A second language is a maintenance commitment, not a launch

Adding a second language to a blog is usually budgeted as a launch: translate the archive, wire up the language switcher, announce it. That is the cheap part. The real commitment is that from launch day onward, every correction, retraction and change of position has to happen twice, and the language that gets left behind keeps publishing the old version with your full authority.

Translation is paid once. Divergence is paid every time something changes.

We argued earlier that machine translation has made the second language cheap enough to change the decision. That is still true, and it is also only half the ledger. Producing a Finnish version of a post is a one-off cost. Keeping it saying the same thing as the English version is a recurring one, and it recurs on a schedule nobody controls: whenever a price moves, a product is renamed, a claim turns out to be wrong, or the company quietly stops offering something.

A single-language archive already decays; we wrote about how every post has a half-life and why the cost sits in noticing rather than fixing. A second locale does not add a second, independent half-life. It adds a gap. The two versions start identical and drift apart one small edit at a time, and the drift is invisible from inside either language.

A stale locale is worse than a missing one

A Finnish reader cannot see the English page. That is the whole reason the Finnish page exists. So when the English version is corrected and the Finnish one is not, the Finnish reader gets the uncorrected claim with nothing on the page to warn them. The fact that you fixed it elsewhere does not reach the person it was fixed for.

Search engines make this sharper rather than softer. Google's own documentation on localized versions of a page says translated pages are only treated as duplicates if the main content stays untranslated. A properly translated post is its own page, indexed on its own, and the hreflang annotations exist precisely so that Google shows the Finnish searcher the Finnish version. The system is designed to send each reader to the copy in their language, which is exactly the copy nobody has looked at since March.

A missing Finnish post costs you a reader who never arrives. A stale one costs you a reader who arrives, believes you, and later finds out you had already changed your mind.

Divergence starts with the small edits

Nobody forgets to translate a retraction. A post pulled for being wrong gets pulled in every language, because the decision is big enough that someone walks it through. The drift comes from the edits too small to walk through: a price adjusted in one sentence, a product name updated, a broken link swapped, a hedge added after a customer complaint. Those get made in the language the person editing reads, and they are done the moment the save button is pressed.

We have this exposure ourselves, and it is worth being specific about it. On this blog the title, body and search fields are stored per locale, while the hero image is shared. When we update a post, the update names one locale and touches only that locale's fields. That is the correct design, because a Finnish post that has been rewritten natively cannot be patched by overwriting it with English. It also means the system will happily accept a fix to the English text and consider the job finished.

Make the change the unit, not the language

The repair is to stop treating an edit as something that happens to a page and start treating it as something that happens to the argument. A correction is recorded once, as a change to what the post claims, and the post is not considered updated until every locale carries it. The status that matters is not "edited" but "edited in two of two".

This is where an engine earns its keep and where it stops. Rewriting the Finnish paragraph to carry the corrected price is mechanical, and we would rather a machine did it the same day than a person got round to it next quarter. Google does not penalise a page for being machine-written; it has no patience for a page that is wrong. What the engine cannot decide is which edits are meant to travel. A corrected figure must reach both versions. A Finnish example that replaced an American one must not be overwritten by the English original on the next sync. That line is a judgment about intent, and it belongs to whoever approves the change, recorded once so the machine can apply it everywhere.

Count languages by what you can keep current

The usual question before adding a locale is whether you can afford to publish in it. The better question is whether you can afford to correct in it, at the speed of your fastest correction, for as long as the archive stays online.

If the answer is yes, a second language is one of the best-value decisions a small company can make. If the answer is "mostly", publish in fewer languages and keep them in step. Two locales that say the same thing are a stronger position than four that each describe a slightly different company, and a reader in any one of them can only ever judge you by the version they were sent.