Multilingual websites
Multilingual website launch for Google and AI answers
This playbook shows how to launch a multilingual website that Google and AI answers can read. It covers one source of facts, a URL per language, native copy, hreflang, a sitemap, structured data and checks before every release. AI for Marketing Ops used this method for RNJ Transport, whose site went live 2 days after the first call.
Updated:
- Who it is for
- Marketing teams at companies that sell in more than one language and plan to build or rebuild their website.
- Time
- About 4 weeks to launch, then a few hours every two weeks for new content.
- Outcome
- A live website in every language, with hreflang, a sitemap, structured data, llms.txt and checks that run before each release.
- Tools
- Google Search Console · Rich Results Test · Lighthouse · A static site framework such as Astro · Claude Code
Steps
- 01 Week 1
One source of facts
Write every company fact once, in one file, before anyone writes a page. Each language then reads the same phone number, address and service list, so the versions cannot disagree.
The old RNJ Transport site showed conflicting phone and address details, and "30 years" where the true figure was 3. With one facts file, a correction in one place reaches every language.
What to do
- List names, phone numbers, addresses, opening hours and legal details.
- List the services, prices and numbers you will state on the site, each with its source.
- Mark facts that are not confirmed and do not publish them.
- Name one person who approves changes to the file.
When it is done: One facts file that every language version reads from.
- 02 Week 1
Routes and slugs per language
Give every page its own URL in each language, and keep all routes in one map. Google recommends a different URL for each language version. It warns that it might not find all versions when content changes with the browser’s language settings.
Google describes folders such as /es/ as easy to maintain. Write each slug in the language of the page, so a Spanish page has a Spanish address.
What to do
- Choose one URL structure: folders, subdomains or country domains.
- Write a route map with each page key and its slug in every language.
- Make the language switch link to the same page in the other language.
- Remove automatic redirects based on browser language or location.
When it is done: A route map that the links, the language switch and the sitemap all read from.
- 03 Week 2
Native copy per language
Write the main text of each page for the people who search in that language. Google counts localized pages as duplicates only when the main content stays untranslated, and it detects the language from the visible text.
Start with the words buyers type in each market. English buyers search "multilingual website" and Spanish buyers search "web multilingüe", so each version needs its own keyword research.
What to do
- Research the search queries in each language with a keyword tool or Search Console.
- Write titles, headings and descriptions in each language, as well as the body text.
- Ask a native speaker to read every page before it goes live.
- Keep the same sections and facts in every language, so no version has less content.
When it is done: Copy files for every page in every language, with the same structure.
- 04 Week 3
Hreflang and sitemap
Tell Google which pages are versions of each other with hreflang, and list every URL in an XML sitemap. Each version must list itself and all other versions. If two pages do not point to each other, Google ignores the tags.
Google treats HTML tags, HTTP headers and the sitemap as equal methods, so one method is enough. The RNJ Transport sitemap carries the hreflang links, with English as x-default. On 3 November 2026 it listed 1,945 pages in 11 languages.
What to do
- Use ISO 639-1 language codes such as en, es and lt, and add a region code only when you need one.
- Add x-default for the page that serves visitors in other languages.
- Use full absolute URLs, and split the sitemap above 50,000 URLs or 50 MB.
- Submit the sitemap in Search Console and add it to robots.txt.
When it is done: One sitemap with every URL and its language versions, submitted in Search Console.
- 05 Week 3
AI crawler access, llms.txt and Markdown copies
Let AI and search crawlers read the site, and give AI tools a short map of it. llms.txt is a proposed standard from 2024. It is one file that lists your key pages for AI agents, with a clean Markdown copy of each page.
Google says you do not need AI text files or special markup to appear in AI Overviews or AI Mode. We add these files for other AI tools that read websites, because they cost little to build. RNJ Transport serves a Markdown copy of every page and allows 11 named AI and search crawlers in robots.txt.
What to do
- Decide which AI and search crawlers you allow, and name them in robots.txt.
- Write llms.txt with your company facts and links to the key pages in each language.
- Generate a Markdown copy of every page from the same copy files.
- Check that no page blocks crawlers by mistake.
When it is done: robots.txt, llms.txt and a Markdown copy of every page, in every language.
- 06 Week 3
Structured data in every language
Add schema.org structured data that describes each page type, in the language of the page. Google’s rule is that markup must match the visible text and must not describe content readers cannot see.
Common types are Organization, WebSite, BreadcrumbList, FAQPage and Article. Google does not guarantee rich results, even when the markup passes its test.
What to do
- Choose the schema types for each page template.
- Fill names, descriptions and FAQ answers from the same copy files as the page.
- Test each template in the Rich Results Test.
- Check one page per language with the URL Inspection tool in Search Console.
When it is done: Structured data on every template, tested and matching the visible text.
- 07 Week 4
Checks before every release
Write a script that blocks a release when a rule breaks. People forget checks, and a script runs them every time.
On RNJ Transport, one command runs before each release. It checks copy rules, the same pages in every language, types, the build, title and description lengths, menus and the sitemap. On 19 October 2026, Lighthouse on mobile scored the site 97 for performance and 100 for SEO.
What to do
- Stop the release when a page is missing in one language.
- Check title and description lengths for every page in every language.
- Check that every hreflang link has a return link and that every internal link works.
- Run Lighthouse on mobile for the main templates.
- Check copy rules, such as banned words, in every language.
When it is done: One command that must pass before anything goes live.
- 08 After launch, every two weeks
Content on a schedule
Publish new guides and blog posts on a fixed schedule in every language, and review Search Console per language each month. A fixed schedule keeps every language growing at the same pace.
RNJ Transport has blog posts scheduled every two weeks in each language. A monthly SEO and AI search audit is scheduled too.
What to do
- Plan topics from the questions buyers ask in each market.
- Write each post in each language, from the same facts file.
- Run the release checks on every new post.
- Review clicks and indexed pages per language in Search Console each month.
When it is done: A content calendar per language and a monthly report.
Mistakes to avoid
- Redirecting visitors by browser language or location. Google warns that this can stop people and crawlers from seeing all versions.
- Translating only the menu and footer while the main text stays in one language.
- Hreflang without return links, or pages that do not list themselves. Google ignores tags that do not point both ways.
- A language switch that links to the home page or to "#". The old RNJ Transport site had this problem.
- Publishing machine translations with no native review. Google’s spam policies count many low-value pages made mainly to rank as scaled content abuse, no matter how they are made.
What to measure
| Metric | How to measure | When |
|---|---|---|
| Indexed pages per language | Search Console page indexing report, with one URL-prefix property per language folder | Weekly in the first month, then monthly |
| Hreflang errors | Release checks: every alternate URL has a return link and lists itself | Before every release |
| Clicks and impressions per language | Search Console performance report, filtered by the language folder | Monthly |
| Brand named in AI answers | Ask ChatGPT, Gemini and Perplexity the same buyer questions in each language and record whether they name you | Monthly |
| Mobile speed | Lighthouse on mobile for each main template | Before every release |
Rather have us run it
We run this playbook with you or for you
Sources
- Google Search Central: tell Google about localized versions of your page
- Google Search Central: managing multi-regional and multilingual sites
- Google Search Central: build and submit a sitemap
- Google Search Central: general structured data guidelines
- Google Search Central: AI features and your website
- Google Search Central: spam policies for Google web search
- llmstxt.org: the llms.txt proposal
Questions
Common questions
How long does it take to launch a multilingual website?
Plan for about 4 weeks if your team writes the copy and follows the steps above. The RNJ Transport site went live 2 days after the first call, with 944 pages. In our projects, AI does the repeat work and people review it. Most of our website projects go live in 10 working days.
Do I need hreflang if each language has its own URL?
Yes. Hreflang tells Google which pages are versions of each other, so it can show each searcher the right one. Google does not use hreflang to detect the language of a page. It reads the visible text. Each version must list itself and every other version.
Is machine translation enough for multilingual SEO?
Not without review. Google detects the language from the visible text. Its spam policies count many low-value pages made mainly to rank as abuse, no matter how they are made. Use AI for the first draft if it saves time, then ask a native speaker to review every page. Write titles and descriptions for the queries people type in each market.
Does llms.txt help a site appear in Google AI Overviews?
Google says you do not need AI text files or special markup to appear in AI Overviews or AI Mode. llms.txt is a proposed standard that other AI tools can read. We add it, with Markdown copies, because it costs little to build. We make no promise about its effect.
Book a call to plan your multilingual website
On a free 30-minute call, we look at your current site and the languages you need. You keep the short plan, even if we do not work together. If you want us to run this playbook, our website and brand revamp service does it as one project. Most projects go live in 10 working days.
Book a discovery call