AI engines bypass language barriers in a way traditional search never did. Perplexity and Gemini can pull facts from an English-language source and synthesise a fluent answer in French, German, or Portuguese without the user ever visiting an English page. ChatGPT Search launched multilingual web retrieval on 31 October 2024. Perplexity has always done live multilingual retrieval. Claude added web search with multilingual capability in March 2025.
This creates two distinct problems for brands with non-English markets. First, your English content may be cited in response to non-English queries — earning you citation credit without a localised page. Second, a competitor's English content may be cited instead of your localised page because AI engines weight their English-language content more heavily than your translated page for the same query. Both outcomes are possible simultaneously, and both require different fixes.
Alhena AI's analysis of cross-market citation behaviour found a 431% difference in AI citation eligibility between sites that only translate content and sites that properly localise — meaning the content addresses market-specific regulations, local examples, region-specific data, and cultural context, not just language. Translation is table stakes. Localisation is the citation differentiator.
What Is the Difference Between Translation AEO and Localisation AEO?
A translated page contains the same information as the source page, expressed in a different language. A localised page contains information specifically relevant to the target market — local regulations, local examples, local data sources, local search behaviour — expressed in the target language.
AI engines evaluate content for market-specific relevance in addition to language matching. A page in German that references US statistics, cites US regulatory bodies, and uses US examples will lose to an English-language page that contains the same statistics but is otherwise more authoritative — because the German page's market-specific signals are weak. ALM Corp's 2026 international SEO analysis documented this clearly: organisations that continue to rely on translation alone find their international content increasingly invisible in AI-generated responses, as systems default to the most confident global representation regardless of geographic appropriateness.
The practical difference for your content programme: translation requires linguistic conversion. Localisation requires market research, local source citation, local regulatory reference, and local entity signals — the same work that earns citations in English, replicated for each target market.
Does Hreflang Still Matter for AI Citations?
Hreflang matters significantly, but for a different reason than in traditional international SEO. In traditional SEO, hreflang routes the right language version to the right user by signalling language and regional targeting to Google's ranking algorithm. In AI search, hreflang tells AI retrieval systems which locale variant to select when multiple language versions of the same content exist — preventing the AI from serving your English page in response to a French-language query when a French-language version exists.
Ahrefs' 15,000-prompt study measuring Google AI Overview citations found that 76% of AI Overview citations come from pages already ranking in Google's top 10. Strong multilingual ranking — which hreflang implementation directly supports — inherits the AI Overview citation advantage at the locale level. A French page that ranks top 5 for a French query because hreflang correctly signals its locale is far more likely to be cited in a French AI Overview response than the same page without hreflang, which may not rank for the French query at all.
Three hreflang implementation requirements that are specifically important for AI citation in 2026:
Server-rendered, not client-rendered. If your CMS generates hreflang tags via JavaScript after page load, AI crawlers that do not execute JavaScript — which includes most of them — see no hreflang tags. The AI crawlers guide covers the server-rendering requirement in full. For multilingual sites, confirm hreflang tags appear in the initial HTML response using curl -A "Googlebot" https://yourpage.com | grep "hreflang".
inLanguage field in Article schema, matching hreflang exactly. Hreflang declares language at the HTML tag level. Article schema's inLanguage field declares it at the content level. Both should use identical ISO 639-1 language codes. A page with hreflang="fr" and "inLanguage": "fr" sends a consistent double signal. A page with hreflang="fr" and no inLanguage declaration sends a weaker signal that AI retrieval systems may not correctly resolve.
Self-referencing hreflang on every page variant. Hreflang requires each language variant to declare itself and every other variant. A French page missing its own self-referencing hreflang — hreflang="fr" pointing to itself — has an implementation error that causes the entire hreflang cluster to be treated as unreliable by both Google's ranking algorithm and AI retrieval systems. The schema errors guide covers hreflang as one of the ten most common AEO infrastructure failures.
What Are the Market-Specific Signals AI Engines Need?
Beyond language and hreflang, AI engines evaluating cross-market content look for four market-specific signals that translation cannot provide:
Local regulatory citations. A page targeting the German market that cites the Bundesanstalt für Finanzdienstleistungsaufsicht (BaFin) for a financial topic, or the Bundesnetzagentur for a telecom topic, signals specifically German market relevance in a way that citing generic EU regulations does not. AI engines retrieving content for a German-language query on a regulated topic weight local regulatory authority references more heavily than generic EU or international equivalents.
Local data sources. Statistics from local statistical offices, local industry associations, or local market research firms signal market-specific research rather than globalised content. A page about UK ecommerce market size citing ONS data earns stronger UK market signals than the same page citing US or global ecommerce statistics. Use local sources for local market claims wherever they are available.
Local entity signals in Organisation schema. Your Organisation schema should include locale-specific fields: areaServed with the specific country or region, address with a local address if applicable, and telephone with the local format. Person schema for local authors should include sameAs links to their profiles on locally relevant professional platforms — XING for German markets, LinkedIn is global but should reference local professional associations where applicable.
Local example and case study references. A how-to guide that uses a German company as the example, references German market conditions, and links to a German case study earns stronger local market signals than the same guide with globalised examples. AI engines assess market specificity through the named entities in the content — local brand names, local market conditions, local regulatory examples — not just through language.
Which AI Engines Handle Multilingual Content Differently?
Engine-specific multilingual behaviour varies enough to require separate tracking for primary international markets:
Google AI Overviews and AI Mode. Run on Google's multilingual index, which has the deepest coverage of non-English content. Hreflang implementation has the strongest direct effect on Google's AI surfaces because Google's ranking algorithm — which AI Overviews pull from — uses hreflang for locale resolution. A properly localised page with correct hreflang consistently outperforms an English-language page for the same query in the target language market.
Perplexity. Does live multilingual retrieval through its own crawler. Perplexity responds to language-specific queries by retrieving content in that language first, then falling back to English-language content if localised content is weaker or absent. The fallback behaviour means English-language content with strong topic coverage can still earn Perplexity citations in non-English markets — but dedicated localised pages outperform the fallback when they are well-structured and properly signalled.
ChatGPT Search. Uses Bing's multilingual index for retrieval. Bing Webmaster Tools allows explicit locale targeting and sitemap submission per language — an additional locale signal that Google does not have an equivalent for. For ChatGPT citations in international markets, Bing Webmaster Tools setup with locale-specific sitemaps is a specific technical step beyond what Google-focused international SEO requires.
Claude. Uses Brave Search for web retrieval. Brave's multilingual crawler coverage is less comprehensive than Google's or Bing's for many non-English markets. For Claude citations in non-English markets, building a strong training-data presence — entity coverage in local Wikipedia editions, coverage in locally respected publications, Wikidata entries in the target language — matters more than technical hreflang implementation alone.
How Do You Track Multilingual Citation Rate?
Standard AEO tracking runs prompts in English across the five major AI engines. International AEO tracking requires running the equivalent prompts in each target language and market, then comparing results across markets. A brand that earns citations for "best AEO tool" in English but earns no citations for "meilleur outil AEO" in French has a specific French market gap that English-language prompt tracking cannot surface.
The tracking discipline is the same as the English-language approach described in the AEO prompt tracking guide, applied to each target locale: identify 10-15 prompts in the target language representing queries your buyers run in that market, run them weekly across ChatGPT, Perplexity, and Google AI Mode, and record citation presence, competing sources, and citation tone.
The SI-UK international case study in this series — how 35 global domains were managed for AI citation consistency — covers the operational challenge of maintaining citation quality across many markets simultaneously. The core principle: each market requires its own prompt matrix, its own citation tracking, and its own freshness maintenance cadence. An international AEO programme cannot be managed from a single English-language dashboard.
The NotioncCue Prompt Tracker supports non-English prompt tracking across all five engines — meaning you can run your French, German, Spanish, or Japanese prompt matrix alongside your English one, with results in the same dashboard. This is the infrastructure gap that most AEO tracking setups leave unfilled: English-only tracking makes international performance invisible until a competitor alerts you with a lost-deal conversation.
Multilingual AEO assessment starts with the same audit as any other market: confirm AI crawlers can access your localised pages (server-rendered content, no JS-gated hreflang), confirm schema declares language correctly, and run five prompts in each target language through Perplexity and Google AI Mode. Those ten minutes of manual checking surface whether localised pages are in the citation pool at all before you invest in localisation content quality improvements. Run the AEO audit checklist in each target language market as a separate audit — the technical checks are identical, but the language-specific elements need to be verified independently.
Frequently Asked Questions
Should non-English markets have separate domains, subdomains, or subdirectories?
For AI citation purposes, subdirectory (yourdomain.com/fr/) is the strongest architecture because it concentrates domain authority into a single entity. Separate country-code domains (yourdomain.fr) split authority across multiple entities and require independent entity-building in each market. Subdomains (fr.yourdomain.com) are intermediate — better than separate domains, weaker than subdirectories. Subdirectory architecture also makes it easier to link internal AEO pages across markets from a single domain, strengthening the topical cluster signals that support all language versions.
How do you handle machine-translated content for AEO?
Machine translation is insufficient for AI citation-worthy content in most markets. AI engines can detect low-quality translation through entity inconsistency, unnatural phrasing, and the absence of local signals. Pages that were machine-translated and not reviewed by a native speaker with market knowledge typically have weaker citation rates than original English pages for the same query, even in the target language market. Budget for human translation and localisation review for any page you expect to earn AI citations — the machine-translated version rarely passes the local relevance bar that AI engines apply to non-English content.
Do all AI engines treat hreflang equally for citation decisions?
No. Google AI surfaces use hreflang most directly because they pull from Google's hreflang-aware index. Bing/ChatGPT use Bing's locale targeting tools. Perplexity and Claude use their own crawlers which respect hreflang but apply it less precisely than Google. Across all engines, correctly implemented hreflang reduces the chance of an English-language page being served for a non-English query — but local content quality and market-specific entity signals ultimately determine citation selection more than the technical hreflang tag alone.