Machine-translated content and SEO

What translation changes

Serving a page in another language can open a market that would otherwise be unreachable, and machine translation makes that cheap to attempt. The same automation, however, can produce pages that are technically indexed but useless to a native reader, and a gibberish page helps no one.

The position to hold is simple: translated content is judged by how a human in that language reads it. Search systems rank pages for what users actually experience, so a translation that reads fluently and answers the query is a real asset, while a literal, garbled translation can rank where nobody opens it, or worse, look like a thin page designed only to cover a market.

Before you publish

Run a translated page through the same test you would run on original content:

• Does a native speaker understand every sentence, not just the general idea?
• Are product names, prices, units and legal terms correct for the locale, not just the words?
• Do technical phrases and marketing claims survive without becoming wrong or offensive?
• Is the translation on-topic for the same query it targets in the source language?

Machine translation of short, factual, unambiguous copy often passes this test. Idiomatic marketing, humour, regional slang and long technical prose frequently do not. Plan to have a reviewer check anything you cannot personally verify.

If you use machine translation, keep the source page authoritative and treat each locale as its own page. Consider editing the output rather than publishing it untouched, because editors catch the errors that an automated pipeline cannot.

Hreflang and dupes

Translated pages introduce the two structural pitfalls to handle before publishing:

• Cross-language duplicates: if two locales carry essentially identical content, they can look like duplicates. Make sure each language page is genuinely its own authored version, and use hreflang to tell search engines which URL serves which language and region.
• Wrong hreflang pairs: hreflang must be reciprocal and point to every alternative, including the page itself. A one-way or mismatched set produces a confusing signal and can suppress the pages it touches.

Two further cautions:

• Do not publish a machine translation purely to increase page count. Volume without genuine value risks the same treatment as any thin or near-duplicate content, and it spends crawl budget on pages that convert nothing.
• A page that exists only as an empty shell after translation, with no usable content, behaves like a soft 404. It indexes as a page that should satisfy a query and does not.

Set up hreflang correctly and keep each locale a real, edited page rather than an automated echo of the source.