Machine Translation Only

What this check means

Machine Translation Only means a localized page appears to be pure automated output. Someone ran the source text through Google Translate, DeepL, an AI model, or a plugin like the free tier of a WPML or Weglot auto-translate, and published it without a human touching it afterward. The words are technically in the target language, but nobody who actually speaks that language read them before they went live.

Modern engines are good. They are not good enough to ship unreviewed on pages that represent your brand or sell your product. They miss idiom, they pick the wrong sense of ambiguous words, they mistranslate your product names and calls to action, and they produce sentences that a native speaker clocks as robotic in about two seconds. That gut reaction is exactly what you do not want a paying customer to feel.

Where Google stands on this

Google's own documentation on spam policies calls out "text translated by an automated tool without human review or curation before publishing" as an example of scaled content abuse. Read that carefully: the problem is not machine translation itself, it is machine translation without human review, produced at scale to fill an index. Google has said for years that automatically generated content presented as original can be treated as spam. So this check is not academic. It maps directly to a documented policy line.

The nuance that matters: Google does not penalize you for using a translation tool. It cares about the result. A machine draft that a bilingual editor cleaned up is fine. A thousand pages of raw output dumped into forty languages to blanket the world is the pattern that gets sites flagged for scaled content abuse.

The translation quality spectrum

Raw machine flagged here

Post-edited MT acceptable

Human translation good

Localized best

Move every page to the right of the red zone before you publish it.

Why raw machine translation costs you

Three things go wrong at once. First, rankings: thin, awkward translated copy earns weak engagement, and quality signals in that market suffer. Second, conversions: even if a page ranks, wording that reads as foreign or plain wrong tanks trust, and trust is what makes someone hand over a credit card. Third, risk: at scale, unreviewed output is the exact pattern Google's scaled-content policy targets, which puts the whole localized subfolder or subdomain in jeopardy, not just one page.

There is a compounding factor with hreflang. If you have gone to the trouble of setting up proper hreflang annotations pointing to these translated pages, you are actively telling Google to serve low-quality output to users in that region. Good technical setup pointed at bad content just delivers the bad content more efficiently.

How to detect it

• The only real test is a native speaker. Have someone fluent read a sample of pages. They will tell you in minutes whether it reads as human. No tool replaces this.
• Screaming Frog with language extraction: pull the visible text of your international URLs and spot-check. You can also flag pages where the declared lang or hreflang says one language but patterns look auto-generated.
• Check your translation stack: if you run WPML, Weglot, TranslatePress, or a similar plugin, look at whether "automatic translation" is on with no editorial workflow behind it. That configuration is the source of the problem.
• Search Console by country: a market with impressions but near-zero clicks often means the pages rank a little but nobody engages, a classic symptom of copy that reads wrong to locals.
• Look for tell-tale artifacts: untranslated UI strings mixed with translated body text, wrong-gender articles, literal idioms, and product names that got translated when they should have stayed fixed.

How to fix it, step by step

1. Triage by value. You will not human-edit everything overnight. Rank your international pages by traffic and revenue potential and start at the top.
2. Post-edit, do not restart. Keep the machine draft as a base and have a native-speaking editor rewrite it into natural, on-brand copy. This is faster than translating from scratch and gets you out of the red zone.
3. Fix the terminology. Lock product names, brand terms, and calls to action so they are consistent and correct across the language.
4. Turn off blind auto-publish. Reconfigure your plugin so machine translation creates a draft that a human approves, never a live page.
5. Noindex or delist the worst offenders in low-priority markets until they can be properly edited, so you stop serving weak pages while you work through the queue.
6. Recheck hreflang so your annotations only point at pages that are actually ready.
7. Re-audit with a native reviewer once edits ship, and monitor engagement by country in Search Console.

Machine only vs post-edited vs human

What good looks like

A healthy international site reads like it was written by someone from that market, because effectively it was: a native speaker shaped every published page, even if a machine produced the first pass. Terminology is consistent, calls to action land naturally, and engagement metrics in each country look like a real audience rather than accidental impressions. You still use machine translation, you just never let it be the last step. That is the whole game here, keep the speed of the machine and add the judgment of a human before anything goes live.

FAQ