Machine Translation SEO: Why Translated German Pages Stay Invisible in AI Search
Machine translation SEO rarely fails on grammar. A translated German page loses in classic search and gets passed over as a source in AI answers. How to spot one, what Google actually says and what a localisation audit checks.
Key takeaways
- Google does not penalize translation, it penalizes translation without value. Its spam policies list automated translating that adds little for users as an example of scaled content abuse.
- Translated pages often target the wrong keyword. The translator picks the dictionary match, not the phrase German buyers actually type.
- AI answers quote what they can check. Cautious translation removes exactly those clear, attributable sentences.
- The real problem is ownership. Nobody measures the German version against the German market, so it gets measured against the source.
In this article
- Why machine translation SEO is a visibility problem now
- How can you tell a page was translated rather than written?
- Does Google penalize machine-translated content?
- AI translation SEO: why do AI answers skip translated pages?
- What should a localisation audit check?
- Translation vs localisation: what is the real difference?
- Frequently asked questions
Why machine translation SEO is a visibility problem now
A German page can be grammatically flawless and still read like a customs form. That is the version of your site most DACH buyers see, and it is increasingly the version an AI assistant decides not to quote. Translation quality stopped being only a brand problem. For machine translation SEO it is now a visibility problem, in classic search results and in generated answers alike.
I have spent over ten years producing German content, and a large share of that work was repairing pages somebody had already translated. The pattern repeats across industries. The English or Spanish original is sharp, the German version is technically correct, and nobody in the German market wants to read it. What changed is the cost of that gap. A thin German page used to cost you a few positions. Now it can cost you the ability to appear at all in the answer that sits above the positions.
How can you tell a page was translated rather than written?
You rarely need a linguistics degree for this. The tells are structural. Sentence length stays suspiciously even, because the translator followed the rhythm of the source instead of German rhythm. Compound nouns get unpacked into prepositional constructions no German writer would build. English collocations survive intact and read like costumes: Lösungen liefern, Werte schaffen, Ihre Reise beginnt hier. Each is defensible alone. Together they signal a page produced elsewhere and pointed at Germany afterwards.
The deeper tell is what the page assumes about its reader. A Spanish original written for a Spanish buyer carries Spanish objections, Spanish payment habits, Spanish trust signals. Translate it and you keep the objections of a market you are not selling into. The German reader gets careful answers to questions they never asked, and silence on the one thing actually blocking the decision: who is liable, what the invoice looks like, whether there is a named human behind the company, whether the terms will survive a procurement review. None of that arrives through a translation layer, because none of it was in the source.
Quick check
Read the German H1 and the first paragraph out loud. If you can guess the English sentence behind every line, so can your German reader.
Does Google penalize machine-translated content?
Not for being translated. There is no machine translation penalty to point at, and anyone who tells you there is one is selling something. Google’s spam policies draw a narrower line: under scaled content abuse they list content run through automated transformations such as translating, where little value is provided to users. The question is not who or what translated the page. It is whether the page helps the person reading it.
Most translated pages never get near that line. They lose more quietly. The page ranks, gets its click, and the visitor leaves, because the headline promised a German answer and delivered an English argument wearing German words. A search engine does not need to detect translation for that to hurt. Your reader already did.
Machine-translated content misses German search demand
There is also a demand problem translation structurally cannot solve. German search demand does not map one to one onto English search demand. A single English term often has two or three German equivalents with completely different intent behind them, and a translator picks the dictionary match rather than the phrase people type. The page then competes for a keyword nobody searches in a market where it had a genuine opening. That is not a language failure. It is a research step that was never budgeted.
What Google’s spam policies actually say
Scaled content abuse means many pages generated mainly to manipulate rankings rather than to help users. One of the listed examples is content scraped and pushed through automated transformations like synonymizing or translating, where little value is provided to users. Translation as such is not on the list. Translation without added value is.
AI translation SEO: why do AI answers skip translated pages?
This is the part still being underrated. An assistant or an AI search surface does not hand you ten blue links. It assembles an answer and leans on a selection of sources. What gets picked, in my experience, is what is easy to extract and safe to attribute: a clear claim, a specific figure, a named entity, a sentence that can be lifted without losing its meaning.
Google’s own guidance for AI Overviews and AI Mode lists no extra requirements. A page has to be indexed, eligible for a snippet and follow the same SEO basics as everywhere else. So there is no AI trick to buy, and a translated page enters that selection with the same weaknesses it has in classic search.
Translated pages are bad at exactly what gets picked. They hedge, because the translator softened every claim they could not verify. Market-specific detail rarely survives the move, so the copy stays generic. And instead of answering the German question head on, the page restates the source framing. A page that says nothing checkable in German gives a German answer nothing to cite, so the model cites a competitor who wrote for the market directly. Your brand is then absent from the only surface that buyer looked at, and no ranking report will tell you why.
Being the best page in English offers no protection. If the query is German, German sources win.
Lukas Thies
What should a localisation audit check?
When I audit a German site version, I am not looking for mistranslations. I am looking for the decisions nobody made. The checkpoints that matter:
- Search demand mapped in German, not translated. Build the keyword set from German queries and rank the variants by intent before a single sentence is written. If the German page inherited its target term from the English brief, that is the first thing to fix.
- Headline and opening paragraph answering a German question. The first fifty words decide whether the reader recognises their own situation. A translated intro almost always opens with the company instead of the reader.
- Market-specific proof present at all. Imprint, named contact, VAT handling, contract terms, references a German buyer can verify. These are conversion elements in DACH, not legal decoration.
- At least a few quotable, checkable sentences per page. Specific claims with numbers or named conditions. This is what makes a page usable as a source in a generated answer, and it is the first thing a cautious translation removes.
- One concept, one German term, across the entire site. Navigation, product pages, help centre, emails, invoices. Terminology drift between systems is the clearest sign that nobody owns the German version.
- Register and form of address decided once and held. The Sie or du call is a positioning decision, not a style preference. Nothing looks more outsourced than a site that switches between them mid funnel.
- Structure and internal links following German demand. German site trees rarely need to mirror the English one. If the page hierarchy was cloned, hreflang is tidy and the information architecture is still wrong.
If the German blog has gone quiet as well, the same logic applies to the archive: here is what a silent German archive tells buyers, crawlers and AI answers.
Translation vs localisation: what is the real difference?
Translation takes a finished argument and changes its language. Localisation rebuilds the argument for a different buyer and keeps the strategy intact. That is why localisation gets scoped like content work rather than priced by the word, and why handing it to the cheapest per-word vendor produces exactly the pages described above. The vendor did what was asked. The brief was the problem.
Running content for German alongside Dutch and Belgian markets taught me the same lesson repeatedly: the pages that performed were never the ones that mirrored the source most faithfully. They were the ones where somebody had been allowed to cut a section that made no sense locally, add one the source never needed, and change the angle of the headline entirely. In regulated German-speaking markets, at bwin, PartyCasino and Sportingbet under Entain, that freedom was not a luxury. It was the only way to produce something a German reader and a compliance reviewer would both accept. In CRM copy the same rule decides whether a German player opens the mail at all.
My honest read: most companies do not have a translation quality problem, they have an ownership problem. Nobody on the team is accountable for whether the German version wins in Germany, so it gets measured against the source instead of against the market. Fixing that is not expensive. It starts with auditing the ten pages that carry the revenue, rewriting them for German demand rather than English structure, and treating the rest as a backlog. The upside is no longer just better rankings. It is being the source the answer quotes. If you want to see how I set that up as a workflow, here is how I work.
Frequently asked questions
Does Google penalize machine-translated content?
Not for being translated. Google’s spam policies list automated translating that provides little value to users as one example of scaled content abuse. A machine-translated page that has been reviewed, adapted to the market and genuinely helps its readers does not fall under that. The bigger risk is quieter: the page targets the wrong keyword and convinces nobody.
Is machine translation good enough for SEO?
As a first draft, often yes. As the finished page, rarely. Machine translation carries over the words of the source, but not the search terms the target market uses, the objections a local buyer has or the proof they expect. Keyword research in the target language and an editorial rewrite still have to happen.
Are machine translations getting better?
At sentence level, yes. Grammar and fluency are no longer the weak spot. What a better engine cannot change is the input: it still translates the source’s argument, keywords and assumptions, so the strategic gap stays where it was.
How do AI answers treat translated pages?
Google says AI Overviews and AI Mode have no extra requirements beyond being indexed and eligible for a snippet. In practice, answers lean on sources with clear, checkable statements. Translated pages tend to hedge and generalise, so the answer often cites a competitor who wrote for the market directly.
Work with me
Not sure whether your German pages were written or just translated?
Send me your three most important German pages and I will tell you what a German buyer sees. Thirty minutes, free of charge, no pitch deck.
