We first published a guide to multilingual SEO a few years ago, and most of the fundamentals in it still stand. What has changed is the thing doing the searching. Since that original piece, AI-generated answers have moved from an experiment to the default experience for a large share of queries – and that shift changes how translated content gets found, or missed. This is a full rewrite for how search actually works now.
Here’s the short version before the detail: getting your medical or technical website found in another language is no longer only about ranking a page. It’s increasingly about whether an AI system reads your content, trusts it, and repeats it. And weak translation is now one of the fastest ways to be left out of that.
What actually changed in search
For most of its history, search engine optimisation meant climbing a list of blue links. That list is shrinking in importance. Google’s AI Overviews now appear above the traditional results for a meaningful share of searches, and instead of offering ten pages they read the pages and generate a single answer. Answer engines – ChatGPT, Gemini, Perplexity – do the same in their own interfaces, and a growing number of people use them the way they used to use a search box.
The practical result is the rise of the “zero-click” search: the user gets their answer in the summary and never visits a website at all. Studies tracking AI Overviews have found organic click-through rates falling sharply on the queries where they appear.
For anyone optimising a website, the goal has widened. You still want to rank. But you also want to be the source the AI selects and cites – a discipline now commonly called Generative Engine Optimisation (GEO) or Answer Engine Optimisation (AEO). The names matter less than the shift they describe: you are no longer only feeding a crawler that indexes words, but a model that reads meaning, weighs credibility, and decides whether your content is worth repeating.
Why this is harder for multilingual content
Here’s the trap for technical and medical companies. AI summaries do not stay within one language. They draw on multiple sources, sometimes across several languages at once, and assemble the response from whichever content reads as the most complete, coherent and trustworthy answer to the query.
Weak or loosely adapted translations get passed over. Not flagged or penalised – simply skipped, because a better-written source in that language wins the citation instead. A page can exist in German or French, be technically correct, and still be invisible in practice because it doesn’t read like something a native expert would write.
Machine-translated content that was “good enough” to fill a page in 2023 is exactly what these systems now step around. For regulated content, where precision was already non-negotiable, that raises the stakes: the same qualities that make a translation compliant and safe are now also what make it visible.
Nine things to actually do
The principles from the original guide still apply – they’ve simply been recast for a world of answer engines. Here’s the concrete version.
1. Research keywords in the target language, not from your English list
Direct translation of keywords misses how people actually search, and AI systems reward content that matches real intent. Work with a native speaker or in-market SEO specialist, and set your research tools (Google Keyword Planner, Ahrefs, Semrush) to the target country and language – not just the language. The phrase a German engineer types is often not the literal translation of the English one.
2. Build a bilingual glossary before translation starts
Agree the correct term for each key concept, product feature and regulatory phrase in every target language, and lock it in a termbase before a single page is translated. Consistency across your pages is one of the signals these models use to judge whether a source is authoritative. Inconsistent terminology reads as unreliable – to a human and to a machine.
3. Structure pages so the answer comes first
Answer engines extract. Give them something clean to extract. Phrase a heading as the question a user would ask, then put a direct, self-contained answer in the first sentence or two underneath before you expand. A short summary near the top of the page helps too. The aim is that any single paragraph can stand alone as a quotable answer.
4. Implement hreflang correctly
Hreflang tags tell search engines which language and region a page serves, so the right version reaches the right user. Get them right across every language variant, include a self-referencing tag on each page, and use the correct region codes where you serve variants of the same language – European versus Brazilian Portuguese being the classic example in regulated markets.
5. Add structured data
Schema markup helps machines understand – and safely extract – your content. FAQ schema, Product schema and, where appropriate, MedicalWebPage schema all give an AI cleaner signals about what a page contains and how credible it is. Make sure the markup itself is translated and localised for each language version, not left in English.
6. Localise properly, don’t just translate
Adapt units, currencies, date formats and, critically, the regulatory references for each market. A technical spec that cites a UK standard means little to a reader in France; content that reflects local standards, terminology and expectations both converts better and reads as more authoritative to the systems ranking it.
7. Earn authority in-market
Backlinks from reputable local publications, industry bodies and institutions still build credibility, and credibility is what makes an AI willing to repeat you. Prioritise genuine in-market authority – a mention in a respected national trade publication is worth more than volume of thin content, which the models have become good at seeing through.
8. Write the way people speak
A large share of AI-assistant queries are spoken and conversational, and they draw on the same sources as text search. Content written in full, natural sentences that mirror how people actually ask questions travels further across both text and voice.
9. Track citations, not just rankings
Rankings are no longer the whole picture. Periodically ask the major AI tools the priority questions your customers would ask, in each target language, and note whether you’re named. In GA4, set up a custom channel to separate referral traffic arriving from AI platforms. If you can’t see whether the AI layer is finding you, you can’t improve it.
Where translation quality sits in all this
Every point above rests on the same foundation, and it’s the one thing AI has not changed: the content underneath has to be accurate, natural, and written with real subject-matter understanding.
For medical and technical companies that isn’t optional. A mistranslated specification or an imprecise instruction for use was always a compliance and safety problem. Now it’s a discovery problem as well, because the content that reads as authoritative to a human expert is the same content the answer engines choose to cite. Machine translation has a role in the workflow – but for the pages you actually want found, quoted and trusted in another language, the gap between “translated” and “translated well” has never been more visible.
If you’re planning international expansion, the useful question is no longer just whether your site has been translated. It’s whether the translated version is good enough to be trusted by the market, and by the systems now doing the finding on the market’s behalf. That’s exactly the standard our medical and technical translation services are built to meet.
Ready to make your multilingual content genuinely discoverable? Request a translation quote and we’ll help you get it right at source.
Frequently asked questions: Multilingual SEO
What is multilingual SEO?
Multilingual SEO is the practice of making translated web content discoverable in international search – both traditional search engines and, increasingly, AI-generated answers. It combines accurate translation with in-language keyword research, technical signals like hreflang, and structuring content so it can be found in each target market.
Is translating my website enough to rank in other languages?
No. Translation makes your content readable; it doesn’t make it discoverable. Ranking in another language also needs keyword and intent research in that language, correct hreflang and structured data, localised content, and in-market authority. A page nobody searches for the way you’ve phrased it won’t be found.
What is Generative Engine Optimisation (GEO)?
GEO – sometimes called Answer Engine Optimisation (AEO) – is the practice of structuring content so AI systems like Google’s AI Overviews, ChatGPT and Gemini will reference and cite it in their generated answers. It focuses on clarity, structure, credibility and extractable facts rather than keyword frequency alone.
How do AI Overviews affect my international traffic?
AI Overviews answer many queries directly on the results page, which reduces clicks through to websites. For multilingual content this raises the bar: the AI selects sources across languages based on quality and clarity, so weak translations are more likely to be passed over in favour of stronger in-language content.
Do I still need hreflang tags in 2026?
Yes. Hreflang tags remain the standard way to tell search engines which language and region each page serves, so the correct version reaches the right user. They also help AI systems interpret your content correctly. Include a self-referencing tag on every page and use accurate region codes.
Why do weak translations get ignored by AI search?
AI systems assemble answers from the sources that read as most complete, coherent and credible. A translation that is stilted, inconsistent in terminology, or clearly machine-generated reads as less trustworthy, so a better-written source in that language tends to be cited instead – even if your page technically exists.
Should I use machine translation for SEO content?
Machine translation can support internal or high-volume, low-stakes content, but for the pages you want found, quoted and trusted – especially regulated medical and technical content – specialist human translation is what meets the quality bar AI search now rewards. Poor translation is now a visibility problem as well as a compliance one.
How is multilingual SEO different for medical and technical content?
Medical and technical content carries compliance, safety and terminology requirements that general web content does not. Terms must be exact, regulatory references must be localised, and accuracy is non-negotiable. That specialist precision is also what makes the content authoritative enough to be cited by AI search, which is why subject-matter expertise matters more here than in most sectors.