The question almost always arrives as a price question: what does it cost to have the website in English as well? That answer is the smallest number in the whole exercise. A second language on your website is not a project with an end, it is a second operation: every change twice, every legal page twice, every line of annotation twice. And the most expensive mistake is rarely a mediocre translation. It is a second language that is not technically recognisable as one. This piece costs out both, with our own measurement across twenty Austrian home pages.
Why is translation the smallest item?
Because it happens once and everything after it happens every year. Let us cost it out on a deliberately small operation: fourteen fixed pages of around 500 words each, twelve blog posts a year at 1,200 words, and one annual revision of a fifth of every fixed page.
For the price we take no quotation but the statutory rate. Section 11(1) of the German JVEG sets the fee for translators appointed by a court: "Das Honorar für eine Übersetzung beträgt 1,95 Euro für jeweils angefangene 55 Anschläge des schriftlichen Textes, wenn der Text dem Übersetzer in editierbarer elektronischer Form zur Verfügung gestellt wird." That is 1.95 euros per started unit of 55 characters. How many characters sit in a word we counted in our own archive: the 59 German posts in this journal come to 68,311 words and 458,703 characters, so 6.71 characters per word.
That gives us this. The initial translation of the fourteen pages is 7,000 words, 855 standard lines, 1,667 euros. The running operation is 15,800 words a year, over five years 79,000 words, 9,646 standard lines, 18,810 euros. The initial translation is therefore 8 per cent of the five-year total.
Two honest caveats. The JVEG rate is what the state pays translators in court proceedings, not a market price for copy that has to sell, so the calculation is a floor. And it contains words and nothing else: the second navigation, the second set of forms, the second layer of search annotation and the coordination time are not in there. Neither omission improves the ratio.
What runs twice from then on?
Not the text. The text is the visible part, and it is the part you can buy from a supplier. What runs twice is everything around it: addresses, navigation entries, form fields with their error messages, the confirmation email after submission, page titles and descriptions for search, image alt text, sitemap entries, and the language annotation itself.
The effort is not in creating those, it is in the sequence: a single changed figure on a service page passes through five stations, and almost every operation forgets the last one.
That fifth station is why a second language often stands worse after two years than on the day it went live. Not because anyone was careless, but because on every change only the first three stations hurt. This is the classic shape of technical debt: a backlog nobody notices until it costs visibility.
Why does Google sometimes not see your second language at all?
Because Google does not read a page's language where most people write it down. The documentation puts it plainly: "Google uses the visible content of your page to determine its language. We don't use any code-level language information such as lang attributes, or the URL." Not the lang attribute, not the /en/ folder in the path, but the visible text.
What that means in practice turned up during the measurement for this piece. One of the twenty companies we checked serves a page under an English address that carries lang="en-US" while its title and content are entirely German. For a visitor that is a small disappointment. For Google it is a German page, and the second one at that.
Two further sentences from the same documentation dispose of popular workarounds. On redirecting by browser language: "Avoid automatically redirecting users from one language version of a site to a different language version of a site." And the technical reason a language switch without its own addresses goes nowhere: "the Googlebot crawler usually originates from the USA. In addition, the crawler sends HTTP requests without setting Accept-Language in the request header." A visitor in Vienna gets German because their browser asks for it. The crawler asks for nothing and therefore only ever sees one version. The rule out of that: one address per language, no redirect, a visible language switch as a link.
How often is the second language annotated wrongly?
More often than the effort involved would suggest. On 10 September 2026 we fetched the home pages of the twenty ATX companies, the largest listed businesses in Austria, and looked at what they declare in hreflang. All twenty serve a page with HTTP 200 under a second language address. Only thirteen tell Google that those addresses belong together.
The three properties checked are not niceties, they are the conditions under which the annotation works. The self-reference, because every version has to list itself. The return link, because Google writes: "If two pages don't both point to each other, the tags will be ignored." And the x-default, which names the version visitors land on when no language setting matches.
There is one more trap inside the language code itself. The same documentation lists incorrect region codes among the common mistakes: "using EU, UN, or UK in hreflang annotations doesn't have an effect on Google Search". The frequently typed en-uk therefore does nothing, en-gb is the correct form. In our sample not a single code was invalid. What was missing were the links.
Do you have to translate your legal pages?
Not according to the wording of the statutes. Section 5 of the Austrian ECG only says that a service provider has to make the listed information permanently, easily and directly accessible, and then lists everything from name to VAT number. No language is named anywhere, the word does not appear in the provision. The German counterpart, section 5(1) DDG, states the requirement with the same silence on language: information "die leicht erkennbar und unmittelbar erreichbar sein müssen, ständig verfügbar zu halten".
What is interesting is where language does turn up. Neither regime requires you to offer a particular language, both require you to say which ones exist: section 9(1)(4) ECG names "die Sprachen, in denen der Vertrag abgeschlossen werden kann", and article 246c(4) EGBGB matches it with "über die für den Vertragsschluss zur Verfügung stehenden Sprachen". Run an English website but conclude contracts only in German, and that is precisely what you have to write down.
The privacy notice sits below the level of statute. The transparency guidelines of the European supervisory authorities (WP260 rev.01, adopted on 11 April 2018) say in paragraph 13: "A translation in one or more other languages should be provided where the controller targets data subjects speaking those languages." As an indicator of that targeting the footnote names, explicitly, that someone "operates a website in the language in question". That is an expectation of the regulators, not a statutory rule. Anyone turning it into a duty is overstating it, and an English website is still exactly the indicator the guideline names.
The item that appears in no quotation is a different one. On 7 December 2010 the Court of Justice ruled in joined cases C-585/08 and C-144/09 (Pammer and Hotel Alpenhof) on when an activity is "directed to" another country, and held on language: where a website allows consumers to use a language or currency other than the one customarily used in the trader's own member state, that language or currency may be taken into account as an indicator. In the operative part language sits in the list of indicators, tied to the possibility of making and confirming a booking in that other language. A translation without a checkout therefore weighs less than an English shop. Cross that threshold and you invite the consumer law of the target market into your house.
Who reads the second language anyway?
That question usually gets answered with a feeling and rarely with a figure. Two surveys help, and both point in an uncomfortable direction.
The first concerns your own market. In Eurobarometer 540 (fieldwork September to October 2023, 26,523 respondents) 54 per cent of respondents in Austria say they speak English well enough to hold a conversation. In Germany it is 65 per cent, across the EU 47. More interesting than the level is the movement: in 2012 Eurobarometer 386 still put Austria at 73 per cent. The newer report claims in one place that "Compared to 2012, in all EU Member States, English grows substantially" and contradicts itself a few pages earlier, where it says of foreign language skills overall: "In six countries it has decreased: Austria (61%, -17)."
The second survey concerns the target market, and there the second language becomes the argument for it. In 2020 CSA Research surveyed 8,709 consumers in 29 countries with Kantar, filtered from 31,933 vetted participants: "76% of online shoppers prefer to buy products with information in their native language", and "40% will never buy from websites in other languages". For Germany the same analysis reports the highest value in the study, 57 per cent buying only at local-language websites. Honestly stated: the report sits behind a membership, what is publicly verifiable is the CSA release of 7 July 2020, and the 76 per cent applies, per CSA, to a choice between two similar products rather than to buying behaviour in general.
Together those two give you an order of operations. The second language is rarely there for your own audience, it is there for people who do not read your first one. You have to be able to name them: not "international clients", but this market, these enquiries last year, this offer.
When is the second language worth it anyway?
Once that answer exists, there is a smaller cut that is almost always better than the full copy: limit the second language to the pages that sell something. Home page, services, contact, legal texts. The blog stays in one language until demand is proven. That takes by far the largest part out of the calculation above, because those twelve posts a year are exactly the item that comes back every year. Localisation never meant translating everything anyway, it meant adapting the right things.
Three levers when the decision is due:
- Cost out the running item, not the quotation. Multiply the words you publish in a year by five. That number decides, not the initial translation.
- Give every language its own address and check the annotation at three points: does every version list itself, does the other version link back, is there an
x-default? In our sample four out of twenty passed all three. - Cut the scope before you translate. Selling pages first, blog later, and put the sentence about which language contracts are concluded in into the legal texts.
If you are weighing up an English version right now: we will cost out the running item with you before anyone translates a word. 🌍