3,182 local businesses, 561 broken or missing websites: how to tell if yours is one of them

Search “signs your website is obsolete” or “when to redesign your website” in French and you'll get twelve pages written by twelve web agencies that sell redesigns. Every one of them tells you your site is finished, and every one of them backs it with the same unsourced numbers: 53% of visitors leave after three seconds, you should redesign every three to five years, conversion goes from 0.5% to 2%. I'm not going to repeat those figures, because none of the pages carrying them says where they came from — and that missing source is the actual point.

I sell redesign work too. Same chair. The only thing I can offer that those twelve pages don't is a count: I ran three automated checks against 3,182 local businesses, I'm naming the ones that failed, and I'm publishing the method including the four ways it produces wrong answers, so you can re-run it on your own city and disagree with me.

Two things to state before anything else, because they change how you should read the rest.

This is a first-party market observation, not a client case study. None of these 3,182 businesses is a client of mine. None was manually audited. None was contacted for this article. What follows is what a script saw from the outside.

The geography is wrong for you, and I'm not going to bury that. The 16 cities are all in the United States and Canada. Zero France, zero Belgium — while most people reading this run an agency or a local service business in one of those two countries. This is a North-American observation whose transposability is genuinely up for discussion. That's why the second half of the article moves to two Belgian agency sites I audited by hand, and why those two turn out to be the most useful part.

The three rules, and nothing else

A business was flagged if it failed at least one of these:

1. No website at all — the Google Maps listing has no website field.
2. Website unreachable — the request failed three times in a row.
3. Frozen copyright — the page text carries a copyright year several years in the past.

That's the entire test. It says nothing about design, load time, mobile layout or conversion rate. It detects neglect, not quality. Keep that distinction in mind; it's the hinge of this whole article.

What 3,182 records actually look like

Sixteen cities. Canada (6): Toronto, Vancouver, Calgary, Ottawa, Edmonton, Winnipeg. USA (10): New York, Miami, Los Angeles, Chicago, Houston, Philadelphia, San Francisco, Boston, Seattle, Atlanta.

561 businesses qualified — 17.6% of the 3,182. That total is not an estimate: the eight city batches sum to 561, and the deduplicated final file (deduplicated on the Google Maps URL) also contains 561 records. The two numbers match exactly, so there are 561 distinct businesses and no cross-batch duplicates.

The per-vertical figures are accurate to roughly ±6 — I located the block boundaries in a sorted file rather than counting every line — while the 561 total is exact.

One mandatory warning on that table: restaurants were only queried in 6 of the 16 cities, against 16 cities for the three other verticals. The restaurant number is therefore not comparableto the others, and nobody — including me — gets to read it as “restaurants are doing better.”

Two censuses, side by side, and why an average would lie

I did not read all 561 records. Each line is expensive to read because of the Google Maps URL column, and the full file runs past 120,000 tokens. What I did instead was read two batches in full — every single line — and refuse to extrapolate from them.

Those 102 records are 18.2% of the 561. The other 459 I did not read line by line.

Now look at what the two rows do to each other. Batch 1 is majority alive but frozen — 34 of 58, about 59%, are working sites whose copyright stopped years ago. Seattle and Atlanta flip it: 23 of 44, about 52%, have no website at all. Same three rules, same script, opposite diagnosis.

That's why I won't publish a sentence like “X% of local businesses have no website.” Pick either batch and you'd get a completely different X, and both would be honest arithmetic on a sample that doesn't support the claim. Geography doesn't change how many businesses fail — it changes which failure they have.

The pattern that actually carries this: the reputation is alive, the website isn't

The interesting finding isn't the count. It's the gap between an excellent Google reputation and an abandoned web presence. These are all named records I read myself:

Twenty-two thousand reviews at 4.7, and the request to the website fails. A real estate team with a 4.9 and a copyright that stopped in 2013.

The oldest copyright year I read is © 2003 — Universal Tours Inc, Seattle, 4.0 / 39 reviews. Not “the oldest in the dataset”: 459 records went unread, and something older may well be sitting in them.

And one record that stops this from becoming a tidy story: M. Fishman & Co., Chicago — 2.1 / 119 reviews — no website. Not every qualified lead is a thriving business whose only problem is its website. Some of them are just struggling, and the missing site is a symptom, not the cause.

Four ways this test is wrong

This section is the reason to trust the rest of it. The script has four documented failure modes, and I found all four by reading it rather than by running it.

1. The copyright check keeps the maximum year found in the page text. An abandoned site that mentions a recent date anywhere slips through — a false negative. A perfectly maintained site whose legal page still says “© 2019” gets flagged — a false positive.
2. The script reads the raw HTML the server returns. Footers rendered in JavaScript — React, Vue, Wix, any single-page app — are simply invisible to it. No copyright found, no flag raised. Silent false negatives, and probably the largest category.
3. HTTP 403 anti-bot responses count as reachable but copyright-undetectable.And “site unreachable” only ever meant “the request failed three times from my environment.” The script itself carries a comment warning that above a concurrency of 3, the DNS resolver saturates and returns false “unreachable” results. The question mark in that label is not decorative.
4. “No website” comes from a missing field on Google Maps. The business may well have a site it simply never listed there.

Two Belgian agencies the test would have missed completely

Here's where the automated method meets the readership it's supposed to serve — and loses.

Etimo, Charleroi(etimo.be) has been in business since 1963. Its social proof is better than most of its competitors': it's a member of Opinion System, an ISO 20252-certified review body, which is a stronger signal than a Google star average. On the site itself: the H1 of the homepage is literally the word “Accueil”, the parser found zero CTAs on the entire page, five different phone numbers appear in the markup with several of them truncated, and the certified review block sits at the bottom of the page, below the new listings. There is no copyright year in the footer at all. Rule 3 has nothing to read. Etimo would never appear in my 561.

urbano-immo.be is the mirror image. Its copyright says 2025 — perfectly fresh, passes rule 3 without blinking, site loads fine, website field present on Maps. Passes all three rules. And yet: no H1 anywhere on the page, an empty meta description, an empty OG title, and a property search that displays listings marked “VENDU” — sold. A buyer landing there sees nothing available.

Put those two next to the 561 and the limits stop being abstract. A recent copyright year is not proof of a working website, and a missing one is not proof of an abandoned business. The three rules find neglect that leaves a visible trace. They do not find a site that loads, looks current, and quietly fails to convert anyone — which, from where I sit auditing these things by hand, is the far more common and far more expensive condition.

Honest note: I have no conversion data, no traffic figures and no revenue impact for any business in this article — not the 561, not the two Belgian agencies. Nobody gave me their analytics. What I have is a reproducible external observation on 3,182 records, two hand-read censuses totalling 102 of them, two manual audits, and a published list of everything the method gets wrong. Anyone telling you that a frozen copyright year costs you a specific percentage of leads is making it up, and so would I be.

Run the three checks on your own site, then the two that matter more

1. Open your homepage source and search for the copyright symbol. If there's no year, you have Etimo's problem: a visitor gets no signal that anyone still maintains the site. If there's a year older than last year, fix it today — it costs one line and it's the single cheapest trust signal on the page.
2. Check whether that footer is rendered in JavaScript. View source, not inspector. If the year appears in the inspector but not in the raw source, some automated readers never see it at all.
3. Open your own Google Maps listing and confirm the website field is filled in. Thirty-nine of the 102 records I read in full had none. Some of those businesses almost certainly have a site; they just never put it there.
4. Load your site from a network that isn't your office— mobile data, a VPN, anything. “Works on my machine” is exactly the failure mode rule 2 was designed to catch, and exactly the one it also gets wrong.
5. Then read your H1 out loud.If it's the word “Home,” or your company name, or nothing at all, you've just found a bigger problem than any copyright year — and it won't show up on any automated test, including mine.
6. Count the CTAs above the fold.Zero is a real number, and I've now found it on a live agency site that's been trading since 1963.

Hamza Benjaaba

Web designer and no-code developer, specialized in websites for real estate agencies and local service businesses. The 3,182-record observation in this article comes from my own lead-qualification tooling, run on public Google Maps data in July 2026; the Etimo and Urbano findings come from manual audits in my own audit pipeline. Neither company is a client, and neither was contacted for this piece.