This is largely structural rather than negligent: a manufacturer selling through a dealer network rarely accumulates reviews in its own name, because the buyer reviews the dealer. It is the hardest dimension to move and the one least worth panicking about.
How the 37 scored
| Band | Score range | Companies |
|---|---|---|
| Competitive | 60–74 | 7 of 37 |
| Developing | 41–59 | 21 of 37 |
| Emerging | 27–40 | 9 of 37 |
Whether an assistant names them at all
AI answer presence is the heaviest of the six dimensions, worth 30 of the 100 points. Across 37 companies:
- Reliably namedNamed across most tested queries19
- Named sometimesSurfaced on narrow or regional queries only13
- Effectively absentNot named on any tested query5
Which dimensions fail
Each dimension carries a different weight, so the figures below are expressed as a percentage of what was available — making them comparable to each other. Strongest first.
Entity consistency (10 pts)
75.1%The strongest dimension by a wide margin. These are long-established companies with consistent names, addresses and phone numbers across the web. Identity is not what is holding them back.
Cited-source presence (20 pts)
59.6%Most appear in trade directories and association listings. The corroboration exists; it is often thin, but it is rarely absent.
Freshness (5 pts)
60.5%Mixed, and frequently self-contradicting — a current blog post sitting above a footer copyright several years stale.
AI answer presence (30 pts)
49.1%The heaviest dimension, and barely half earned. Companies are commonly named for a narrow regional query and absent from the broad category question that carries the volume.
Site readability (15 pts)
48.1%The dimension they control most completely, done worst. Specification data published as PDFs and images of tables, with marketing copy where the numbers should be.
Review signal (20 pts)
42%The weakest dimension in the set. Twenty-one of 37 scored under half. Manufacturers selling through dealers rarely accumulate reviews under their own name.
By category
Directional rather than definitive — some of these groups are small, and the limits below say which.
Windows, doors, siding & cladding
median 64· n=6 · range 60–69Trusses & engineered wood
median 55· n=9 · range 36–72Concrete, precast & masonry
median 53· n=11 · range 37–67Lumber & building supply
median 50· n=6 · range 27–74Metal roofing & steel
median 35· n=5 · range 29–61What this does not show
Thirty-seven companies is a real sample for this category but a small one in absolute terms, and the category groupings are smaller still — the metal roofing median rests on five companies. Treat category comparisons as directional. The first edition of this Index reported 28; widening it to 37 moved the median by half a point and left the ceiling and the dimension ranking unchanged, which is some evidence the shape is stable.
Each company was scored once, at a single point in time between 28 July and 13 September 2026. AI answers change. A company scanned today might score differently.
Some individual dimension values in the underlying scorecards are marked as estimates, principally review ratings read from aggregators rather than verified on each platform directly. Those are flagged in the source records and are carried here unchanged.
Scoring involves judgement. The six dimensions and their weights are fixed and applied consistently, but reading whether a company was "named" in an answer is not a purely mechanical call.
The set is Ontario-only and skews toward independent manufacturers rather than subsidiaries of national groups. It is not a census of the province.
About the Index
What was actually measured?
Each company was scored out of 100 across six weighted dimensions: whether AI answers name it, whether independent sources corroborate it, its review signal, how readable its site and product data are to a machine, whether its business details are consistent across the web, and how current its content is. The same instrument was applied to every company in the set.
Why are no companies named?
Because publishing a named ranking of companies we may later approach would turn a measurement into a pressure tactic, and that is not a trade we are willing to make. The aggregate findings are the useful part: they describe a category, and any manufacturer can locate themselves within the distribution without being listed in it.
How were the companies chosen?
They were drawn from a working list of Ontario building-product manufacturers and suppliers built from trade-association member directories and regional sources, filtered to companies that manufacture or supply building products and sell to contractors, builders or dealers. Companies were scanned in the order they entered the list, not selected by expected result.
Can I find out where my own company would score?
Yes. The same six-dimension scan runs against any manufacturer, and the result is a written scorecard with the evidence attached rather than a number on its own — including which competitors get named when you are not. That is the audit described elsewhere on this site.
Will this be updated?
The intention is a second edition once the scanned set is substantially larger, which also makes the category comparisons more reliable than this first edition can claim. Any future edition will state its own sample size and dates rather than quietly absorbing them into this one.
If you make building products, the mechanics behind these scores are set out in GEO for building-product manufacturers, and the four reasons an assistant stays silent are in why AI doesn't recommend your products. If you sell software instead, see GEO for SaaS companies.
Thirty minutes. We run your category live on the call.
No pitch deck. We ask the engines what your buyers ask, in front of you, and show you which competitors get named. You keep the findings either way.