Hreflang for DACH: Why AI Platforms Barely Use the Tags

Hreflang is considered the standard tool for multilingual and multiregional websites. In the DACH region, it is supposed to ensure that Austrian users see the de-AT version, German users the de-DE version, and Swiss users the de-CH version. This logic works in traditional Google Search but runs into a new limit: AI platforms like ChatGPT, Perplexity, Copilot, and Claude largely don't evaluate hreflang annotations. This article explains how to set up hreflang correctly for DACH, why AI crawlers ignore the signals, and how you can still make de-AT content visible in AI answers and traditional search.

Why hreflang matters for DACH and where the new limit lies

The DACH region is a special case: three countries, one language, but different markets, prices, legal frameworks, and in part different terminology. This is exactly where hreflang comes in. It tells search engines which URL is intended for which language-region combination, so that Google serves the de-AT page in Austria and the de-DE page in Germany.

The Austrian market is almost entirely online. At the start of 2025, 8.69 million people in Austria used the internet, a penetration rate of 95.3% (Austria). If you address this market with your own de-AT localization, you have a legitimate interest in the right variant being served, in Google and increasingly in AI systems as well.

That is exactly where the playing field shifts. Hreflang no longer works as a hard constraint once content is delivered through AI-mediated retrieval. The spread of this AI search is measurable: in Germany, 50% of internet users at least occasionally use AI chats instead of traditional search engines (Germany), and 42% of these AI searchers have already received false or fabricated information from the AI (Germany). Wrong language versions are part of this problem: when a system returns de-DE content for an Austrian query, prices, VAT rates, or legal terms are often wrong.

Hreflang basics for DACH: correct annotation and the reciprocity requirement

A clean hreflang implementation for the DACH region follows clear rules. The language-region codes combine ISO 639-1 (language) with ISO 3166-1 alpha-2 (region).

• Correct language-region codes: de-DE for Germany, de-AT for Austria, de-CH for Switzerland. Plain de without a region targets all German-speaking users regardless of country and can make sense as an additional generic variant.
• x-default as fallback: The x-default value marks the page that is served when no other variant fits, for example for German-speaking users outside DACH or for a language selection page. x-default is the most frequently missing component.
• Self-referencing: Every page must also reference itself via hreflang. The de-AT page therefore lists de-DE and de-CH, plus itself as de-AT.
• Reciprocal references (return-link requirement): Hreflang is bidirectional. If the de-DE page points to the de-AT page, the de-AT page must point back to the de-DE page. If the return link is missing, Google ignores the annotation.

A DACH minimum therefore consists of de-DE, de-AT, de-CH, and x-default, each with complete, reciprocal, and self-referencing tags. Despite these clear rules, hreflang is error-prone. An international analysis of 374,756 domains found that 67% have at least one hreflang implementation issue (international). The most common errors: 56.3% are missing x-default, 18% lack self-referencing tags, and on 16.9% hreflang tags point to redirected or broken pages (international).

Why AI crawlers largely ignore hreflang

The decisive difference between Google and AI platforms lies in processing. Google renders pages, executes JavaScript, and selects a locale based on hreflang, location, and language settings. AI crawlers don't.

An analysis of crawler traffic across the Vercel network shows that none of the major AI crawlers renders JavaScript (international). GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot fetch content but don't execute it. GPTBot alone generated 569 million requests across the network in one month (international). Some crawlers do download JavaScript files: for ChatGPT, JS files account for 11.50% of requests, for Claude 23.84% (international), but they lack the rendering environment to execute them. A broader study of 23 major AI crawlers reaches the same result: 69% of AI crawlers cannot execute JavaScript (international), explicitly including GPTBot and ClaudeBot.

This has three consequences for hreflang:

• No JS rendering: Hreflang signals that only reach the DOM through client-side JavaScript are invisible to AI crawlers. They only see the initial HTML.
• No locale selection: AI platforms don't interpret hreflang as a selection mechanism. They process the URL they just fetched without actively looking for the de-AT variant of a de-DE URL.
• Hint instead of constraint: What is a weak hint at Google becomes practically ineffective in AI retrieval. The platform delivers the version it has indexed, not necessarily the regionally appropriate one.

The upside matters: content in the initial HTML, including embedded JSON data or server-side rendered components, can be indexed because AI models also interpret non-HTML content. So the lever is not hreflang but writing the right language signal directly into the delivered HTML.

content-language for Bing and Copilot

A common misconception is that hreflang works the same way in every search engine. By its own account, Bing treats hreflang only as a weak signal (international) and detects language mainly through the content-language signal, links, and where visitors come from. In the DACH region, this is no side issue: in Austria, Bing ranks a clear second with a 9.01% market share, behind Google at 81.87% (Austria, May 2026). Since Microsoft Copilot is built on the Bing index, the content-language signal also carries over into this AI system.

• Correct ll-cc format: content-language also uses the language-region scheme, for example de-AT. Multiple values are separated by commas.
• Consistency across all levels: Content-Language in the HTTP header, the lang attribute on the html element, and supplementary meta tags should carry the same value so that Bing doesn't receive a contradictory language signal.
• Don't forget the lang attribute: The lang="de-AT" attribute on the html tag is the simplest and most reliable on-page language signal. It sits in the initial HTML and is therefore visible to crawlers without JS rendering.

In practice, this means: if you serve DACH variants, maintain hreflang for Google and content-language plus a correct lang attribute for Bing, Copilot, and AI crawlers together.

Hreflang as a hint, not a directive

Even at Google, hreflang is not a command. Google explicitly describes hreflang tags as hints, not directives (international). Google's John Müller makes clear that hreflang does not guarantee indexing and that with nearly identical variants, Google often picks one of them as the canonical.

This is particularly relevant for DACH. The de-DE, de-AT, and de-CH variants are often almost identical in language. If two variants have the same content, Google typically picks one of the same-language versions as canonical because its systems simplify things (international). As a result, the de-AT page can drop out of the index even though the hreflang annotation is technically correct.

Genuine differentiation helps against this. The more the de-AT variant differs in content, the more likely Google is to treat it as a standalone page:

• Standalone content: Austrian prices in euros with the correct VAT rate, local contacts, Austrian legal terms, and references from the Austrian market.
• Local terms and spellings: Where Austrian German differs, the text should reflect that. This sharpens relevance and distinction at the same time.
• Separate URL per variant: A clearly separate URL per variant is the prerequisite for a variant to be indexed at all and referenced in AI systems.

Implementation methods compared and their AI visibility

Hreflang can be implemented in three ways. For AI crawlers, what matters is whether the signal is in the initial HTML.

• HTML link elements in the head: The hreflang annotations sit as link tags in the head section. If they are rendered server-side, they are part of the initial HTML and therefore visible to all crawlers. For DACH, this method is the most transparent.
• HTTP header: Hreflang can be delivered via the Link HTTP header, mainly for non-HTML documents like PDFs. The header is visible regardless of the crawler but harder to maintain in practice.
• XML sitemap: Hreflang can be mapped completely in the XML sitemap. This keeps the HTML code lean but has a drawback for AI retrieval: the signal is not in the individual page. An AI crawler that fetches a URL directly without evaluating the sitemap doesn't see the annotation.

For traditional Google Search, all three methods are equivalent. You can combine them, but according to Google this brings no advantage in search and makes maintenance more work, because every change has to be carried over consistently in several places. One method, used consistently, is the better choice. For AI visibility, the server-side rendered head variant is the most reliable because it ties the language signal directly to the delivered document. What remains decisive, however, are the lang attribute and clear on-page language signals, since these work even without hreflang interpretation.

Hreflang in Next.js and headless architectures

In modern Next.js and headless setups, the rendering strategy determines whether language signals reach AI crawlers at all.

• Server-side rendering instead of client-side: Locale signals must be in the initial HTML, not generated by client-side JavaScript after hydration. Since AI crawlers don't render JS, any signal that is only set on the client side gets lost.
• alternates.languages in the metadata: Next.js lets you output hreflang server-side via the metadata object with alternates.languages. This reliably places the annotations in the head of the initial HTML.
• Set the lang attribute server-side: The lang="de-AT" on the html element should be set by the server per locale. It is the most stable language signal for crawlers without rendering.
• Clean head structure: Important metadata must sit at the top of the head. An international survey found that 10.1% of desktop pages have invalid elements in the head (international). If hreflang, canonical, or meta robots tags come after an invalid element, they are moved into the body and thus become ineffective.

GEO and AEO consequences for de-AT content

If AI platforms don't evaluate hreflang as a selection mechanism, the task shifts from locale control to pure visibility. The goal is for de-AT content to be recognizable as Austrian content in ChatGPT, Perplexity, and Copilot, without hreflang acting as a filter.

• Server-side rendering as a must: Only what is in the initial HTML gets processed by AI crawlers. The de-AT content must be delivered entirely server-side.
• Explicit on-page language signals: Instead of relying on hreflang, the Austrian connection should be in the text itself: Austria, Austrian locations, euro prices with the Austrian VAT rate, Austrian legal and technical terms. This makes the localization directly readable for language models.
• Separate, stable URLs: Every variant needs a unique, permanently accessible URL so that it can be cited and linked.
• Structured, sourced content: Clear definitions, lists, and sourced statements increase the likelihood of being named as a source in generative answers.

For this, the subfolder approach with example.com/at/ is usually superior to the separate ccTLD approach because it consolidates domain authority while providing a clear, indexable URL per region. This preserves the localization without spreading visibility across several weak domains.

Hreflang in combination with canonicals, redirects, and sitemaps

Hreflang never stands alone. Conflicts with canonical tags, redirects, and sitemaps are among the most common sources of errors.

• Don't let canonical and hreflang contradict each other: Every variant should canonicalize to itself. A canonical pointing from the de-AT page to the de-DE page cancels out the hreflang logic and leads to consolidation.
• No references to redirects: Hreflang must point to the final URL that returns status 200, not to an address redirected via 301. References to redirects or broken pages are a documented top error.
• Consistency with the XML sitemap: URLs in the sitemap, in the canonical, and in the hreflang annotations must be identical, including protocol, trailing slash, and capitalization.

The scale of these problems is considerable. An international study of 18,786 websites with hreflang code found contradictory hreflang directives on 31.02% and a missing x-default on 47.95% (international). Such conflicts lead Google to ignore the signals and decide on its own.

Monitoring, validation, and measurement

Hreflang needs ongoing control, precisely because errors often stay invisible. A combination of tooling and log file analysis covers the most important problems.

• Google Search Console: There is no longer a hreflang error report there, since Google retired the International Targeting report. The Page indexing report shows which URLs were excluded as duplicates, and the URL Inspection tool shows the index status and the Google-selected canonical URL for each DACH variant. You find the hreflang errors themselves with a crawler.
• Crawlers like Screaming Frog: A technical crawl uncovers missing self-references, missing x-default, non-reciprocal references, and hreflang tags that point to redirects or non-canonical pages.
• Log file analysis of AI bot access: Server logs show whether and how GPTBot, ClaudeBot, and PerplexityBot fetch the individual variants. This lets you check whether the initial HTML is delivered with the correct language signals.
• Content render check: Fetching the raw HTML without executing JavaScript shows whether the lang attribute, hreflang in the head, and the local connection are really in the initial document.

For context on adoption: according to the international Web Almanac survey from 2022, 9.6% of pages on desktop used hreflang tags, with de among the four most common language values (international). So hreflang is widespread, but as the error rates show, rarely error-free.

Further reading and conclusion

Hreflang remains useful for traditional Google Search in the DACH region if it is implemented cleanly: de-DE, de-AT, de-CH, and x-default with complete, reciprocal, and self-referencing tags. The key insight is that hreflang is only a hint even at Google and becomes practically ineffective in AI-mediated retrieval, because AI crawlers don't select a locale and don't see signals set on the client side at all.

The consequence for companies serving the Austrian market is clear: language signals belong in the server-side rendered HTML, the Austrian connection belongs in the visible text, and every variant needs its own stable URL. In addition, content-language and the lang attribute should be maintained for Bing and Copilot. Sensible next steps are a render check of the initial HTML, a hreflang audit with a crawler, and a log file analysis of AI bot access to check whether the right variant reaches the AI systems at all.