Volver al blog
Guía12 min de lectura

Machine Translation for Website Localization: What to Check

Machine translation covers the wording of a localized site, not the whole job. See what Google allows, what to set up for SEO and how costs scale.

Equipo Lantis

Machine translation can produce readable website text quickly, but it covers only the linguistic side of localization, so style, adaptation and brand tone still need review. For SEO, the bigger risks are structural: one URL per language, hreflang on every relevant page, and sitemaps that list the versions. Google does not ban machine-translated pages, but large volumes of low-value translated pages can fall under its scaled content abuse policy.

  • Machine translation handles wording. Menus, product names, legal text, calls to action and brand copy still need a human read.
  • Google does not treat machine translation as spam by itself, but a large set of thin, unreviewed translated pages can break its scaled content abuse policy.
  • Give each language its own URL, add hreflang to every applicable page, and use XML sitemaps for hreflang if your platform cannot output it in the page code.
  • Compare cost by active languages and word allowance, not by the headline price.

What can machine translation do for a website, and where does it fall short?

Machine translation handles the wording of a page quickly, but it covers only part of a localization job. Smartling's guide to website translation notes that machine translation, and post-editing to a lesser extent, mainly addresses the linguistic side of translation rather than the whole localization work.

Illustration about “What can machine translation do for a website, and where does it fall short?”

Quality varies. The same Smartling guide says translation quality may vary and that clients sometimes need to step in editorially.

A small academic study examined Google Translate output for the menus of one flight-charter website, from English to Indonesian. The output split into readable and less readable text. The weaker part came from the tool's inability to adjust to stylistic forms and to apply adaptation strategies. It is one small case, so it illustrates a risk rather than giving a general error rate.

Vendors are more optimistic. Motionpoint, a localization company, claims that neural machine translation is often of good quality and has fewer errors than other machine translation methods. That is the vendor's own claim, and the practical way to test it is to run your own pages through the tool and have a native speaker read them.

Menus, product names, legal text, calls to action and anything with a brand voice are the places where a human read pays off most. Plain informational paragraphs carry the least risk when they go out with lighter review.

Does Google penalize machine-translated pages?

Google does not treat machine translation as spam in itself. Google told Search Engine Land in 2025 that its policies do not strictly define content translated by AI as spam, while it still points to its scaled content abuse policy for the limits.

In the same year, Multilingual reported that Google had softened its automated translation policy by removing earlier guidance that told site owners to block AI-translated content, provided it offers user value.

The risk sits in the scaled content abuse policy. GSQI points out that the policy lists automated transformations, including translating, among the techniques that can violate it when many pages are produced with little value for users.

The production method is not the test. ClearPost's explainer notes that the policy applies whether content is produced through automation, human effort or a mix of both. What matters is the value of the resulting pages: a large set of thin, unreviewed ones is the problem.

Older advice points the other way. In 2020, Search Engine Land reported that Google's Gary Illyes said Google did not want machine-translated content in its index. Date any guidance you find before you apply it.

A workable test for each translated page has three questions. Does it answer something a local visitor would look for? Has someone reviewed it for accuracy and tone? Does it exist for a reason beyond filling out a language list? For a longer look at how translated pages relate to duplicate content, see Lantis's article on duplicate content and translations.

How should URLs and hreflang be set up before translating?

Google's guidance on structure is direct:

"Google recommends using different URLs for each language version of a page rather than using cookies or browser settings to adjust the content language on the page." — Google, Search Central documentation

Three further points come from the same documentation:

  • Redirects: Google advises against automatically redirecting users to a language version based on their assumed language, because that can keep search engines and users from seeing every version.
  • Dynamic switching: if the language changes on the fly, Googlebot may miss variants, because it usually crawls from the USA and does not send an Accept-Language header.
  • Partial translation: translating only the boilerplate of a page while the bulk of the content stays in one language can produce a poor experience in search results.

Hreflang belongs on every applicable page. Seobility's 2026 guide says to add hreflang tags to all applicable pages, not only the homepage. The same guide adds that some infrastructure cannot output hreflang in the page source but can support XML sitemaps that reference hreflang tags. Confirm which of the two your platform supports before launch, because a limitation found later means rework on every URL.

Tools differ in how much of this they handle for you, and the sources consulted do not detail URL, hreflang or sitemap handling for Webflow's localization, Shopify Translate & Adapt or Weglot, so check each one's documentation against the Google guidance above. Lantis creates a separate URL per language and generates hreflang and sitemaps automatically. It works on Framer, Webflow, Shopify, WordPress and any other website, and it is installed by pasting one line of code, without changing DNS or the design. If your store runs on Shopify, a step-by-step hreflang guide for Shopify covers the manual route.

What limits do native localization apps have compared with other tools?

The comparison below applies four criteria to every option: price, language cap, translation engine and glossary. Where a cell says "Not detailed," the sources do not cover that criterion.

OptionPriceLanguagesTranslation engineGlossary
Webflow Localize Essential$9 per locale per month, billed yearly, per Superblog's analysisUp to 3 localesNot detailedNot detailed
Shopify Translate & Adapt (free app)Free2 extra languages, or 8 for stores in the UK, Switzerland or 10 EU countries, per Weglot's app reviewGoogle Cloud Translate, per Transcy's reviewNo dedicated glossary, per Outfy's review
Weglot (third-party tool)Paid plans from €15 per month, per Weglot's own pageNot detailedNot detailedNot detailed
Lantis3 € per month for 1 language; 2,50 € per language for 2 to 9; 2 € per language for 10 or moreMore than 60 availableGemini by default; Claude, DeepL or DeepSeek also availableYes, per language

On Webflow, Superblog's pricing analysis states that as of September 2026 Localize Essential cost $9 per locale per month billed yearly and covered up to 3 locales. According to Weglot's comparison of Webflow tools, Essential includes machine-powered translation capped at 10,000 words per locale per month.

Webflow's Localization add-on is also unavailable on Ecommerce sites. Creative Corner's analysis explains that Ecommerce products, categories and pages cannot be localized, so a Webflow store needs another route for those pages. For a Webflow site without a store, native localization may be enough if three locales cover your needs and you are happy with its machine translation allowance.

On Shopify, Weglot's app review says the free Translate & Adapt app automatically translates content into up to 2 additional languages. For a store that needs only those languages and no glossary or image translation, the free app can be the better fit than a paid tool. Weglot's own pricing is a vendor claim, and its competitor status applies to the Shopify and Webflow comparisons it publishes. For a fuller view of the app's gaps, see the limits of the free Translate & Adapt app and alternatives to it.

Lantis charges only for active languages and places no limit on the number of websites or visitors. It includes 500.000 words per language and has no lock-in period, with a full refund within 14 days after each payment. Its limits mirror its pricing: the Agency plan covers up to 20 websites and up to 5 languages per website, and words beyond the included amount are bought in blocks of 500.000. It is a paid, separate tool, so a small site that fits within Webflow's native locales or Shopify's free languages may not need it.

How does cost grow with each active language?

Pricing models fall into three groups: a fixed price per language or locale, tiers by number of languages, and a word allowance. The same site can cost very different amounts depending on which applies, so start by identifying the model.

Per-locale pricing scales linearly. At the Localize Essential rate in Superblog's analysis, three locales come to $27 per month, which is also the most that tier covers. A fourth language requires a different tier.

Tiered pricing lowers the marginal cost. Lantis charges 3 € per month for one active language, 2,50 € per language per month from 2 to 9 languages, and 2 € per language per month from 10 languages. Webflow's figures are in dollars and Lantis's in euros, so compare the arithmetic for your own language count rather than the headline prices.

Word allowances matter as much as price. Weglot's comparison puts the Essential machine translation cap at 10,000 words per locale per month. Lantis includes 500.000 words per language, and each extra 500.000 words cost 1 € as a one-time payment. Count the real words on your site, including product pages and blog posts, before choosing a plan.

Free tiers cap languages rather than price. According to Weglot's app review, Shopify's free app covers two extra languages in most locations, which suffices for a test but not for a multilingual catalogue. Weglot says its paid plans start at €15 per month and scale with what a business needs.

How much human review does each page need?

Smartling notes that translations sometimes need more editorial intervention from the client, so plan a human step rather than publishing the first draft. Match the level of review to the type of page:

  • Highest weight: home, pricing, category and product pages, and checkout or legal text. Have a native speaker who knows the product review these.
  • Informational pages: blog posts and plain explanatory content. Check a sample after launch and widen it if the same errors repeat.
  • Images with text: banners and infographics. Outfy's review says Shopify's free app has no image translation features, so plan a separate process if your tool lacks them.

Build a glossary before translating. Product names, feature labels and terms that must stay in English should be fixed for each language, so the same term does not appear three different ways across the site. Shopify's free app has no dedicated glossary. Lantis includes a glossary per language, a visual editor to correct translations, and the option to exclude pages or blocks from translation.

Whatever tool you choose, check that you can correct the output without breaking the site. Fixing a translated page before it is indexed is easier than fixing it afterwards.

Next steps

  • Inventory: count the pages and words you plan to translate, and mark which pages are informational and which carry brand or legal weight.
  • URL structure: decide how each language will be addressed, and check whether your platform outputs hreflang in the page code or needs sitemaps for it.
  • Pilot: translate a small set of pages first, review them with a native speaker and a glossary, and check how they appear in search results and sitemaps before scaling up.
  • Cost: calculate the monthly cost for your language count and word volume, including any extra word packs, rather than comparing headline prices.
  • Tooling: if you want the URL, hreflang and sitemap setup handled automatically, Lantis does it with one line of code and no DNS changes, and its multilingual SEO checklist lists what to verify after launch.

Frequently asked questions

Does Google penalize machine-translated websites?

Google told Search Engine Land that its policies do not strictly define AI-translated content as spam. The risk lies in publishing many pages with little value for users, which can fall under the scaled content abuse policy that lists translating among automated transformations.

How do I set up hreflang for a multilingual website?

Give each language version its own URL, then add hreflang to all applicable pages, not only the homepage, as Seobility's guide advises. If your platform cannot output hreflang in the page code, XML sitemaps can reference it instead.

Can I translate a Shopify store without a paid app?

Yes, within limits. Shopify's free Translate & Adapt app translates into up to 2 additional languages, or 8 for stores based in the UK, Switzerland or 10 EU countries, according to Weglot's review. Outfy's review notes it has no dedicated glossary or image translation features.

Is machine translation good enough for a marketing website?

It works as a fast first draft. Smartling says quality may vary and that machine translation mainly addresses the linguistic side of localization. Pages that carry brand voice, calls to action or legal text should be reviewed by a native speaker before publication.

Sources

Comprobador de SEO multilingüePega la dirección de cualquier página de tu web. La leemos como lo hace Google y revisamos hreflang, lang, canonical y cada versión traducida, con la solución a cada problema.

Pon tu web en varios idiomas hoy.

Una línea de código. SEO real por idioma. Traducción con IA. Sin tarjeta para empezar.

Sigue leyendo