Qaurus is a Canadian go-to-market agency in the Toronto, Waterloo corridor. In September 2026 we measured 39 pages across 12 Ontario marketing agency websites, our own included, against the on-page signals that AI answer engines draw citations from.
The market has solved the technical layer almost completely and skipped the evidence layer almost completely. We score 0% on the same measure everyone else does.
What did we measure, and why those signals?
Eleven on-page signals that determine whether a page can be quoted rather than just read. An answer engine assembling a reply needs a sentence it can lift and a source it can attribute. The signals split into three groups:
- Machine-readable structure, JSON-LD schema depth,
@idgraph connectivity - Answer shape, question headings, a direct answer beneath them, an FAQ block, a clear entity definition in the opening lines
- Evidence, named tools or platforms, first-hand operator voice, a quantity with a stated source
Every page was fetched once from the public site, parsed, and scored by the same code we run against our own pages.
What does the Ontario agency market actually look like?
Structurally healthy, evidentially empty. Across 35 competitor pages:
| Signal | Pages carrying it |
|---|---|
| Schema depth (typed JSON-LD) | 29/35, 83% |
| Answer-shaped headings | 22/35, 63% |
| Direct answer under a question | 21/35, 60% |
| Named tool or platform | 21/35, 60% |
| Media with proper alt text | 18/35, 51% |
| Clear entity definition | 15/35, 43% |
| Contextual internal links | 15/35, 43% |
| FAQ block | 6/35, 17% |
| First-hand operator voice | 3/35, 9% |
| A quantity with a stated source | 2/35, 6% |
@id graph connectivity | 0/35, 0% |
Not one site was missing JSON-LD entirely. Every agency has done the technical work. Two pages out of thirty-five carry a number a reader could trace to a source.
Which signal is the market missing most?
Sourced data, and it is not close. Only 2 of 11 agencies had a single page carrying a quantity attributed to a named measurement source. Nine had none at all.
This matters more than it looks. An answer engine choosing between eleven agency pages that all describe the same services in the same structure has nothing to distinguish them. A page that says "we tracked 83 URLs over six weeks in Search Console" gives it something to quote and someone to credit. A page that says "we deliver results" does not.
We are in the empty column too. Zero of the four Qaurus pages in this sample carry sourced data. We measured the market and found our own gap in it, which is why the two studies published alongside this one are the first attempts to close it.
Is anyone doing anything unusual?
One thing, and it is ours. Of 35 competitor pages, none connects its schema nodes with
@id references, the mechanism that tells a search engine that the organisation on the services
page and the organisation in the article byline are the same entity rather than two coincidentally
similar names.
All four Qaurus pages do. We are not claiming this wins anything on its own; entity graphs are a foundation, not a ranking factor. But in a sample of 39 pages it is the only structural difference that separates any site from the rest, and it took one afternoon to implement.
How long are agency pages?
Median 896 words, with a long tail at both ends. The shortest page measured 155 words; the longest 2,597. Seven of 35 pages fell under 300 words, the floor below which our own indexing gate refuses to queue a page for submission.
Length is not quality. But a 155-word service page has roughly one paragraph in which to define an entity, answer a question and cite something, and in practice it does none of the three.
What should a business owner take from this?
If you are choosing between Ontario agencies, their websites will not help you much, they are near-identical in structure and almost uniformly silent on evidence. Three questions will separate them faster than any amount of browsing:
- Show me a number and where it was measured. 94% of the pages we scanned could not.
- What did it cost per outcome? Cost per lead, with the date range.
- What happened that did not work? The answer is more informative than the case study.
FAQ
Which agencies were included? Eleven Ontario marketing agencies, labelled Agency A through K, plus Qaurus. The agencies are anonymised deliberately: the finding is about a market pattern, and naming firms in a small regional market turns a measurement into an accusation. Qaurus is the one site identified, because we are the only participant who agreed to be. Three further sites returned HTTP 403 or refused the connection to our research user agent; they were excluded rather than retried, and are not among the eleven.
Does a low score mean an agency is bad at marketing? No, and that is worth stating plainly. This measures what a website publishes, not what an agency delivers. An agency with excellent client outcomes and a thin website will score poorly here. What the score predicts is whether an AI answer engine has anything to quote, a narrower claim.
Why does @id connectivity matter if it is not a ranking factor?
Because entity resolution is how a search engine decides that two mentions refer to the same
organisation. It is a prerequisite for being treated as a known entity rather than a string, and it
is invisible to readers, which is probably why nobody implements it.
Could the sample be skewed? Yes, in two ways we can name. The page selection came from an earlier audit rather than a random crawl, so it favours homepages and primary service pages. And three sites blocked our user agent, which may correlate with more sophisticated infrastructure, meaning the sites most likely to score well are slightly more likely to be missing.
Method and limitations
Pages were fetched once each in September 2026 from the public web with a research user agent,
then scored by index_readiness.py, the same module that gates our own pages before they enter an
indexing queue. No Google SERP was queried, no rank data was scraped, and no login or paywall was
crossed. The full signal definitions are in that file.
39 pages across 12 agencies is a description of a small market sample, not a census. Ontario has far more than 12 marketing agencies. These are the ones already in our competitive tracking, which skews toward Kitchener, Waterloo and toward firms serving small and mid-sized businesses.
Signals are proxies. "First-hand operator voice" measures phrasing, not expertise; a page can satisfy it without the underlying work having happened. "Sourced data" checks that a quantity has a stated origin, not that the origin is accurate. Both would be trivial to game, which is a reason to treat a high score as necessary rather than sufficient.
One page per URL, one moment in time. Sites change, so these figures describe September 2026 and nothing after it.
Anonymisation costs verifiability, and that is a real trade. Because the agencies are labelled rather than named, you cannot independently reproduce our per-agency figures, you have our word for them. We judged that a fair price for not publishing a league table of named local businesses. The per-page scores sit in our repository under the same labels, and the scoring module is the one we run against our own pages, so the method is inspectable even where the sample is not.
We publish what we measure, including the parts where we come out badly. If you want this run against your own site, get in touch.
