In September 2026, a crawler drew 41,841 companies at random from the Swiss commercial register, looked for their websites, and audited the first 10,000 it could tie to their company with certainty. Here is what it found, in proportions, without a single name.
companies drawn at random from the commercial register: limited companies, corporations, sole proprietorships, general partnerships
websites tied to their company with proof, then audited
cantons, each in proportion to its weight in the register
margin of error on national percentages
The crawler does not use Google. It builds the addresses a customer would try from the company name, and keeps a website only if it finds proof there that it belongs to the company: the registered name, the UID number, the address or the town.
| Companies | Proven site | Uncertain | No address |
|---|---|---|---|
| All companies | 24% | 32% | 44% |
| By legal form | |||
| Limited company (Sàrl, GmbH) | 29% | 36% | 35% |
| Corporation (SA, AG) | 28% | 34% | 38% |
| Sole proprietorship | 13% | 24% | 63% |
| General partnership | 13% | 23% | 64% |
| By region | |||
| German-speaking Switzerland | 26% | 31% | 43% |
| French-speaking Switzerland | 20% | 33% | 47% |
| Ticino | 20% | 35% | 45% |
“Proven site” is a floor, not an adoption rate. What it measures is narrower and more useful: how many companies have a website that a machine can tie to them without getting it wrong. Among sole proprietorships, one in eight.
Share of the 9,970 reachable websites that have each element on their home page. A file that returns a web page instead of its content counts as missing.
Mobile version: the page declares a screen width (viewport tag). Structured data: a JSON-LD block describes the business for machines.
of websites give no structured data at all. Not the trade, not the address, not the opening hours, in a form a search engine or an AI can read without guessing.
have no privacy page the crawler could find. In French-speaking Switzerland, 54%. Since September 2023, Swiss data protection law requires a website to inform visitors as soon as it collects personal data, through a contact form for instance.
have an llms.txt file, almost always without asking for one: 96% of Wix websites have one, 87% of Shopify websites. The platform puts it there, not the business. On WordPress, 8%.
still shows a year of 2024 or earlier in its footer. For a visitor, it is the first sign of an abandoned website, even when it is not.
A gap of more than three points between German- and French-speaking Switzerland is beyond the margin of error. In Ticino, with 440 websites, the margin is about five points.
| Measure | German-speaking | French-speaking | Ticino | Switzerland |
|---|---|---|---|---|
| Websites audited | 7,267 | 2,263 | 440 | 9,970 |
| HTTPS | 94% | 95% | 95% | 94% |
| Mobile version | 91% | 93% | 89% | 91% |
| robots.txt file | 79% | 87% | 77% | 81% |
| Sitemap | 73% | 77% | 68% | 74% |
| Meta description | 62% | 64% | 54% | 62% |
| Structured data | 45% | 50% | 45% | 46% |
| llms.txt file | 17% | 19% | 14% | 17% |
| Privacy page | 59% | 46% | 43% | 55% |
| Year of 2024 or earlier in the footer | 11% | 11% | 16% | 11% |
French-speaking websites describe themselves to machines slightly more often (structured data, robots.txt) and show a privacy page far less often: 46%, against 59% in German-speaking Switzerland.
One eighth of the commercial register, open data from the Federal Commercial Registry Office, chosen by a hash of each company’s identifier: uniform and reproducible. 82,059 limited companies, corporations, sole proprietorships and general partnerships remain. Excluded: bankruptcies, liquidations, holding companies, listed companies, pharmacies, group subsidiaries.
In shuffled order, the crawler looks for each company’s website and stops at the 10,000th proven site. 41,841 companies processed.
Home page in desktop and mobile versions, contact and privacy pages, robots.txt, sitemap.xml, llms.txt. One page at a time, about twenty at most per site. Four websites refused the crawler through robots.txt: they are left out.
Proportions only: no names, no website addresses, no figure for a group of fewer than thirty websites. The raw data is deleted six months after publication at the latest.
The crawler identifies itself as HallebardierEtude, reads robots.txt before any other page and follows it. It opens nothing behind a login, submits no form and does not test security. Details are on the privacy page.
Every figure, by region, legal form and canton: etude-pme-2026.json (labels in French). The same data as CSV: GitHub, Hugging Face and Zenodo (DOI 10.5281/zenodo.23192819). To cite the study: Hallebardier, “10,000 Swiss small-business websites”, September 2026, hallebardier.com/en/swiss-sme-website-study/. Licence CC BY 4.0: free to reuse, including commercially, with the source and a link to this page.
The page code as the server sends it, without running JavaScript, like many AI crawlers.
Structured data placed elsewhere on the site is not counted. The home page is where a crawler starts.
A meta description can be there and poor, a privacy page there and incomplete.
A “legal notice” or “Impressum” page that includes privacy information without naming it is not counted as a privacy page.
Nothing here says whether a website is attractive, clear or convincing.
The crawler times the delivery of the code, not the display of images and scripts.
No certificates, no security headers: the study does not test websites.
WordPress, Wix or a custom build: the study counts platforms, it does not judge them.
The free test reads your home page the way the study’s crawler read these 10,000 websites, and tells you in a few seconds what ChatGPT, Claude and Google can get from it.