Original research · published 19 November 2026 We scored 18 real websites on whether an AI can read them
Every agency selling AI search visibility says it works. Almost none of them publishes a measurement. These are 18 live client sites, scored out of 100 on three checks, anonymised to sector and score. The median was 55. Every figure below is generated from the audit file, not typed in.
- 18 live sites
- Method stated in full
- No sign-up to run it yourself
Most sites are not unreadable. They are unquotable.
Not one site in this set was broken. They loaded, they ranked, people used them. They simply did not state the few facts an assistant needs in a form it could lift.
Median score across the 18 sites. The mean was 57, and the range ran from 40 to 76.
Had no single extractable sentence saying what the business is and where — 15 of 18, the commonest fault.
Had no linked profiles corroborating the business elsewhere. Every site in the set failed this check.
Scored 70 or better. 6 of 18 came in under 50, which is where an assistant starts guessing.
Three checks, and how often each one failed
An assistant answering a question about a business has to establish three things. Each site was scored on whether it made them establishable.
One plain sentence saying what the business is
A sentence an assistant can lift that states what the business does and where. Not a tagline, not a heading, not words inside an image — extractable text. This was the most common failure of the three, and it is the cheapest to fix.
Failed on 15 of 18 sites (83%)
Organization or LocalBusiness markup
Structured data telling a machine the business name, address and contact details as facts rather than as page furniture. Without it an assistant has to infer the business from prose, and frequently infers wrong.
Failed on 12 of 18 sites (67%)
Linked profiles confirming the business exists elsewhere
A sameAs list pointing at the business’s own profiles — Google Business Profile, LinkedIn, an industry directory. It is how a model corroborates that a website belongs to a real organisation. Every single site failed this one.
Failed on 18 of 18 sites (100%)
All 18 sites, by sector and score
The complete dataset, lowest to highest. Sector and score is all that was recorded: there is no client column to withhold, because one was never kept against these numbers.
| Sector | Score | Out of 100 |
|---|---|---|
| Medical tourism | 40 | |
| Clinic | 44 | |
| Photography studio | 44 | |
| Restaurant | 46 | |
| Clinic | 47 | |
| Reviews platform | 49 | |
| Diagnostic centre | 51 | |
| Clinic | 51 | |
| Acupuncture clinic | 54 | |
| Leadership coaching | 55 | |
| Café | 60 | |
| Mobile tyre fitting | 63 | |
| Dietitian | 67 | |
| Immigration consultant | 67 | |
| Retail shop | 67 | |
| Clinic | 68 | |
| Manufacturer | 69 | |
| Law firm | 76 |
Median 55. Mean 57. Range 40–76. Audited 2026, published 19 November 2026 by MSJ Info Support Corporation, Surrey, British Columbia.
How the scoring worked
Each site was fetched the way an assistant fetches a page, then scored on three checks: whether one sentence of extractable text states what the business is and where it operates; whether the page carries Organization or LocalBusiness structured data; and whether it links to profiles corroborating the business elsewhere. The composite is expressed out of 100.
What the sample is. 18 live websites from MSJ Info Support’s own client list, across healthcare, hospitality, legal, retail, manufacturing and professional services, in several countries. It is not a random sample of Canadian businesses and nothing here should be read as describing that population. The figures describe this set, and the set is published in full above so a reader can judge it.
What was withheld, and why. The businesses are real paying clients. The dataset records sector and score only; no client is named against a score here or anywhere else MSJ Info Support publishes this audit. That is a deliberate limit on the research, and a reader weighing the numbers is entitled to know it exists.
What would make it stronger. A larger sample, sites not drawn from one agency’s clients, and a second scoring pass by someone with no interest in the result. The intention is to re-run the audit quarterly and publish the direction of travel, which is more informative than any single snapshot.
You can run the same scoring on your own site with the free AI visibility checker, which needs no sign-up and reports the specific faults it finds. The checks themselves are set out in the AI search checklist, and the work of fixing them is our AI SEO service.
What did the 18-site AI visibility audit measure?
MSJ Info Support audited 18 live client websites in 2026 for how legible they are to AI assistants, scoring each out of 100. The three checks were whether the page states in one extractable sentence what the business is, whether it carries Organization or LocalBusiness structured data, and whether it links to profiles confirming the business exists elsewhere. Scores ranged from 40 to 76, with a median of 55.
What was the most common failure?
15 of the 18 sites audited had no single plain sentence saying what the business is and where it operates — 83 per cent of the set, and the most common fault of the three measured. The information was usually present, but set in an image, a slogan or a heading rather than in text an assistant could extract and quote.
Is this a representative sample of Canadian small business?
No, and it should not be read as one. These are 18 websites from one agency’s client list across several countries and sectors, not a random sample of Canadian businesses. The figures describe this set. They are published because measured figures with a stated method are more useful than the unmeasured claims common in this market, not because they generalise.
Why are the businesses not named?
Because they are real paying clients and a named business paired with a low score would be a disclosure made at their expense rather than with their consent. The dataset is anonymised to sector and score, and MSJ Info Support applies the same rule everywhere the audit appears, including the figure that runs on the homepage.
Which sector scored best?
Law firm scored highest at 76 out of 100. Professional-services sites tend to state plainly who the practitioner is, what they do and where they are licensed, which is close to exactly what an assistant needs to extract — though in most cases it appears to be a by-product of how such sites are written rather than a deliberate choice.
How can a business check its own score?
MSJ Info Support publishes the same checks as a free tool at /ai-visibility-checker.html. It needs no sign-up, reads the page the way an assistant would, and returns the score with the specific faults found. The scoring is the one used in this study, so a result is directly comparable with the figures here.
How often is this study updated?
This edition of the MSJ Info Support AI visibility study covers 18 sites and was published on 19 November 2026. The firm intends to re-run the audit quarterly and republish, including the direction of travel. Every figure on this page is generated from the audit file at build time rather than typed in, so the page and the data cannot drift apart.
Where would your site sit on this table?
Run the same three checks on your own page. It is free, there is no sign-up, and the score is directly comparable with the 18 above.

