Why this exists
We wanted Canadian numbers, so we published our own.
There is a great deal of published advice about being visible in AI-generated answers, and very little published measurement, particularly for Canada.
Between June and August 2026 we ran commissioned discoverability audits for businesses in Alberta, British Columbia and Ontario. Each audit scores the client and every competitor a buyer-query search surfaces, on the identical rubric, with the same evidence standard. That produced a dataset as a by-product: 75 businesses, eight sectors, one scoring method.
This page publishes it. Businesses are anonymized; the numbers, method and limits are not.
The headline
Ranking first on Google does not mean being named by AI
In four of the eight scored sets (exactly half), the business with the strongest classical SEO was not the business AI engines named most often. Two different businesses were winning two different games in the same market.
The split appeared in counselling, artificial turf, accounting, and wealth management. In the accounting set, the firm with the best classical SEO foundation was beaten in AI answers by a competitor with weaker on-page fundamentals but a stronger presence on the sources engines actually cite. Whichever game a business is winning, it may not be able to see the one it is losing.
This is the practical reason a search-rankings report is no longer a discoverability report. They measure different things, and half the time they produce different winners.
The blind spot
The machine-readable layer is effectively unclaimed
Every business we score gets a machine readability score out of 20: an llms.txt file, an AI manifest, Schema.org structured data, and whether core facts (hours, address, phone, principals, services) sit in structured data an agent can parse rather than in HTML a human has to read. One caveat: llms.txt is still a proposal, and Google Search does not use it.
The median business scores 25% of the available points on this pillar: the weakest of the four by a wide margin, in every sector we scored. Classical SEO, by comparison, has a median of 68%.
The reason this matters more than it sounds: the other three pillars are contests. Competing on SEO or reviews means outworking businesses who have been at it for years. The machine-readable layer is not a contest yet. It is a pillar where the median competitor scores a quarter of the available points, and in our experience the work is measured in days.
The distribution
The median business scores 48 out of 100
Across all 75, the median Discoverability Index score is 48, with an interquartile range of 38 to 62 and a full range of 18 to 85. 53% score below 50. Narrowing to the 51 Canadian independent local businesses, the median falls to 44.
| Sector | Calibration | n | Median SDI | Median machine readability |
|---|---|---|---|---|
| Lawn care / home services | retail | 9 | 65.0 | 40% |
| Sports league software* | advisory | 8 | 62.0 | 32.5% |
| Counselling / mental health | retail | 9 | 58.0 | 35% |
| Accounting / advisory | advisory | 15 | 49.0 | 20% |
| Group benefits / insurance | retail | 9 | 47.5 | 17.5% |
| Artificial turf / landscaping | retail | 9 | 43.5 | 32.5% |
| Nonprofit / housing consulting | advisory | 7 | 39.0 | 30% |
| Wealth / family office | advisory | 9 | 37.0 | 20% |
Sector medians rest on 7 to 15 businesses each. Treat them as directional, not authoritative. Note also that the two lowest-scoring sectors are both professional services, where the advisory calibration weights AI answer visibility highest.
Scale
Being big helps with authority, not with agents
Fifteen of the 75 are national networks, multinationals or US platforms, including Big Four accounting firms and enterprise software companies. Their median score is 60, against 44 for Canadian independent local businesses.
But the advantage is not evenly spread. They lead comfortably on authority signals (median 87% of available points versus 53%) and on classical SEO (80% versus 60%). On the machine-readable layer they score a median of 25%, the same as everyone else.
A national brand's advantage is built from decades of citations, press and domain authority, and a local business cannot buy its way past that quickly. The machine-readable pillar has no such moat. It is the one pillar where a nine-person firm and a Big Four firm start from the same place.
Honest limits
What this data cannot tell you
Published research is worth what its caveats are worth. Ours are real, and they cut against our own numbers.
Seven limits, stated plainly
- This is not a random sample. Each set is one business that commissioned an audit plus the competitors a buyer-query search surfaced. Competitors are in the sample because they were visible, which biases it upward. The true median for Canadian small businesses is very likely lower than 48.
- Audit subjects are self-selected and score lower. Their median is 33, against 49 for the competitors around them. Businesses often commission an audit when they already suspect a problem. We do not present the subject median as a market figure.
- n = 75 across eight sectors is small. Sector figures rest on 7 to 15 businesses. This is a first cut, not a national census.
- Not every business here is small. Fifteen are national or enterprise. Every claim about local businesses on this page uses the 51-business Canadian independent cut.
- One set is not Canadian. The sports league software sector is mostly US-based. It is in the overall figures and out of every Canadian figure, marked in the table.
- Scores are point-in-time, June to August 2026. The engine set expanded from three to five partway through; earlier runs were revalidated against that change, but two carry three-engine evidence.
- Client confidentiality. Every report is confidential to the client who commissioned it. Nothing here names a business, and nothing here reproduces a client's own score.
We would rather publish a small honest number than a large invented one. If the sample grows enough to change these figures, this page will say so.
Method
How each business was scored
One 0–100 score, four pillars, applied identically to every business in a set so they sit on one leaderboard.
- Pillar 1Classical SEO: 30 pts retail / 25 advisoryIndexability and named-crawler access, on-page completeness, content depth against the buyer-query set, ranking presence, and an authority proxy. On-page and schema verified by live browser render, never raw fetch. A JavaScript-rendered site returns an empty shell to a non-browser fetcher, and scoring that shell produces a wrong answer.
- Pillar 2AEO / GEO answer visibility: 35 pts retail / 40 advisoryShare of voice across Perplexity, ChatGPT, Gemini, Claude and Copilot on a fixed, geo-locked buyer-query set, with a new session per query to avoid context contamination. Plus citation-source health, review or thought-leadership cadence, and inclusion in the listicles and registries engines pull from.
- Pillar 3machine readability: 20 pts, both calibrations
llms.txt, a published AI manifest (an MCP description file; there is no agreed standard location for one yet), Schema.org completeness for the business class, and whether core facts are agent-parseable. - Pillar 4Trust: 15 ptsLocal trust signals for consumer-facing businesses (Google Business Profile, category-relevant Canadian review platforms, local recognition), or authority and credibility for professional and B2B firms (published thought leadership, named proof, third-party recognition).
Every factual claim in an audit traces to a logged source or carries an explicit estimated, modeled or unverified tag. The same discipline governs this page: every figure here recomputes from the underlying scored rows, and each row is checked against its four pillars before it enters the dataset.
FAQ
Questions about the benchmark
How discoverable are Canadian businesses in AI search?
Across 75 businesses scored on one rubric between June and August 2026, the median Discoverability Index score was 48 out of 100, and 53% scored below 50. For the 51 Canadian independent local businesses in the set the median was 44. These figures come from audit sets where competitors were selected because a buyer-query search surfaced them, which biases the sample toward more visible businesses, so the true market median is likely lower.
Does ranking first on Google mean AI will recommend you?
No. In four of the eight scored sets (exactly half), the business with the strongest classical SEO was a different business from the one AI engines named most often. Ranking and being cited in a synthesized answer are related but separate outcomes, driven by different signals.
What is the weakest discoverability signal for Canadian businesses?
The machine-readable layer, by a wide margin. 63 of 75 businesses (84%) scored below half on machine readability, which covers llms.txt, an AI manifest, Schema.org structured data, and core facts an agent can parse. Of the 25 sites we probed, none published an MCP description file, and there is no agreed standard location for one yet. Of 33 businesses whose components we logged individually, six publish an llms.txt.
Are big national brands more discoverable in AI search than local businesses?
Partly. The 15 national, enterprise and multinational businesses in the set had a median score of 60 versus 44 for Canadian independent local businesses, but their advantage sits in authority signals and classical SEO. On the machine-readable layer they are just as weak, with the same 25% median. That pillar is open to any business willing to do the work.
How is the Discoverability Index scored?
One 0–100 score from four pillars: classical SEO, AEO/GEO answer visibility, machine readability, and a trust pillar. Local businesses are scored on a retail calibration (30/35/20/15) and professional or B2B firms on an advisory calibration (25/40/20/15). AEO/GEO carries the heaviest weight in both. Every business in a set is scored identically so the results sit on one leaderboard.