Building a Multilingual Website: hreflang Done Right

A multilingual website publishes each language version on its own indexable URL, connected by reciprocal hreflang tags. Three decisions determine whether it works: subdirectory URL structure (site.com/en/), correctly paired hreflang annotations, and human-reviewed translation instead of raw machine output. Research shows 31% of international sites get the hreflang part wrong.

You have probably seen the failure pattern. A company adds a language switcher, runs its pages through an automatic translator, and six months later nobody can explain why the new language version gets zero search traffic. The translation is rarely the whole problem. The technical structure underneath it usually is.

This guide walks through every decision in order: which languages to add, how to structure URLs, how to implement hreflang without breaking it, what translation actually costs, and how to keep the whole thing alive after launch.

Table of Contents

• Multilingual vs. Multiregional
• Choosing Which Languages to Add
• URL Structure: Subdirectory, Subdomain or ccTLD
• How to Implement hreflang Correctly
• The Eight Most Common hreflang Mistakes
• Translation Quality: Machine, AI and Human Editing
• Localization Goes Beyond Translation
• Language Switchers and the Auto-Redirect Trap
• Technical Foundations: CMS and Framework Choice
• What a Multilingual Website Costs
• Language and AI Search Visibility
• Post-Launch Measurement and Maintenance
• Frequently Asked Questions

Multilingual vs. Multiregional

These two terms get used interchangeably, and that confusion is where most broken setups begin. A multilingual website serves the same content in different languages — English and German, for example. A multiregional site serves the same language to different countries with market-specific conditions: different pricing, shipping times, tax rules or legal disclosures.

Most companies should start multilingual and stay there as long as possible. Two languages, one price list, one shipping policy. The moment you introduce country-specific currency, stock levels or regulatory text, complexity multiplies and your multilingual website turns into a multiregional platform with a very different maintenance profile.

Scale matters too. According to W3Techs data, as of June 2026 English accounts for 49.7% of all websites whose content language can be identified, followed by Spanish and German at 6.0%, Japanese at 5.0% and French at 4.6%. Adding English makes you visible — and drops you into the most crowded pool on the web. Plan for your English pages to need more effort than your home-language pages, not less.

Choosing Which Languages to Add

Do not pick languages by intuition. Every multilingual website starts with this question, and you already have three data sources that answer it.

Analytics and Search Console. Look at the country and browser-language breakdown of your existing visitors. If Germany already sends you triple-digit monthly organic sessions, a German version is a testable hypothesis rather than a guess. Search Console's country report shows where you earn impressions but lose clicks — the clearest signal of a language mismatch.

Your commercial reality. Where do your leads, exports or clients actually come from? A manufacturer shipping to the EU needs German long before it needs Japanese. A SaaS company selling to North America may only ever need English.

Your inbox. Which language do inbound enquiries arrive in? Which ones go unanswered because nobody on the team can reply? No analytics tool captures this, and it is often the most accurate indicator you have.

The business case is well documented. CSA Research's long-running consumer studies found that roughly 72% of buyers spend most of their time on sites in their own language, and four in five consumers say they will not buy from a brand without local-language support. Language is not a nice-to-have feature — it is a conversion variable.

One firm rule: do two languages properly rather than five badly. Every additional language multiplies the cost of each price update, each new blog post and each legal revision.

URL Structure: Subdirectory, Subdomain or ccTLD

This is the most expensive decision to reverse in a multilingual website project. Getting it wrong means hundreds of 301 redirects and months of ranking volatility later.

For the large majority of businesses, the answer is the subdirectory. The authority you have accumulated on your main domain flows straight into the new language folder. Buying a fresh domain and starting authority from scratch throws away your single most valuable SEO asset.

Subdomains carry most of the cost of a ccTLD without delivering the local trust signal that makes a ccTLD worth having. Choose one only when there is a genuine technical constraint — for instance, when the new language version has to run on a completely separate platform.

A ccTLD is a serious commitment. If you are going head-to-head with local competitors in Germany, opening an office there and publishing German content continuously, .de makes sense. Otherwise you end up maintaining a second site that quietly goes stale.

Translate your URL slugs too. site.com/de/services/web-design is weaker than site.com/de/leistungen/webdesign for a German audience. Our own site follows exactly this pattern: Turkish content at the root, English under /en/, and every slug written in its own language. Building that mapping at the start is far cheaper than migrating hundreds of URLs later.

How to Implement hreflang Correctly

The hreflang annotation tells search engines which language and region a page targets. On a multilingual website its job is to stop language variants from competing with each other and to make sure the right version surfaces for the right user.

Three implementation methods

Google treats all three as equivalent. Pick one and apply it consistently.

1. HTML link tags in the head — the most common approach, ideal for small and mid-sized sites.
2. HTTP headers — the only option for non-HTML files such as PDFs.
3. XML sitemaps — keeps markup lean on sites with thousands of URLs and centralises management.

A two-language page looks like this in HTML:

<link rel="alternate" hreflang="en" href="https://site.com/en/services/web-design" />
<link rel="alternate" hreflang="de" href="https://site.com/de/leistungen/webdesign" />
<link rel="alternate" hreflang="x-default" href="https://site.com/en/services/web-design" />

That same block must appear on every page in the set. If the English page points to the German page but the German page does not point back, the annotation is discarded.

Language and region codes

Google's documentation defines the format precisely: an ISO 639-1 language code, optionally followed by an ISO 3166-1 Alpha-2 region code. Google also warns that you cannot specify a country code on its own — the first code always represents the language.

• hreflang="en" — all English speakers, regardless of country
• hreflang="en-GB" — English speakers in the United Kingdom
• hreflang="de-AT" — German speakers in Austria
• hreflang="UK" is invalid; the ISO code for the United Kingdom is GB
• hreflang="en_US" is invalid; the separator is a hyphen, not an underscore

Google adds one more caveat worth remembering: it determines a page's language algorithmically, not from your hreflang or lang attribute. A page that claims to be English but still contains German copy cannot be rescued by markup.

When you need x-default

The x-default value defines the fallback page shown when a user's language matches none of your versions. On an English–German site, it decides what a French visitor sees. Add it whenever you run a language gateway page or target an international audience broadly.

The Eight Most Common hreflang Mistakes

An analysis of 18,786 sites published on Search Engine Land found that 31% of sites serving multiple languages carry conflicting hreflang directives. The same study found 16% were missing self-referencing tags and 47.95% were not using x-default at all. In other words, most teams that build a multilingual website implement hreflang incorrectly.

The eight failures you will actually encounter:

1. Missing return links. If the pages do not reference each other, Google ignores the whole set. The rule exists so that a third party cannot unilaterally declare itself an alternate version of your page.
2. No self-referencing tag. Every page must list itself alongside the other language versions.
3. hreflang fighting canonical. If your German page canonicalises to the English URL, the two signals cancel out. Each language version must canonicalise to itself.
4. Relative URLs. hreflang requires absolute URLs, including protocol and domain.
5. Pointing at redirected or broken URLs. A 301 or 404 inside the set breaks it.
6. Invalid codes. UK, EU and UN are not ISO language or region codes and get ignored.
7. Inconsistent sets across pages. Three languages on the homepage and two on inner pages leaves Google unable to resolve the correct set.
8. Correct markup, untranslated content. The most deceptive failure of all: the tag says de, the page still reads in English.

Translation Quality: Machine, AI and Human Editing

Translation is where most multilingual website budgets go, and Google's stance on automated translation has softened in recent years. Machine translation is no longer treated as spam by default; the test is whether the resulting page serves the reader. That said, Google's scaled content abuse policy still lists "automated transformations like translations" as a potential violation when the output provides little or no value. Publishing raw machine output remains risky.

A three-tier workflow handles this efficiently:

Editorial review costs less than most teams assume. According to 2026 industry rate guides, professional human translation runs roughly $0.10 to $0.30 per word, while machine translation post-editing lands between $0.04 and $0.12 per word — typically 30-70% of the full human rate. Reserve the expensive tier for the pages that close deals.

One warning: an outdated language version can do more damage than no version at all. Stale pricing, discontinued services or a dead phone number on your German pages actively misinform the market you are trying to win.

Localization Goes Beyond Translation

Translation moves words. Localization adapts the page to how the target market actually behaves, and a multilingual website needs both.

Research keywords in each language rather than translating them. A literal translation of your primary keyword is frequently a term nobody searches. Pages built on translated keywords can be grammatically flawless and still target zero search volume. Run separate keyword research per market before writing a single line.

The details that get overlooked most often:

• Currency and number formats. Decimal separators, thousand separators and currency position all differ by locale.
• Date and time formats. 07/14/2026 and 14.07.2026 mean different things to different readers.
• Phone numbers and addresses. Include the country dialling code in international format.
• Legal pages. Privacy policies must satisfy the target market's regulations, not only your own.
• Case studies and references. Foreign visitors do not recognise your domestic client logos; lead with project scale and outcomes instead.
• Imagery. Stock photography and colour associations carry different cultural weight across markets.

If you work with a nearshore partner, localization quality is worth checking before the contract is signed. Our guide to what a corporate website should include covers the baseline standards every language version should meet.

Language Switchers and the Auto-Redirect Trap

The single most common design mistake on a multilingual website is forcing visitors to a language version based on their IP address. This backfires in three ways. IP geolocation is unreliable, so travellers and VPN users land in the wrong place. Googlebot generally crawls from US IP addresses, so it may never see your other language versions. And a forced redirect removes the visitor's choice entirely.

The correct pattern keeps the user where they are and offers a dismissible suggestion banner instead: "View this page in German?" That informs the visitor without confusing the crawler.

For the switcher itself:

• Write language names in their own language: "Deutsch", "Français", "Türkçe" — not "German", "French", "Turkish".
• Skip the flags. A flag represents a country, not a language, and picking one flag for English or Spanish is always wrong for someone.
• Switching language must send the visitor to the equivalent page, not to the homepage. If no equivalent exists, use the nearest category page.
• Put the switcher in the header where people expect it. A switcher buried in the footer does not get found.

Technical Foundations: CMS and Framework Choice

Your platform choice determines the long-term maintenance cost of a multilingual website more than any other technical decision.

WordPress plugins. Adding a second language to an existing WordPress site is fastest through a multilingual plugin. Setup is straightforward, but each plugin adds database overhead and can conflict with theme updates. Performance degrades noticeably as page counts grow — measure it against your Core Web Vitals baseline before and after.

Framework-based custom builds. Modern frameworks such as Next.js treat internationalized routing as a core feature. Locales, translated URLs and hreflang tags are configured in one place, and adding a language becomes a matter of adding a translation file. We use this approach on our own site: Turkish at the root, English under /en/, localized slugs per locale, and hreflang emitted only for pages that genuinely exist in that language.

Headless CMS. If a large content team feeds both a website and a mobile app from the same source, headless architecture pays off. Setup costs more, but past three languages it scales better than anything else.

Whichever route you take, plan for partial translation from day one. Having 200 posts in your main language and 40 in the second is normal. The second-language blog index should list only translated posts, and hreflang should be emitted only where a translation actually exists. Publishing empty "translation coming soon" pages just accumulates thin content in the index.

If your existing platform is old and slow, rebuilding may cost less than retrofitting; we compare the two paths in our website redesign guide.

What a Multilingual Website Costs

The budget for a multilingual website breaks into three parts: technical implementation, translation and content, and ongoing maintenance. Typical 2026 ranges:

Those figures reflect Western agency rates. Working with an established Turkish agency typically delivers the same scope for 40-70% less, since senior developer rates run $30-45 per hour against $90-150 in Western Europe. A two-language corporate build quoted at $30,000 in London commonly lands between $8,000 and $18,000 in Istanbul with identical deliverables.

To size your own project, run the numbers through our website cost calculator — it accounts for page count, language count and integration scope. For online stores, our e-commerce development cost breakdown goes deeper on the per-module figures.

The line item teams forget is maintenance. Every new service page, price change and blog post doubles in effort. If you cannot absorb that, apply the second language to your 8-10 highest-value pages rather than the entire site.

Language and AI Search Visibility

Search no longer ends at the results page. Research reported by Search Engine Land in August 2026 shows that AI assistants handle language versions very differently. ChatGPT's retrieval bot fetches English versions 65-79% of the time on multilingual sites — roughly 2.6 times more often than Googlebot. Copilot is close to neutral, while Google's AI systems lean slightly toward local-language pages.

The measured effect of adding an /en/ folder was substantial: a 122% lift in ChatGPT citations, 52% for Copilot and 28% for Google's AI surfaces. The researchers attach an important caveat — outdated English content is riskier than having none, because assistants will happily cite stale information.

The practical implication is that a multilingual website now drives visibility in AI assistants as well as in classic search results. Keeping language versions current has shifted from a technical SEO task to a content operations commitment. Our SEO service covers this class of auditing.

Post-Launch Measurement and Maintenance

Launching a multilingual website is where the real work starts. Run this checklist across the first 90 days.

Week one. Confirm each language version is indexed independently using a site: query. Add every version to Search Console and submit the sitemaps. Inspect the hreflang tags in the rendered source of a sample page per language.

Month one. Check the coverage report for "alternate page" or "duplicate content" warnings. Open the country performance report — are impressions arriving from the target market? Measure page speed separately per language, since translation plugins usually surface their performance cost here first.

Every quarter. Audit hreflang sets with a crawler. Verify that newly published pages were added to the annotation set. List untranslated pages and prioritise them. For a fast overall health check, run our free site analysis tool.

Avoid one measurement trap: do not benchmark language versions against each other. English pages will rarely match your home-language traffic, because English competition is far denser. The right metric is qualified demand per language, not raw sessions. A version drawing 200 monthly visitors and three serious enquiries beats one drawing 2,000 visitors and none.

Frequently Asked Questions

Does a multilingual website cause duplicate content penalties?

No. Versions in genuinely different languages are not duplicate content. The risk appears with regional variants of the same language, such as en-US and en-GB carrying near-identical copy. Correct hreflang annotation resolves that ambiguity.

Is a Google Translate widget enough?

No. Browser-side translation widgets do not create separate URLs, so search engines cannot index the translated versions. If you want search traffic, each language needs its own permanent, crawlable address and its own stored content.

Will a multilingual website work without hreflang?

Partially, but not reliably. Google can identify language versions algorithmically, but it has to guess which version to serve to which user. hreflang removes that guesswork and substantially reduces the chance of the wrong language version ranking.

Should I choose a subdirectory or a subdomain?

Subdirectory, in almost every case. It inherits the root domain's authority in full, adds no hosting cost and runs from a single codebase. A subdomain only makes sense when the language version must run on a technically separate platform.

How long until new language pages start ranking?

Indexing usually takes one to four weeks. Meaningful rankings and traffic take three to six months. The timeline depends on keyword competitiveness, your domain authority and the depth of the content you publish.

Does AI translation hurt SEO?

Publishing unedited output can. Google's scaled content abuse policy treats automated translations that provide little value to users as a violation. Using AI translation as a first draft and putting it through an editorial pass is both safe and economical.

How many languages should I add?

Start with two and measure. Each language you add to a multilingual website multiplies content production and maintenance. Prove that your second language generates real demand before adding a third, or you will accumulate neglected pages.

Should I retrofit my existing site or rebuild it?

If your platform is under three or four years old and performs well, adding a language to it is cheaper. If the codebase is dated, slow or unmaintained, moving to a site architected for multiple languages from the start usually costs less over three years.

A multilingual website is an infrastructure decision, not a translation task. It touches URL architecture, technical SEO, content operations and ongoing maintenance — and the most common mistake is running those steps in the wrong order. Translation gets commissioned first, and the technical structure is patched on afterwards. The correct sequence is URL structure, then hreflang, then content operations.

A multilingual website with two languages executed properly will out-earn five languages executed halfway, every time. Define the market you actually want, research keywords natively in that market's language, and commit to keeping those pages current. To scope your project, take a look at our web design service or get in touch — we review your current site's readiness for a multilingual build at no charge.