We scored fifty nine local business websites, and the failures were not the ones we expected

We built a website test that scores a page out of 100 on nine fixed checks, then pointed it at a sample of ordinary Quebec local business websites to find out what the real distribution looks like. Sixty one attempted, fifty nine scored. The median came out at 76, which is higher than we expected, and the two checks the most sites failed are the two that decide whether a stranger stays. Nobody is named here, and nobody is going to be.

How this was run

All of it is reproducible, which is the only reason publishing a distribution is defensible. The checks are fixed rules with no model involved anywhere, so the same page scores the same every time, and the script that produced these numbers is in the repository this site is built from.

  1. The sample. Sixty one Quebec local business homepages across eight trades: plumbing, electrical, roofing, auto repair, barbershops and salons, dental clinics, pilates and yoga studios, landscaping and snow removal, and veterinary clinics.
  2. How they were found. Ordinary web searches, the same way a customer would find them. Directories, marketplaces and franchise landing pages were removed, so every site in the sample belongs to a single independent business.
  3. What was fetched. One page per site, the homepage, once. Never a crawl. robots.txt was checked first and honoured, and requests were made one at a time with an identifiable user agent.
  4. What was scored. The nine checks behind the website test, weighted to total 100. Two sites are excluded from the results: one returned a server error, one disallowed us in robots.txt.
  5. What is published. Scores and pass rates. No hostnames, no screenshots, no identifying detail. These are working businesses that did not ask to be graded in public.

The distribution

The headline is that most of these sites are fine. Half scored 76 or better and seven scored a perfect 100, which is not the picture usually painted by people who sell websites.

The interesting part is not the middle, it is the tail. Seven sites out of fifty nine scored below 50, and one scored 22. Those businesses are not losing customers because of design taste. They are losing them because a stranger cannot tell what they do or how to contact them.

  • Median 76, mean 72. The mean sits below the median because a small number of very poor sites drag it down, which is exactly the shape you would expect.
  • Range 22 to 100. The gap between the worst and the best is not a design gap. Every check the worst site failed is fixable in an afternoon.
  • Seven perfect scores. None of them were expensive-looking sites. Passing these nine checks has nothing to do with budget.
Where the fifty nine scored sites landed
0 to 24 1 of 59 cannot be found or understood at all
25 to 49 6 of 59 failing several fundamentals
50 to 74 20 of 59 workable, leaking in one or two places
75 to 100 32 of 59 clears the floor

Source: Automatik, nine-check website test run across 61 Quebec local business homepages on 14 August 2026; 59 scored, 2 excluded (one server error, one robots.txt disallow). Aggregate data in scripts/sample-scores.json.

Which checks people actually fail

This is the part worth the exercise. The distribution tells you how businesses are doing overall; the per-check pass rates tell you what to fix, and the order is not what most people would guess.

Pass rate by check, weakest first, across 59 scored sites
CheckPassedWhat failing it means
Someone other than you vouches for you51 percentNo reviews, testimonials or case studies anywhere on the homepage
The first screen says what you do54 percentThe main heading is a name or a slogan rather than a service and a place
The page is not heavy before it starts58 percentA heavy document, or a long list of scripts and stylesheets before anything renders
There is an obvious way to book or ask76 percentNo booking, quote or enquiry link that a rule can find
The phone number is tappable on a phone76 percentNo tel: link, so the number has to be memorised and typed into the dialler
A real business is visible behind the site80 percentFewer than three of address, phone, hours, email and a social profile
Nothing on the page looks abandoned83 percentA stale year or leftover placeholder copy still on the page
The phone number is text, not a picture85 percentNo phone number in the page text, so it is almost always inside an image
Getting in touch takes no hunting92 percentNo contact route near the top of the page

The three findings that are worth acting on

Everything below comes straight out of the table above. None of it is a hunch.

  • Half of these businesses publish no proof at all. The single most-failed check, at 51 percent, is whether anyone other than the business vouches for it. These are trades and clinics with years of satisfied customers and, on the page a stranger actually lands on, nothing but their own claims. This is the cheapest fix on the list and the one most consistently skipped.
  • Nearly half fail the first sentence. Only 54 percent have a heading that tells a stranger what the business does. The rest lead with the business name, a slogan, or a full-screen image with the useful information somewhere below it. Everything else on the page is downstream of this, which is why it is the most expensive failure in the set.
  • Page weight is worse than reputation suggests. At 58 percent it is the third most-failed check, and it is the one nobody in these businesses can see, because their own site is cached in their own browser. A customer on a phone on mobile data is having a different experience from the owner, every time.

What surprised us

Two things, and we are reporting them because they cut against what an agency would prefer to find.

The floor is higher than the industry pretends. A median of 76 with a third of the sample above 85 does not support the usual claim that local business websites are universally terrible. Most of these are adequate, and the businesses running them do not need to be sold a rebuild.

Modern sites failed differently, not less. The oldest-looking sites in the sample often had the phone number in text, in the header, next to the opening hours, because that was simply how sites were built. Several newer, better-designed sites failed contactability while looking considerably more professional. Design competence and conversion competence turned out to be close to unrelated in this sample.

What we would not claim from this

The limitations matter more than the findings, so they get their own section rather than a footnote.

  • It is not a random sample. Sixty one sites found through ordinary searches, across eight trades, in one province. It is a snapshot, not a census, and it is biased toward businesses that are already findable.
  • It is one page per site. A business whose homepage hides the phone number may well have it on a contact page. The homepage is where search traffic lands, which is why it is the page tested, but the score is about that page rather than the whole site.
  • It reads the document, not the rendered result. A site that paints its content in with JavaScript scores worse here than it looks in a browser. We think that is the more useful measurement, since it is closer to what a crawler and a slow connection get first, but it is a choice and it should be stated.
  • Nine rules are not a judgement of quality. These are hygiene checks. A site can pass all nine and still fail to persuade anybody, and a site can score 60 and be doing perfectly well because its customers all arrive by referral.
  • We built the test. The rules encode what we think matters. They are all listed on the tool page so you can disagree with any of them, and the weights are visible rather than implied.

Run it on your own site

The whole point of a deterministic test is that you get the same treatment these fifty nine got. Paste your URL into the website test and you get the nine checks, pass or fail, each with a one-line fix. Nothing is stored and there is no email gate.

If it fails the two big ones, proof and the first heading, the fixes are writing rather than development. The four second test covers what a visitor decides before scrolling and in what order, and the voice pass covers how to write a heading that says something only your business could say.

And if the site turns out to be fine, the leak is further down. What happens after somebody gets in touch is a different problem with its own arithmetic, in the speed-to-lead guide.

FAQ

Why not name the sites?
Because the story is the distribution, not any individual business. Publishing a list of local companies with low scores would be a cheap way to get attention at the expense of people who never agreed to take part, and it would make the finding less useful rather than more.
Can I reproduce these numbers?
Yes. The checks are in the repository, the scoring script takes a list of URLs, and the aggregate output from this run is committed alongside it. Point it at your own sample and you will get comparable output, because there is no model in the path to introduce variance.
Why 61 attempted and 59 scored?
One site returned a server error while we were testing, and one disallowed our user agent in robots.txt. Both are excluded rather than counted as failures, because neither is a finding about the page.
Is a score of 76 good?
It is the median here, which makes it ordinary rather than good. The more useful reading is per check: a 76 achieved by failing proof and offer clarity is a worse position than a 76 achieved by failing page weight and a stale year, even though the number is identical.
Does this mean most local businesses do not need a new website?
That is roughly what the data says, yes. Most of this sample needs a heading rewritten, some reviews added and a few scripts removed. Those are afternoons, not projects, and a rebuild would not fix any of them on its own.

Try the tool