MACHINE 404 / The index

We scanned 42 Greek business websites. 96.6% have no price a machine can read.

Hotels, retail, clinics, law firms and food brands. Names you know, with budgets. The average score was 51.5 out of 100, the range ran from 21 to 70, and 86.2% scored under 65. A further 11 refused an automated reader outright and could not be scored at all, and 2 answered nothing at all.

By sector, worst first

What fails, and how often

How the sample was chosen, and what that means

This is a convenience sample of well-known Greek brands across five sectors, not a random draw from the business register. That matters, and it matters in one direction: these are companies with marketing departments, agencies and budgets. If anything, the real average across Greek business websites is worse than what you are about to read, not better.

Each site was measured with the same eighteen checks we sell, on 2026-09-13, from one address, three requests each, sequentially with a pause. The rubric is published in full with the weight of every check, so any number here can be re-derived by anybody who wants to argue with it.

The refusals, and what we are not claiming

11 of the 40 sites would not answer an automated reader at all: a challenge page, a firewall, or a silence. We are not claiming those sites block ChatGPT. Verified AI crawlers are often allowlisted where a generic reader is not, and we did not impersonate one to find out, because impersonating a crawler to get past a wall is exactly the behaviour that gets scanners banned.

What it does show is a posture. A site that answers browsers and refuses everything else has made a decision about machine readers, usually without knowing it, and that decision now sits between them and every assistant that is not on their firewall's allowlist.

What we take from it

The failures cluster. It is almost never the whole site that is unreadable; it is the same two or three things missing everywhere, and they are the expensive ones. The pattern across this sample is that pages are described well and companies are not described at all, which is why an assistant can summarise what a business does and still not name it as the supplier.

The second thing is that spending does not fix it. Several of the lowest scores here belong to sites that clearly cost a great deal to build. Nobody bought a bad website; they bought a website for an audience that has since changed, and no dashboard told them.

The three holes, in order

96.6% publish no machine-readable offer. Not a cheap price or an expensive one: no price at all, as far as the thing comparing three suppliers is concerned. It is the heaviest check in the scan at 12 points, and the shops fail it as often as everybody else. One site in the sample does publish a price a machine can read, and it finished at the top of the sample.

48.3% do not declare the company at all, so a model can read what they do and still not know who they are. And 96.6% have never made a decision about AI crawlers, which means the decision was made for them by silence.

Limits, stated plainly

One moment in time. A convenience sample, not a random one. Five sectors, not the economy. Homepages, not whole sites. And measured by the company that sells the fix, which is why the rubric, the weights and the method are published rather than described. Re-run it yourself on any domain, including ours.

If one of these is yours

No business is named here and none will be. If you want to know whether you were in the sample, ask: you get the same thing the free form on the product page gives anybody, your score out of 100, your band and the three checks costing you the most, at no charge and whether or not you ever buy anything. Publishing a named low score to sell somebody a fix is a way to get attention once and deserve none of it afterwards.

The other fifteen checks with their findings, and the instructions written for your own platform, are the report, and that is the part that is paid for. Same line as everywhere else on this site: how bad it is, is free.