Content localization

Content localization that starts with what to translate

Most content localization starts by translating the most popular articles. This workflow starts one step earlier: it works out which articles have search demand in which language, checks that nothing already exists or competes, and only then translates, checks and publishes them as drafts in the right language.

It was built for Reply.io's blog, in Claude Code with Python checks and a WordPress publisher for multilingual sites.

Why translating your top articles wastes most of the budget

An article that performs in English can have almost no search demand in German, and a modest one can be huge in Brazilian Portuguese. Translation is priced per word, so choosing by English traffic spends most of the budget on pages nobody will search for.

The numbers from the plan make the point. Of 105 possible translation jobs, the top 38 carried 91% of the estimated visits for 39% of the words. The rest of the list was mostly cost.

A content localization strategy built on search data

The selection runs as a pipeline, and each step removes a reason not to translate:

1. Take the Search Console export for about 2,900 URLs, keep the English pages, and crawl the top 600 by clicks, which carry 99% of them.
2. Classify each page, merge redirects and duplicates, and drop anything already translated or not worth translating, like company news or dated seasonal posts.
3. Rank the remaining articles on clicks and impressions, with a boost for pages already ranking in the top 10 or 20.
4. Write three candidate keywords per article per language: the native phrasing, the literal translation, and a different angle.
5. Check volumes and difficulty across ten countries, estimate winnable traffic, and sort every job into tiers.
6. Check again that each target isn't already live, and that it wouldn't compete with an existing page in that language.

The last step can block a job outright. It stopped four, where the translated keyword would have competed with the client's own AI assistant landing pages in those languages.

Two things the data showed that intuition didn't

The literal translation is often the wrong keyword. The second or third variant won 64 of the 105 jobs. In German, the literal phrase for finding a phone number lost to a phrasing for reverse phone lookup that gets 39,000 searches a month.

Country choice changes the answer too. Checking Spain alone made Spanish look seven times smaller than French. Adding four Latin American countries took the Spanish volume from 2,990 to 11,980 searches, and one sales term drew 80 searches in Spain against 2,800 in Mexico.

Translation that has to pass a checker

A translated article isn't finished until the checking script passes it, and the script is strict:

The checker has its own tests: 27 deliberately broken articles, each breaking one rule. A second script flags things to re-read without failing the batch, such as a translation much shorter than its source or a keyword repeated in back-to-back sentences, which the first checker once let through.

Translation isn't word for word either. Examples are localized to the market, and sections are added where local searchers ask something the English article never covered.

Publishing into a multilingual WordPress site

Finished articles go into WordPress as drafts in the right language, with a person pressing publish. The English post's images are reused rather than uploaded again, with translated alt text, and English e-book covers are dropped, which cut one batch's page weight from 6.7 MB to 2.2 MB.

One step stays manual on purpose, because there's no API for it. Linking each translation to its English original in the multilingual plugin has to be done in the admin screen, so a verification script lists every pair and fails until all of them are linked.

Where it stands

The translation batches so far were chosen from keyword sheets the client supplied for Brazil. The first 10 Brazilian Portuguese articles went live in September 2026, and the second batch of 36 passed every check and is being drafted into WordPress in groups.

The prioritization plan covers what comes next. None of its 38 recommended jobs has been commissioned yet, and German, Spanish and French are waiting on keyword lists.

Companies ready to localize

This suits a company with a proven English blog and real customers in other languages, where the question is where to spend a translation budget. It fits multilingual SEO best when the site already runs a multilingual setup, so translations can live next to their originals.

A site with little English search traffic is the wrong place to start, since the ranking relies on pages already earning clicks. And it doesn't replace native review for anything legal or regulated.

Where the localization budget goes

Most of the cost is translation and review time, and the selection exists to shrink it. The pipeline itself uses a Search Console export and keyword exports you likely already have.

Localization is billed through execution capacity, a flat $2,000 monthly for agents and automations on top of a fractional Head of Content retainer. The first plan for a new site takes two weeks or more, most of it keyword checks across markets.

Start with the languages you sell in

Tell us which countries your customers are in and send a Search Console page export. On a call we'll look at how much of your English blog has demand in those languages before a single word is translated.