Greek Hospitality Digital Census 2026

I put 5,565 Greek accommodation properties through a protocol of three searches each, to answer one question only: does a website exist that belongs to the property itself. Of those, 4,690 produced a definitive verdict. The remaining 875 stayed unidentified, usually because the name is common and the map gives no settlement, and they are excluded from every percentage rather than counted as negative.

Across the 4,690 definitive verdicts, 1,108 have no website of their own, that is 23.6%. This blended figure hides the finding, because it mixes hotels with studios. Splitting by property type shows that the gap widens as the property gets smaller.

Apartments and studios are the largest category after hotels and carry the highest share of absence, 40.6% across 960 definitive verdicts. Taking all small properties together, the share is 38.1% across 1,839.

The census produced 3,667 website addresses. I downloaded the homepage of each one with the same script and the same settings, within the same week. 3,152 responded and they form the population of every figure below. The other 515 did not respond at all.

The most serious finding is not aesthetic. It is that the websites do not tell machines what they are. Structured data of any type is present on 41% of the 3,152. Structured data that explicitly declares a lodging business is present on 8%. On the rest, an answer engine trying to recommend a place to stay has no way to work out that the page is an accommodation, where it is, and how to book it.

So that the study does not stop at isolated measurements, I combined twelve checks into a single score. Each check is binary, measured mechanically on the HTML of the homepage, and none of them is a matter of taste. Either a tap to call phone number is there, or it is not.

Across the 3,152 websites the median is 7 out of 12. Eight websites, that is 0.3%, passed all twelve points. Another 94 reached 11. The two ends say the same thing from opposite sides: the level is not set by a few sites getting everything wrong, but by almost nobody getting everything right.

Median 7 out of 12 across 3,152 websites. Perfect score: 8 websites, 0.3%. Eleven points or better: 102 websites.

The reference curve used by the public audit tool comes from an earlier measurement of 2,533 Greek accommodation websites with the same script. There the median is 8 out of 12 and 40 websites hold a perfect score, that is 1.6%. The difference against the 3,152 of the census is expected: the census includes far more small island properties, which is exactly the part of the market that measures worse.

Reference population: 2,533 websites, median 8 out of 12, 40 perfect scores, 1.6%.

Every region carries its own figure and can be read on its own. The two tables measure different things and are not comparable with each other: the first measures how many properties have a website, the second measures what the existing websites do.

Properties with no website of their own, by region

Technical website audit, by region

Regions with fewer than 100 measurements are flagged as a small sample. At those sizes a single property moves the percentage noticeably and the figure should not be read as precise.

If you cover a specific region, the figure for your region may be republished freely, with attribution to the study and a link to this page. No permission is needed and nothing has to be requested.

The study was carried out by one person, using open data and scripts I wrote. Every step is described below so that it can be repeated or challenged.

Census figures in one line: 5,565 properties examined, 4,690 definitive verdicts, 3,582 with a website of their own, 1,108 without, 875 unidentified and excluded. Recheck of negatives: 946 rechecked, 92 overturned, 71 ambiguous, 25 domain with no content. Website audit: 3,667 addresses fetched, 3,152 responded.

Overturn rate on rechecked negatives: 9.7%.

The technical audit behind this study is public. The MyStaySite audit tool applies the same twelve points, with the same script and the same scoring logic described above, and places the result on the same reference curve. So any property owner and any journalist can verify the study against a real website instead of taking it on trust.

One difference, also recorded above: in the tool, the booking path check also credits an internal link to a booking or contact page. In the study only a form or a booking engine on the homepage was counted.

The study measures specific things and nothing beyond them. Below is what it does not measure, so that it is not read as something it is not.

• It does not measure design. No measurement concerns aesthetics, photography or how attractive a website is. A beautifully designed website can score low, and the reverse holds too.
• It does not measure content quality. The audit counts words, not whether the text is correct, useful or true.
• It does not measure business success. A property with no website can be full all summer through platforms. The study says nothing about occupancy, revenue or profitability, because it did not measure them.
• The population is as OpenStreetMap records it, not an official register. The Greek accommodation register exists and is more complete, but it is not released as open data. So properties absent from the map were not included, and the type of each property is the type the map assigns to it.
• The technical audit is a snapshot of one week and of the homepage only. A website that happened to be down those days was counted as not responding. Elements present on inner pages but not on the homepage were not measured.
• The audit ran without executing JavaScript. A website that builds its content entirely in the browser was measured on what the server returned, which is also what a machine that does not execute JavaScript sees.
• The census verdicts rest on search. The second, adversarial pass overturned 9.7% of the rechecked negatives, so a small residue of error remains after the correction, in the direction of overstating absence.
• No statistical significance testing was carried out between regions. Regional differences are presented as descriptive figures for this population.

The findings and tables on this page may be republished freely, in whole or in part, in print and online, with attribution to the study and a live link to this page. No permission is needed, no prior notice is needed and there is no time limit. The same applies to the charts.

The census is intended to be repeated with the same method, so that the 2027 figures are comparable with these.