Multilingual SEO that actually ranks in every market
Translation is the easy part. Hreflang, localized content, and an engine that scales to thousands of pages without reading like a machine are what make it rank. I have done it across 91,000 pages in 30 languages.
Most "multilingual" sites are just translated sites, and most translated sites do not rank. They miss the hreflang that maps each version to its market, they leave URLs and metadata in English, and they ship machine-literal copy that Google filters as thin. Multilingual SEO is the system that fixes all three: correct technical signals, native-reading content, and a translation pipeline you can run at scale.
What multilingual SEO actually requires
- Bidirectional hreflang, every locale references every other plus itself, with x-default. Get this wrong and the whole cluster gets filtered.
- Localized URLs, subdirectory or subdomain per locale, not a query param or a cookie. Search engines need a stable, crawlable URL per language.
- Native-reading content, machine translation alone reads as spam. It needs a humanization pass to compete.
- Locale-complete sitemaps and internal links, every version in the sitemap, breadcrumbs and nav pointing within the locale, not back to English.
- Per-market signals, currency, examples, and intent that match the audience, not a find-and-replace of the English page.
Translation vs localization vs multilingual SEO
Translation converts words. Localization adapts tone, examples, currency, and intent to a market. Multilingual SEO wraps both in the technical structure that lets search engines and AI assistants serve the right version to the right person. You need all three. A beautifully localized page with broken hreflang still loses; a technically perfect page in machine-literal Spanish still loses. The win is doing both, at scale, repeatably.
How I do it at scale
The engine is a pipeline, not a plugin. Machine translation (DeepL or an LLM like Claude) does the first pass. A glossary locks brand names and domain terms so they never drift. A humanization step rewrites the output until it reads as native, this is the part almost everyone skips, and it is why most translated content gets filtered. Hreflang, locale URLs, and sitemaps are generated at build time, so 30 languages stay consistent without anyone hand-maintaining tags. I have run exactly this across 91,000 pages in 30 languages on a live programmatic platform.
Engine vs plugin vs SaaS, in one line each
- Weglot, the safe SaaS pick for small-to-mid sites: easy, SEO-correct out of the box, recurring per-word cost that climbs with content.
- GTranslate, cheaper, but only the paid indexable plan does anything for SEO; the free widget is invisible to Google.
- WPML, fine for one WordPress site you translate by hand.
- Custom engine, wins at scale or when you need control over quality and cost. The build-vs-buy crossover is usually a few thousand pages.
The full breakdown: Weglot vs GTranslate vs a custom i18n engine.
What it costs
Three shapes. An audit of an existing multilingual estate (hreflang, locale URLs, sitemaps, content quality) with a prioritised fix plan: 4,000-10,000 USD. Implementation on a SaaS or plugin (Weglot, WPML) done SEO-correct, glossary and quality gates included: 8,000-30,000 USD depending on page count and language count. A custom translation engine, machine translation plus humanisation plus build-time hreflang, the setup I run across 91,000 pages in 30 languages: 30,000-120,000 USD to build, then API and infrastructure costs that undercut SaaS per-word pricing from roughly the third language onward. The first language is the expensive one; every language after it gets cheaper.
When multilingual is the wrong call
Four honest cases. Your English site does not rank yet: fix that first, because translation multiplies existing authority, it does not create it. You want one extra market: a single well-localised landing cluster beats a half-translated site and costs a tenth as much. Nobody on your team can spot-check the target language: pure machine output with no review loop is how sites get filtered, so line up a native reviewer or pick fewer languages. And your content changes daily with no pipeline: translations rot within weeks, so build the automation first or accept permanent lag.
Frequently asked questions
What is multilingual SEO?
Multilingual SEO is the practice of making one site rank in multiple languages and markets at once. It is more than translation: it needs correct hreflang, localized URLs, per-locale content that reads as written by a native, and a sitemap and internal-linking structure that tells search engines which version belongs to which audience. Translation is step one; multilingual SEO is the whole system.
Is translation the same as multilingual SEO?
No. A translated page with no hreflang, a machine-literal tone, and a URL Google cannot map to a locale will not rank, and may get filtered as duplicate or thin. Multilingual SEO adds the technical signals (hreflang, canonical, sitemap, locale URLs) and the quality bar (humanized, native-reading copy) that make translated content actually compete.
How do you translate at scale without it reading like a machine?
A pipeline: machine translation (DeepL or an LLM like Claude) for the first pass, a glossary to lock brand and domain terms, and a humanization step that rewrites the output so it passes as native. I have run this across 91,000 pages in 30 languages. The humanizer is the part most teams skip, and it is the difference between ranking and getting filtered.
Should I use Weglot, GTranslate, WPML, or a custom engine?
For a small-to-mid site that needs to ship fast and SEO-correct, Weglot is the safe call. GTranslate is cheaper but only if you pay for the indexable plan; the free widget does nothing for SEO. WPML suits a single WordPress site. A custom i18n engine wins once you are at real scale (thousands of pages) or need control over translation quality and cost. See my full build-vs-buy comparison.
What are the most common multilingual SEO mistakes?
Broken or one-directional hreflang (every version must reference every other, including itself, plus x-default), translating the body but leaving the URL and metadata in English, client-side-only translation that search engines never index, missing locale versions in the sitemap, and machine-literal copy that reads as spam. I have cleaned up sites with 30,000 of these at once.
How much does multilingual SEO cost?
Three bands. An audit of an existing multilingual estate: 4,000-10,000 USD. SEO-correct implementation on Weglot or WPML: 8,000-30,000 USD. A custom translation engine like the one I run across 91,000 pages in 30 languages: 30,000-120,000 USD to build, with per-page costs after that far below SaaS per-word pricing. The crossover from SaaS to custom is usually a few thousand pages or three-plus languages.
If you are going multilingual, the first call is short
Tell me your content estate, your target markets, and your stack. I will tell you whether a SaaS, a plugin, or a custom engine fits, what the hreflang architecture needs to look like, and what it costs to get every market ranking. I build and run this at scale through Seahawk Media.