International SEO: hreflang, country targeting, multi-language

A comprehensive installation and audit reference for international SEO: deciding which markets to enter, which domain architecture to deploy, how deeply to localize, which regional search engines to…

The Canonical 2026 Reference for Strategic International Search Optimization Across Market Selection, Domain Architecture, Content Localization, Regional Engines, and AI Citation Coverage

A comprehensive installation and audit reference for international SEO: deciding which markets to enter, which domain architecture to deploy, how deeply to localize, which regional search engines to chase, and which AI engines actually reach each target country. Distinct from framework-hreflang.md, the operational tag and URL cluster reference, this document is the strategic and editorial reference. Decisions here precede tag generation, drive market selection, and define which countries get content investment versus which are listed as out of scope.

Cross stack implementation note: code samples are plain HTML and bash. For React, Vue, Svelte, Next.js, Nuxt, SvelteKit, Astro, Hugo, 11ty, Remix, WordPress, Shopify, and Webflow equivalents, see framework-cross-stack-implementation.md. For pure client rendered SPAs (no SSR/SSG) see framework-react.md. For per market schema patterns see framework-schema.md.

1. Document Purpose

International SEO in 2026 encompasses three intersecting decisions, and AI engines have layered a fourth.

The first decision is market selection. Which countries justify investment given total addressable market, competitive density, cost of entry, regulatory burden, and language adjacency. SimilarWeb traffic share and Statista market sizing are the two most commonly cited TAM inputs.

The second decision is domain architecture: ccTLD inventory, subdomain country prefix, subdirectory locale path, or hybrid. Each optimizes for a different signal stack and operational profile. The decision is rarely reversible without 12 to 24 months of migration cost.

The third decision is content localization depth: translation, partial localization, full localization, or independent in market content production. Search Engine Land's 2024 analysis of 1,200 multilingual sites found localized content earned 3.4 times the organic traffic of straight translated content over an 18 month window.

The fourth decision, new in 2026, is language model coverage. Google AI Overview reached 113 countries by February 2026. Perplexity is available in 96 countries. ChatGPT Search reaches 188 countries with restrictions in China, Russia, Iran, North Korea, Cuba, Syria, and parts of Belarus. Claude with web search reaches 105 countries. Gemini grounding via Search reaches 174 countries. A country with strong Google reach may be unreachable through ChatGPT entirely. International SEO in 2026 must account for both surfaces independently.

This framework specifies the strategic, editorial, regulatory, and infrastructure decisions. Mechanical tag generation lives in framework-hreflang.md. Per engine citation tradeoffs live in framework-multiengine-tradeoffs.md. Per country E-E-A-T lives in framework-eeat.md. Per market schema variants live in framework-schema.md.

1.1 Three Operating Modes

Mode A, Install. Build the international strategy from a single market starting point. Decide markets, architecture, localization depth, AI engine coverage, and per market budget. Follow Sections 2 through 13 in order.

Mode B, Audit. Evaluate an existing international footprint. Skip to Section 14.

Mode C, Hybrid. Audit existing markets, install for new markets being added. The most common engagement pattern when a US first business is adding Canada, UK, Australia, or the EU.

1.2 Conflict Resolution Rules

1.3 Relationship to Neighboring Frameworks

This framework is the strategic header for international work. framework-hreflang.md is the operational tag and URL cluster reference. framework-localseo.md handles within country local SEO. framework-multiengine-tradeoffs.md handles AI engine prioritization at the country level. framework-aicitations.md, framework-aioverviews.md, framework-searchgpt.md, and framework-perplexityspaces.md handle per engine signals. framework-technicalseo.md handles non locale specific technical foundation. framework-schema.md handles per market schema variants. framework-contentbriefs.md handles per market brief generation. framework-brandvoice.md handles per market voice. framework-eeat.md handles per country trust signals.

2. Client Variables Intake

The intake captures the market portfolio, domain inventory, content localization state, regulatory posture, and AI engine reach assumptions. Stored at /var/www/sites/[domain]/audit/international/intake.yaml.

# INTERNATIONAL SEO CLIENT VARIABLES

business_name: ""
primary_domain: ""
hq_country: ""                          # ISO 3166-1 Alpha 2
hq_locale: ""                           # e.g. en-US
business_model: ""                      # b2b, b2c, saas, local_service

# Target market portfolio
current_markets: []
target_markets_year_1: []
target_markets_year_2_3: []
deferred_markets: []                    # explicitly out of scope with reason
language_country_matrix: []             # ["en-US","en-GB","es-ES"]
primary_market: ""
fallback_market: ""                     # x-default candidate

# TAM inputs
similarweb_traffic_share_pulled: false
statista_market_size_pulled: false
ahrefs_keyword_volume_per_country: false
competitor_density_per_country: false
regulatory_burden_assessed: false

# Domain architecture
url_structure: ""                       # cctld, subdirectory, subdomain, mixed
cctld_inventory: []
subdirectory_pattern: ""                # "/en-us/", "/de/"
subdomain_pattern: ""                   # "uk.example.com"

# Content localization
locales_with_full_translation: []
locales_with_localized_content: []
content_identity_us_vs_gb: ""           # identical, spelling_only, fully_localized

# Hreflang (see framework-hreflang.md)
hreflang_present: false
hreflang_method: ""                     # html_head, http_header, xml_sitemap
x_default_present: false

# Regional search engines
baidu_target: false                     # China; requires ICP
yandex_target: false                    # Russia
naver_target: false                     # South Korea
yahoo_japan_target: false               # Japan
icp_license_obtained: false

# AI engine reach (2026)
google_ai_overview_target_countries: []
perplexity_target_countries: []
chatgpt_search_target_countries: []
claude_target_countries: []
gemini_target_countries: []

# Currency, units, date format
currency_per_market: {}                 # {"US":"USD","GB":"GBP","DE":"EUR"}
unit_system_per_market: {}              # imperial, metric
date_format_per_market: {}              # MDY, DMY, ISO

# Regulatory posture
gdpr_compliance: false  # EU and UK
ccpa_compliance: false  # California
lgpd_compliance: false  # Brazil
pipl_compliance: false  # China
appi_compliance: false  # Japan

# Infrastructure (Bubbles hosted)
origin_server_location: "Arkansas USA"
regional_replica_servers: []
nginx_geoip_enabled: false
geoip_database_path: ""                 # /etc/nginx/GeoIP2/GeoLite2-Country.mmdb
postgres_locale_dimension: false

The intake is gating. Sections 3 through 14 cannot run cleanly without it.

3. Market Selection Methodology

Market selection is the most consequential decision because it commits a budget envelope, translation pipeline, regulatory exposure, and operational team profile per country. The reversibility curve is shallow. Once a country is launched, the cost of unlaunching includes 301 redirect maintenance for 24 to 36 months, brand goodwill damage, and customer support cleanup. This section produces a defensible market ranking before any tag or URL work begins.

3.1 The Four Input Sources for TAM

SimilarWeb traffic share by country. Exposes the percentage of category demand sitting in each country. SimilarWeb's country data accuracy is highest in the top 20 markets and degrades below the top 50 by traffic volume. For markets below the top 50, supplement with Statista.

Statista market sizing. Provides revenue and unit volume estimates per country in absolute terms. Cross referenced against SimilarWeb's traffic share, Statista produces a per country revenue per traffic ratio. Germany typically shows 1.4 to 1.8 times the per traffic revenue of the United States in B2B SaaS due to higher average contract values; Japan shows 0.6 to 0.8 in B2C ecommerce due to entrenched domestic competition.

Ahrefs or Semrush keyword volume per country. Both expose per country search volume for the priority keyword set. Run the top 50 priority keywords through each tool's country dimension. Highest aggregate volume is not always highest revenue; volume must be cross checked against intent quality and conversion rate by country.

Competitor density per country. Run the priority keyword set through Ahrefs Site Explorer with country filter and count competing pages with domain rating above 40 in the top 10. High volume with low density signals opportunity. High volume with high density requires a 24 to 36 month investment runway.

3.2 The Cost of Entry Matrix

Each candidate market carries a fixed cost of entry independent of marginal content per locale cost.

The matrix produces a defensible cost floor from 12,000 USD for low regulatory low translation markets to 150,000 USD plus for markets requiring full localization, regulatory work, and in country team. Compare against the 18 month revenue forecast from SimilarWeb plus Statista TAM. Payoff under 18 months: first wave. 18 to 36 months: second wave. Above 36 months: deferred unless strategic.

3.3 The Five Country Ladder Pattern

A defensible launch sequence for a US first business adding international markets.

Rung 1, Canada. en-CA hreflang variant of en-US with minimal localization (CAD currency, fr-CA for Quebec required by Charter of the French Language Bill 96 since June 1, 2022). Lowest cost, fastest payoff.

Rung 2, United Kingdom. English with spelling and vocabulary differences, GBP, UK GDPR (functionally identical to EU GDPR). Mid cost.

Rung 3, Australia. English with localization (AUD, metric units, regional vocabulary), Australian Privacy Principles (lighter than GDPR). Often paired with New Zealand at en-AU plus en-NZ.

Rung 4, EU member state. Germany, France, or the Netherlands typically. Full translation, EU GDPR, VAT handling for ecommerce, payment processor regional coverage. Mid to high cost, high payoff in B2B SaaS.

Rung 5, Japan, Mexico, or Brazil. First non English non European market. Cultural distance increases, search engine landscape shifts (Yahoo Japan in JP, Google in MX and BR), payment processor work intensifies. High cost, high opportunity for businesses with proven product market fit.

3.4 The Deferred Markets Discipline

The intake captures deferred markets explicitly with reason. A common failure mode is allowing "we will get to it" to substitute for explicit deferral. The deferred list typically includes China (PIPL, ICP license, content moderation), Russia (sanctions, distinct Yandex ecosystem), Iran (sanctions), India (English coverage adequate without separate Hindi or regional language investment at most B2B scales), and the MENA cluster outside the Gulf Cooperation Council states.

3.5 The Market Selection Output

The deliverable is the per market scorecard at /var/www/sites/[domain]/audit/international/market_scorecard.yaml:

markets:
  US: { tam: "$10B+", cost_of_entry: "home", payoff: 0,  decision: active }
  GB: { tam: "$1B-$5B", cost_of_entry: "28K USD", payoff: 9,  decision: active_y1 }
  CA: { tam: "$500M-$1B", cost_of_entry: "18K USD", payoff: 6,  decision: active_y1 }
  DE: { tam: "$1B-$5B", cost_of_entry: "65K USD", payoff: 18, decision: active_y2 }
  CN: { tam: "$5B+", cost_of_entry: "180K+ USD", payoff: ">36", decision: deferred_regulatory }

4. Domain Architecture Decision

The architecture decision sits second in the dependency chain. It is locked once content is published; reversal requires migration with 301 redirect maintenance for 24 to 36 months and authority loss in the 8 to 25 percent range per Search Engine Roundtable migration aggregations of 2024 and 2025.

4.1 The Three Primary Architectures

ccTLD (country code top level domain). Pure country signal. example.com (US), example.co.uk (UK), example.de (Germany), example.fr (France), example.com.au (Australia), example.co.jp (Japan). Strongest geographic trust signal. Best for established multinationals with budget for separate per country presence and operations teams. Authority builds in country; local press cites the in country domain. Search engines weight the ccTLD signal substantially. The German Federal Court of Justice in Karlsruhe has rulings affirming .de domains carry higher consumer trust for German businesses (BGH Urteil, multiple cases, 2020 to 2024). Downside: authority does not share across the portfolio; each domain rebuilds backlinks, brand search, and engagement signals independently; operational cost increases per domain.

Subdomain country prefix. uk.example.com (UK), de.example.com (Germany), fr.example.com (France). Lower operational cost than ccTLD. Geographic targeting in Search Console can be set per subdomain. Subdomains have historically been treated by Google as separate properties for authority, though Mueller has publicly stated since 2019 that Google has reduced the distinction. The reality in 2026 is mixed; authority sharing between subdomain and root is partial.

Subdirectory locale path. example.com/en-us/, example.com/en-gb/, example.com/de/, example.com/fr/, example.com/en-au/. Authority concentrates on the single domain. Authority sharing is total; easiest operational management; cheapest to launch a new market because no DNS or certificate work is required. Downside is the weakest geographic signal; the .com domain reads as American by default in many markets. Hosting in Arkansas USA does not signal European market intent.

4.2 The Per Market Reasoning Matrix

The decision is rarely uniform across the portfolio. The hybrid pattern combines architectures based on per market reasoning.

4.3 The Hybrid Pattern

A common hybrid for mid market businesses with three to seven markets: example.com hosts the US primary plus /en-gb/, /en-au/, /en-ca/, /fr-ca/. example.de, example.fr, and example.co.jp run as ccTLDs because Germany, France, and Japan have local trust expectations that exceed the operational cost. The hybrid is operationally complex. Hreflang must point across domain boundaries. Search Console properties multiply. Backlink portfolios diverge. Going hybrid is defensible only when at least one market has strong cultural or regulatory pressure for ccTLD that exceeds the operational cost.

4.4 The URL Parameter Anti Pattern

example.com/?lang=en and example.com/?country=us&lang=en are universally a failure mode. URL parameters are weak ranking signals, frequently stripped by canonical tags, and historically misindexed. The Ahrefs 2024 hreflang study found URL parameter patterns had the highest hreflang error rate at 78 percent versus 31 percent for subdirectory and 23 percent for ccTLD.

5. Hreflang Reference

This section summarizes hreflang strategically. The full operational reference (14 patterns, canonical and pagination interaction, validation, audit rubric) lives in framework-hreflang.md.

5.1 What Hreflang Does

Hreflang tells Google which page version to show users based on language and country. Without hreflang on a multi locale site, Google may serve the wrong language version, treat locales as duplicate content with split ranking signals, or rank whichever locale has the strongest authority regardless of user country. With hreflang correctly deployed, Google serves the user the locale that matches their language preference and approximate country location.

Hreflang is not a ranking factor. It is a serving signal. It does not lift ranking in country A; it routes the user to the page that matches them once ranking is determined. The common failure mode is treating hreflang as a growth lever; it is a hygiene requirement.

5.2 The 14 Most Common Hreflang Patterns

The full pattern reference lives in framework-hreflang.md Section 5. Summary:

1. Language only across a single primary market (en, es, fr, de)
2. English variants (en-US, en-GB, en-CA, en-AU)
3. Spanish variants (es-ES, es-MX, es-AR, es-CO)
4. Portuguese variants (pt-PT, pt-BR)
5. French variants (fr-FR, fr-CA, fr-BE, fr-CH)
6. German variants (de-DE, de-AT, de-CH)
7. Chinese variants (zh-CN mainland, zh-TW Taiwan, zh-HK Hong Kong)
8. Arabic variants (ar-SA, ar-AE, ar-EG, ar-MA)
9. Mixed locale subset with x-default
10. Locale plus generic language fallback (en-US, en-GB, en, x-default)
11. Cross domain hreflang (example.com to example.co.uk to example.de)
12. Hreflang at the homepage only versus per page hreflang
13. Hreflang in XML sitemap versus HTML head versus HTTP header
14. Hreflang with paginated series (rel=next, rel=prev interaction)

5.3 The English Tripartite Market

The en-US, en-GB, en-CA market trio is the most common multi locale entry point for US first businesses. Treating all three as identical is a frequent failure mode. The three markets diverge in spelling, vocabulary, currency, regulatory expectation, and citation pattern. en-US uses dollar, color, elevator, gasoline. en-GB uses pound, colour, lift, petrol. en-CA blends with Canadian English preferring colour and centre but using miles per gallon in informal contexts while metric in formal. The en-AU and en-NZ markets are sometimes treated as fourth and fifth variants; en-AU prefers AUD currency and metric units; en-NZ prefers NZD with regional vocabulary divergence.

5.4 When Hreflang Fails

The most common failure modes, with the full catalog in framework-hreflang.md:

• Missing self reference. Each locale must reference itself plus all others.
• Missing bidirectional links. Locale A references B, B does not reference A. Google ignores.
• Wrong ISO codes. UK is GB. Brazil is BR. China is CN. Wales has no separate code; en-GB-WLS is not valid.
• Hreflang to non canonical URLs. Target must be a 200 status indexable canonical page, not a 301 redirect, not a noindex page.
• Missing x-default. The fallback for users not matching any specific locale.

The Ahrefs 2024 study of 374,756 domains found 67 percent had at least one implementation error. Most international sites in 2026 have broken hreflang. The audit work to confirm the cluster is bidirectional and self referencing is non optional and ongoing.

6. Content Localization vs Translation

Translation is the conversion of words from one language to another. Localization is the adaptation for cultural context, idiom, units, currency, date format, regulatory framing, citation patterns, examples, case studies, and tone. Search Engine Land's 2024 analysis of 1,200 multilingual sites found localized content earned 3.4 times the organic traffic of translated content over an 18 month window. Common Sense Advisory's 2023 survey of 8,709 consumers across 29 countries found 76 percent prefer to buy in their native language, 40 percent will not buy from a website not in their native language, and 65 percent prefer content with cultural references they recognize.

6.1 The Five Localization Layers

Layer 1, language. Spelling differences (color/colour), vocabulary (elevator/lift, gas/petrol), grammar conventions, formality registers. The layer most often handled in straight translation.

Layer 2, units and formats. Currency (USD, GBP, EUR, JPY, MXN, BRL), unit system (imperial vs metric), date format (MDY US, DMY UK and most of Europe, YMD Japan and ISO 8601), time format (12h with AM/PM vs 24h), phone format, address format (state vs county vs prefecture, ZIP vs postcode vs PIN).

Layer 3, examples and cultural references. Sports references (baseball/football US, cricket/rugby UK, hockey Canada, soccer Europe and Latin America), local landmarks, cultural touchstones (Thanksgiving in US and Canada on different dates, Boxing Day in UK and Commonwealth), case studies and customer logos that resonate in the target market.

Layer 4, regulatory and legal. GDPR (EU, UK), CCPA/CPRA (California with extending state laws across Colorado, Virginia, Connecticut, Utah, Texas, Florida, Oregon, Iowa, Indiana, Tennessee, Delaware, New Jersey, Montana, New Hampshire, Kentucky as of 2026), LGPD (Brazil), PIPL (China), APPI (Japan), PIPEDA (Canada), the Australian Privacy Principles. Privacy policy and terms of service variations per jurisdiction.

Layer 5, business specifics. Pricing for local market with local currency and local price elasticity. Service availability by region. Customer support hours and timezone. Local phone number. Payment processor coverage. Fulfillment and shipping coverage.

6.2 The Localization Depth Decision

The five layer model implies an investment ladder. The cheapest engagement runs Layer 1 only (machine translation reviewed by a native speaker). The deepest runs all five layers with in country content writers. Most engagements hybrid: full localization Layer 1 through 4 for transactional and high intent pages; Layer 1 with format adjustment Layer 2 for evergreen; machine translation with light review for archive depth.

high_intent_pages:   { language: full_native, units: full_adapted, examples: full_local,
                       regulatory: full_local, business: full_local }
medium_intent_pages: { language: full_native, units: full_adapted, examples: partial_local,
                       regulatory: global_referenced, business: partial_local }
evergreen_blog:      { language: professional_translation, units: format_adjusted,
                       examples: original_retained, regulatory: global_referenced,
                       business: global_referenced }

6.3 The Common Localization Failure Modes

Auto translation only. DeepL and Google Translate produce serviceable output in 2026 across major language pairs but lack idiom recognition and domain specific terminology. The Common Sense Advisory 2023 survey found 71 percent of consumers can identify machine translated content and 59 percent associate it with lower brand trust.

Identical content with currency converted. Misses cultural relevance. Search Engine Land 2024 found currency-only localization performed only 1.2 times better than no localization, far below the 3.4 times multiplier of full localization.

Direct copy of US site to UK with hreflang only. Identical content treated as duplicate. Ranking signals split, citation share is lower, and the UK locale typically ranks below the US locale because the US locale has older authority. The fix is layer 1 spelling plus layer 2 currency and unit adjustment plus layer 5 phone number and address variant; this minimal localization typically lifts UK to parity within 90 to 180 days.

Forgetting non content elements. Forms with date format, phone format, address format, state vs county vs province dropdown, support hours, payment and shipping options. Often the highest friction points in conversion.

Translating boilerplate without translating the high authority pages. Footer, legal pages, FAQ translate, but the primary pillar pages remain in the source language. Prioritize high intent and high authority pages first.

7. Regional Search Engine Optimization

Google holds the dominant share in most markets but several countries have legacy or sovereign search engines holding material share. Country specific engines have their own ranking signals, webmaster tools, content moderation, and citation behaviors. International SEO that ignores a regional engine holding more than 15 percent share leaves traffic on the table.

7.1 Baidu in China

Baidu held approximately 60 to 65 percent search share in mainland China in 2025 per StatCounter Global Stats and the China Internet Network Information Center reports. The engine has substantially different ranking signals, content moderation requirements, and infrastructure expectations.

ICP license requirement. Hosting in mainland China requires an Internet Content Provider license from the Ministry of Industry and Information Technology. The license process takes 20 to 90 days and requires a Chinese legal entity or an ICP filing through a Chinese partner. Without ICP, the site cannot be hosted in mainland China; it can still be indexed by Baidu but loads significantly slower from outside the Great Firewall and ranks lower as a result.

Content censorship matrix. Baidu filters content per the State Internet Information Office Cybersecurity Law of 2017 and the 2021 Data Security Law. Triggers include Tiananmen Square, Xinjiang and Uyghur references, Tibet independence, Taiwan independence, Falun Gong, criticism of senior CCP officials, and a moving target of contemporary topics. International sites entering China must scrub filtered topics or expect inconsistent indexing.

Baidu Tongji and Webmaster Tools. Baidu equivalents of Google Analytics and Search Console. Webmaster Tools at ziyuan.baidu.com. Chinese language interface. Setup requires a Chinese mobile phone number.

Practical ranking signals. Title tag and meta description are weighted more heavily than on Google in 2025 audits. Internal anchor text is weighted heavily. Server location and ICP status are direct ranking factors. Baidu's E-E-A-T understanding is less developed than Google's; older domains with long history rank well even with thinner content.

7.2 Yandex in Russia

Yandex held approximately 60 to 65 percent share in Russia in 2025 per StatCounter Global Stats. After the 2022 invasion of Ukraine, Yandex was partially divested and now operates with shifting corporate structure. International SEO in Russia in 2026 is constrained by sanctions for US, EU, UK, and other jurisdictions.

Yandex.Webmaster at webmaster.yandex.com is the Yandex Search Console equivalent. IKS metric (Indeks Kachestva Saita) replaced PageRank as the public quality signal; international sites typically score 0 to 100 on launch and need 6 to 18 months to reach 200 plus. Yandex.Metrica is free analytics with session recording.

US, EU, and UK sanctions limit advertising spend and commercial relationships with Yandex. Consult legal counsel before launching a campaign in Russia in 2026.

7.3 Naver in South Korea

Naver held approximately 55 to 60 percent share in South Korea in 2025 per StatCounter Global Stats (Google 30 to 35 percent). The Naver ecosystem is dominant and distinct from Google.

NaverWebmaster Tools at searchadvisor.naver.com. Korean interface with partial English translation in 2025.

Blog ecosystem and Shopping. Naver Blog and Naver Cafe are dominant content distribution surfaces in Korea, producing a significant share of Naver search results pages. Naver Shopping is the dominant ecommerce search surface; onboarding requires a Korean business entity or a Korean fulfillment partner. International brands need a Naver presence not just a localized website.

7.4 Yahoo Japan

Yahoo Japan operates as a separate property from Yahoo US and uses Google's search backend since 2010. Yahoo Japan held approximately 13 to 18 percent search share in Japan in 2025 (Google direct 75 to 80 percent). Optimization for Yahoo Japan is functionally optimization for Google in Japan, with minor differences in front end behavior and shopping integration.

Japanese language specifics. The Japanese writing system mixes kanji, hiragana, katakana, and Roman characters (romaji). Search queries split across the four scripts. The byte limit for visual title truncation in Japanese results is approximately 35 characters versus 60 in English. Keyword research per script is critical.

7.5 Seznam and Coccoc

Seznam held approximately 18 to 22 percent share in the Czech Republic in 2025, down from 40 plus percent in the late 2010s. Coccoc, a Vietnam specific browser and search engine, held approximately 5 to 7 percent share in Vietnam in 2025 (Google 90 plus percent). Both matter for hyper local campaigns but for most international work, Google captures these markets adequately.

7.6 The Regional Engine Decision

The decision reduces to share threshold and market priority. The team's rule:

• Above 30 percent share: required investment
• 15 to 30 percent: justified for high priority markets
• 5 to 15 percent: monitor only, do not invest
• Below 5 percent: ignore

By this rule, Baidu in China and Naver in Korea require dedicated investment. Yandex in Russia requires investment if the market is in scope despite sanctions. Yahoo Japan piggybacks on Google work. Seznam, Coccoc, and other long tail regional engines do not justify dedicated investment for most campaigns.

8. AI Engine Regional Behavior 2026

The fourth dimension of international SEO in 2026 is language model coverage. AI engines vary by country in two ways: coverage (whether the engine is available in the country at all) and behavior (whether, given a query, the engine produces citations that include the target site). Both differ from consumer search engine coverage and behavior.

8.1 The Reach Matrix

As of February 2026:

A country can have strong Google search reach and zero ChatGPT or Claude reach. Mainland China is the largest such market. Russia is another. Iran is a third. For these markets, AI citation strategy through ChatGPT, Claude, or Perplexity is moot. Bing Copilot, which has Chinese localization, becomes the priority AI surface in mainland China.

8.2 The Behavior Matrix

Engine behavior varies by query origin IP and by query language. The team's per country sampling across 2024 and 2025:

ChatGPT. Routes queries through different models depending on geographic region for performance and compliance. EU IPs hit slightly different content moderation thresholds than US IPs. Citations of EU specific sources are weighted higher for queries from EU IPs.

Perplexity. Source diversity heuristic favors in country domains for in country queries. A query from a UK IP about UK consumer law produces citations weighted toward .co.uk domains, Citizens Advice, UK Government, UK news outlets. Same query from a US IP produces citations weighted toward .com domains and US legal services.

Gemini. Google Search grounding produces citations consistent with Google Search rankings in the query region. A German IP querying Gemini about a German legal question produces citations consistent with Google.de top 10 results.

Claude. Web search uses a separate retrieval system that does not match Google's exactly. The team's sampling has found Claude favors Wikipedia, government sources, and major news outlets relatively more than Google AI Overview does. Regional bias is lower than Perplexity.

Bing Copilot. Follows Bing search rankings closely, including the regional bias Bing has in markets where it is established (UK and parts of Western Europe where Bing share is 4 to 8 percent versus 2 to 3 percent in the US).

8.3 The Per Country AI Strategy

Per country AI strategy is necessary, not optional. The team's framework:

High reach countries (US, UK, CA, AU, most of EU, JP, KR, SG, BR, MX, IN). Run the full AI engine optimization stack: Google AI Overview, ChatGPT Search, Perplexity, Claude web, Gemini, Bing Copilot. Apply the per engine decision tree in framework-multiengine-tradeoffs.md.

Partial reach countries (Russia, Vietnam, parts of MENA, parts of Africa). Run a subset. Skip the engines not available. Re prioritize based on local share. In Russia, Yandex specific work matters more than ChatGPT.

Low reach countries (mainland China, Iran, North Korea). AI citation strategy is largely moot. In mainland China, focus on Baidu and Bing Copilot. Skip ChatGPT, Claude, Perplexity.

8.4 Citation Behavior by Query Language and Monitoring Cadence

Distinct from query origin IP, query language drives citation behavior. A query in German receives citations weighted toward German language sources regardless of the IP. A query in English from a German IP receives citations weighted toward English sources. Per locale content matters more than per country IP work because users in country sometimes query in their second language.

AI engine reach matrices shift monthly. The team checks the OpenAI, Google, Anthropic, and Perplexity availability pages on the first of every month and updates the per country AI engine reach matrix in the intake.

9. Currency, Date, and Unit Localization

Server side detection, client side detection, user override, and SEO crawler exposure intersect to produce common failure modes.

9.1 Server Side Detection

Server side detection uses two inputs: the Accept Language HTTP header (browser language preference), and IP address geolocated through MaxMind GeoIP2. The two often disagree (a German speaker traveling in France, an American expat in Singapore).

http {
    geoip2 /etc/nginx/GeoIP2/GeoLite2-Country.mmdb {
        $geoip2_data_country_iso_code default=US source=$remote_addr country iso_code;
    }
    map $geoip2_data_country_iso_code $suggested_locale {
        default "en-US";
        US en-US; GB en-GB; CA en-CA; AU en-AU;
        DE de-DE; FR fr-FR; ES es-ES; MX es-MX; BR pt-BR; JP ja-JP;
    }
    server {
        listen 443 ssl http2;
        server_name example.com;
        location / {
            add_header X-Suggested-Locale $suggested_locale;
            try_files $uri $uri/ /index.html;
        }
    }
}

The configuration exposes the suggested locale as a response header but does not redirect. The redirect decision lives at the application layer where user choice and consent management override IP signal.

9.2 The Do Not Auto Redirect by IP Rule

Auto redirecting users by IP fails for four reasons. First, it blocks Googlebot from seeing all locale versions; Googlebot crawls from US IPs predominantly, and an auto redirect to en-US locks the German, French, and Japanese locales out of the index. The 2024 Google Search Central guidance explicitly states: "Avoid serving different content to users in different countries automatically based on perceived country or language." Second, it traps users who want a different locale than their IP suggests. Third, it produces SEO duplication. Fourth, it can be flagged as cloaking.

The correct pattern is to show a banner suggesting the alternative locale, let the user click through, and persist the choice.

9.3 Currency, Date, and Unit Display

The JavaScript Intl API handles currency, date, and unit display across locales without external libraries.

<script>
new Intl.NumberFormat("en-US", { style: "currency", currency: "USD" }).format(1234.56);
// "$1,234.56"   en-GB GBP: "£1,234.56"   de-DE EUR: "1.234,56 €"   ja-JP JPY: "￥1,235"

new Intl.DateTimeFormat("en-US",
  { year: "numeric", month: "long", day: "numeric" }).format(new Date(2026,4,14));
// "May 14, 2026"   en-GB: "14 May 2026"   de-DE: "14. Mai 2026"   ja-JP: "2026年5月14日"

new Intl.NumberFormat("en-US", { style: "unit", unit: "fahrenheit" }).format(75);
// "75°F"   celsius en-GB: "24°C"   mile-per-hour en-US: "100 mph"
// kilometer-per-hour en-GB: "160 km/h"
</script>

Currency conversion is a separate concern from display. The team uses a server side conversion service that caches rates daily. Product prices are set per locale and updated quarterly to reflect market price elasticity.

9.4 The Accessibility Implication

The IP based redirect anti pattern is also an accessibility failure. Users with VPN routing, users on cellular networks with shifting carrier IPs, and users in offices with corporate routing through other countries fail the IP test. The banner suggestion pattern works for these users; the auto redirect does not. See framework-accessibility.md for the user choice surface design pattern.

10. International Schema

Schema markup varies by market in two dimensions. First, the schema types differ; LocalBusiness in one country, Organization with Place entities in another. Second, the schema values differ; Wikidata IDs, Knowledge Panel IDs, and sameAs URLs are country specific.

10.1 The Per Country Organization Schema

A multinational business produces one Organization schema with hasMap referencing all locations, plus per country LocalBusiness or Place entities. The Organization schema lives on the global homepage and links to the country specific entities.

<script type="application/ld+json">
{
  "@context": "https://schema.org", "@type": "Organization",
  "@id": "https://example.com/#organization",
  "name": "Example Company", "url": "https://example.com",
  "logo": "https://example.com/logo.png",
  "sameAs": ["https://en.wikipedia.org/wiki/Example_Company",
             "https://www.wikidata.org/wiki/Q12345678",
             "https://www.linkedin.com/company/example-company"],
  "subOrganization": [{ "@id": "https://example.co.uk/#organization" },
                      { "@id": "https://example.de/#organization" },
                      { "@id": "https://example.com/au/#organization" }],
  "contactPoint": [
    { "@type": "ContactPoint", "telephone": "+1-555-0100",
      "areaServed": "US", "availableLanguage": "en" },
    { "@type": "ContactPoint", "telephone": "+44-20-5555-0100",
      "areaServed": "GB", "availableLanguage": "en" },
    { "@type": "ContactPoint", "telephone": "+49-30-5555-0100",
      "areaServed": "DE", "availableLanguage": ["de","en"] }
  ]
}
</script>

10.2 The Per Country LocalBusiness Schema

Each country with a physical presence gets a LocalBusiness entity, on the country specific homepage or Contact page, referencing back to the parent Organization.

<script type="application/ld+json">
{
  "@context": "https://schema.org", "@type": "LocalBusiness",
  "@id": "https://example.co.uk/#organization", "name": "Example Company UK",
  "parentOrganization": { "@id": "https://example.com/#organization" },
  "address": { "@type": "PostalAddress", "streetAddress": "123 High Street",
    "addressLocality": "London", "addressRegion": "Greater London",
    "postalCode": "EC1A 1AA", "addressCountry": "GB" },
  "geo": { "@type": "GeoCoordinates", "latitude": 51.5074, "longitude": -0.1278 },
  "telephone": "+44-20-5555-0100",
  "sameAs": ["https://en.wikipedia.org/wiki/Example_Company",
             "https://www.wikidata.org/wiki/Q12345678"]
}
</script>

10.3 Wikidata and Knowledge Panel

Wikidata holds entity IDs for businesses, people, and organizations. A multinational business often has one parent entity in Wikidata plus per country sub entities for major markets. The sameAs links in country specific LocalBusiness schema point to both. Multinational businesses with substantial press coverage per country can hold sub entities; smaller businesses cannot because Wikidata's speedy deletion will remove non notable entities within 48 to 72 hours.

Google's Knowledge Panel surfaces country specific information when triggered. A panel on a US query may show US address, US phone, US executives; the same brand on a UK query may show UK address, UK phone, UK executives. The panel pulls from the per country LocalBusiness schema, Wikipedia or Wikidata entity, and Google Business Profile for physical locations.

11. International Backlinks

Each country and language version needs its own backlink profile. Country specific TLDs and country relevant sites help geographic targeting in measurable ways. The team's working rule: 60 to 80 percent of backlinks pointing at a country specific locale should originate in country, with 20 to 40 percent from global sources with country specific authority signals.

11.1 Per Country Domain Authority Distribution

The per country backlink profile target:

11.2 In Country Journalist Outreach

GB: { publications: [The Times, The Guardian, The Telegraph, FT, BBC News,
                     TechCrunch UK, The Register, Computer Weekly],
      journalists: "30 tech, 15 business, 10 industry",
      platform: "Featured UK, Qwoted, Cision UK" }
DE: { publications: [Frankfurter Allgemeine, Süddeutsche Zeitung, Die Welt,
                     Handelsblatt, Spiegel Online, Heise, t3n],
      journalists: "25 tech, 15 business, 10 industry",
      platform: "PR Newswire DACH, news aktuell" }
FR: { publications: [Le Monde, Le Figaro, Les Echos, Liberation,
                     Franceinfo, BFM Business, 01net, Numerama],
      journalists: "20 tech, 12 business, 8 industry",
      platform: "AFP Forum, Newsroom France" }

11.3 Regional Digital PR Platforms

• Featured at featured.com provides journalist source requests with country filtering; international tier adds .co.uk, .ca, .com.au, .ie.
• Qwoted at qwoted.com similar to Featured with broader international coverage.
• Cision international plan at cision.com is the enterprise option with per country journalist databases for UK, Germany, France, Spain, Italy, Netherlands, Japan, Australia, Canada.
• HARO at helpareporter.com is US weighted but has international queries.
• PR Newswire DACH at prnewswire.com and news aktuell at newsaktuell.de cover German speaking markets.
• AFP Forum covers French and francophone markets.

11.4 The Common Backlink Failure Modes

All backlinks from .com. The entire backlink profile points from .com even for en-GB or de-DE. Google reads this as a US locale serving the foreign market and discounts the geographic relevance signal. Fix: build at least 30 percent of per locale backlinks from in country TLDs.

Per locale link velocity zero. The locale launches and never accrues new backlinks. Fix: quarterly press cycle per locale.

Backlinks from translated mirrors. Machine translated mirror sites with low authority and high spam signal. Fix: Google Search Console disavow file per locale.

Per locale brand search zero. No brand search volume in country; locale ranks but no one searches for the brand by name. Fix: in country brand awareness campaign separate from organic SEO.

12. International Site Performance

Site performance varies by user geography because origin server location and network path interact with the user's network speed. The team maintains performance parity across major markets without using third party CDN, proxy, or DNS based geo load balancing.

12.1 The Self Hosted Constraint

The team self hosts on Debian and nginx at IP 169.155.162.118 from Bubbles in Northwest Arkansas. Approximate round trip times:

First byte time at Bubbles for a static file is approximately 20 to 50 ms internally; user observed TTFB adds the network round trip. For en-US users, total TTFB is 50 to 130 ms; European, 110 to 230 ms; Australian, 240 to 430 ms.

12.2 The Refusal of Third Party CDN

The team does not use any third party CDN, proxy, or DNS based geo load balancing. The trade off is higher latency for users far from the Arkansas origin. The justifications: operational sovereignty (team controls the full request path), trust posture (clients did not contract with a third party vendor directly), cost discipline (edge per request economics underperform self hosting at the network's traffic profile), and operational simplicity (single origin, single deployment pattern, single log location).

International performance optimization happens at the origin, in the HTML payload, and via regional replica servers when latency parity is genuinely required.

12.3 The Multi Origin Pattern

For markets where Bubbles latency is unacceptable, the team deploys regional replica servers running the same nginx and same site content tree, replicated via rsync. nginx GeoIP module routes user requests to the nearest replica.

http {
    geoip2 /etc/nginx/GeoIP2/GeoLite2-Country.mmdb {
        $geoip2_data_country_iso_code default=US source=$remote_addr country iso_code;
    }
    map $geoip2_data_country_iso_code $origin_pool {
        default bubbles_us;
        US bubbles_us; CA bubbles_us; MX bubbles_us;
        GB valkyrie_eu; DE valkyrie_eu; FR valkyrie_eu;
        ES valkyrie_eu; IT valkyrie_eu; NL valkyrie_eu;
        JP m2_apac; KR m2_apac; SG m2_apac; AU m2_apac; NZ m2_apac;
    }
    upstream bubbles_us  { server 169.155.162.118; }
    upstream valkyrie_eu { server 81.x.x.x; }
    upstream m2_apac     { server 103.x.x.x; }
}

The replica pattern is rare. Most engagements run a single Bubbles origin. Reserved for engagements where Australian or Japanese market revenue forecasts justify the operational cost of a second origin plus rsync plus monitoring.

12.4 The Rsync Replication Pattern

# Rsync from Bubbles to regional replicas, daily at 04:00 UTC
for replica in valkyrie.example.eu m2.example.apac; do
  rsync -avz --delete /var/www/sites/ \
    deploy@${replica}:/var/www/sites/ \
    >> /var/log/rsync-${replica}.log 2>&1
done

12.5 The HTML Payload Optimization

Performance optimization at the payload level matters more for international users because the network round trip dominates total page load time over distance. The team's payload discipline: critical CSS inlined in the head under 14 KB for first packet delivery, above the fold image preload links in the head, lazy loaded images below the fold via the loading attribute, JavaScript deferred or async with no render blocking script, total HTML payload under 80 KB before image and font resources.

Lighthouse score target on a slow 4G simulated connection from a far origin should remain above 80 on the major locales. Bubbles measured Lighthouse scores from a simulated EU connection typically run 82 to 90; APAC, 75 to 88; South Africa, 70 to 85.

13. International Site Maintenance

International sites have higher maintenance overhead because every market has its own content velocity, regulatory clock, team profile, and citation surface. The maintenance cadence defines what the team does months 6 through 36 after launch.

13.1 The Per Market Content Velocity

Each market has a different content production rate. Primary market (US): 8 to 16 new pages per month, 4 to 8 refresh per month. Secondary primary (UK, CA, AU): 3 to 6 new per month, 2 to 4 refresh per month. Translated secondary (DE, FR, ES): 2 to 4 new per month, 1 to 3 refresh per month. Deferred markets: 0 to 1 per quarter.

Content velocity matters because the freshness signal is a measurable ranking factor. Locales with zero content velocity for 12 plus months show measurable ranking decay even when the original content is high quality.

13.2 The Regional Update Prioritization

Updates to global pages cascade to per locale pages on a defined order:

update_cascade:
  primary_locale_us: "immediate, day of source change"
  secondary_english_gb_ca_au: "within 7 days"
  major_translated_de_fr_es: "within 21 days"
  minor_translated_other: "within 60 days"

A change to a primary page triggers a translation queue entry for each downstream locale with the deadline calculated from the primary timestamp.

13.3 The Local Team Coordination

For engagements with in country teams, the team coordinates through a weekly cross market sync covering per market performance, content pipeline status, press cycle status, regulatory flags, AI citation drift, and incidents. 45 minutes max. Timezone aware; EU, US, and APAC members alternate the meeting time per quarter.

13.4 The Regulatory Monitoring Per Country

Regulatory frameworks change. The team monitors EU GDPR enforcement through EDPB's monthly report; UK ICO enforcement and guidance through ICO's quarterly newsletter; US state privacy law expansion (Colorado, Virginia, Connecticut, Utah, Texas, Florida, Oregon, Iowa, Indiana, Tennessee, Delaware, New Jersey, Montana, New Hampshire, Kentucky active in 2026); Brazil LGPD through ANPD's monthly bulletin; Japan APPI through PPC announcements; Australia Privacy Act review (Privacy Act 1988 reform was ongoing as of February 2026); California Privacy Rights Act updates from CPPA; China PIPL and DSL data export rules.

A change in regulatory posture in a target market triggers a per market policy review and per locale legal page update. Quarterly minimum, monthly during active enforcement waves.

13.5 The Per Market AI Citation Drift

AI citation behavior drifts as engines update training and retrieval systems. The team samples per locale quarterly through manual queries to ChatGPT, Claude, Perplexity, Gemini, and Google AI Overview from in country IP proxies. Drift surfaces as falling citation share, falling brand mention frequency, or new competitor citations replacing the brand. Drift triggers a citation strategy revisit per framework-aicitations.md.

13.6 The Quarterly Market Review

A 90 minute quarterly review per major market covering TAM refresh, competitor density revisit, revenue vs forecast variance, organic share trend, AI citation share trend, press cycle outcomes, regulatory updates, and team capacity. The review produces a decision: continue, increase, decrease, or sunset. Markets that fail to hit 18 month payoff with no path to 36 month payoff are sunset rather than continued in perpetuity.

14. Bubbles Hosted International SEO Toolchain

The toolchain runs on Bubbles, the team's Debian server at IP 169.155.162.118, with nginx fronting all sites under /var/www/sites/. No third party CDN, proxy, or DNS based geo load balancing. Supports Sections 3 through 13 without external dependencies.

14.1 The Server Profile

bubbles:
  ip: 169.155.162.118     os: Debian 12     webserver: nginx 1.24
  storage:
    primary: /var/www/          # SSD, 500 GB
    external: /mnt/storage/     # 4.5 TB SMR drive, cold archive
  database: { postgres: 16.x, sqlite: modernc.org/sqlite pure Go }
  modules: { geoip2: /etc/nginx/GeoIP2/GeoLite2-Country.mmdb, http2: enabled }
  certificates: Let's Encrypt via certbot, 60 day rotation

14.2 The Per Locale Site Layout

/var/www/sites/example.com/
  index.html                                # global default (US, x-default)
  en-us/  en-gb/  en-ca/  en-au/  de-de/  fr-fr/  es-es/  es-mx/
  hreflang/cluster.xml                      # hreflang via XML sitemap
  schema/organization.jsonld + schema/locales/
  audit/international/
    intake.yaml  market_scorecard.yaml  backlinks/  perf/  schema/

14.3 The Per Locale nginx Configuration

server {
    listen 443 ssl http2;
    server_name example.com;
    root /var/www/sites/example.com;
    index index.html;

    location ~ ^/(en-us|en-gb|en-ca|en-au|de-de|fr-fr|es-es|es-mx)/ {
        try_files $uri $uri/ $uri/index.html =404;
        add_header X-Locale $1;
    }

    location / { try_files $uri $uri/ /index.html; }

    location /sitemap.xml {
        alias /var/www/sites/example.com/hreflang/cluster.xml;
        add_header Content-Type application/xml;
    }
}

14.4 The Pa11y CI Locale Variant Configuration

// pa11y-ci.json
{ "defaults": { "timeout": 30000, "standard": "WCAG2AA" },
  "urls": ["https://example.com/en-us/", "https://example.com/en-gb/",
           "https://example.com/en-ca/", "https://example.com/en-au/",
           "https://example.com/de-de/", "https://example.com/fr-fr/",
           "https://example.com/es-es/", "https://example.com/es-mx/"] }

# Pa11y CI weekly run on Bubbles
cd /var/www/sites/example.com
pa11y-ci --config audit/international/pa11y-ci.json \
  > audit/international/pa11y-$(date +%Y%m%d).json

14.5 The Per Market Postgres Schema

CREATE TABLE international_pages (
    id SERIAL PRIMARY KEY,
    domain VARCHAR(255), locale VARCHAR(10), path VARCHAR(500),
    canonical_url VARCHAR(500),
    hreflang_cluster_size INTEGER,
    word_count INTEGER, last_updated TIMESTAMP,
    UNIQUE(domain, locale, path)
);

CREATE TABLE international_performance (
    id SERIAL PRIMARY KEY,
    domain VARCHAR(255), locale VARCHAR(10),
    measured_at TIMESTAMP, simulated_region VARCHAR(50),
    lcp_ms INTEGER, inp_ms INTEGER, cls_score DECIMAL(4,3),
    ttfb_ms INTEGER, lighthouse_score INTEGER
);

CREATE TABLE international_citations (
    id SERIAL PRIMARY KEY,
    domain VARCHAR(255), locale VARCHAR(10),
    engine VARCHAR(50), sampled_at TIMESTAMP,
    query_text TEXT, cited_url VARCHAR(500), citation_rank INTEGER
);

CREATE INDEX idx_intl_pages_locale ON international_pages (domain, locale);
CREATE INDEX idx_intl_perf ON international_performance (domain, locale, measured_at);
CREATE INDEX idx_intl_citations ON international_citations (domain, locale, engine, sampled_at);

14.6 The Daily Aggregation Cron

# /etc/cron.d/international-aggregation on Bubbles
0 3 * * *       deploy /var/www/scripts/aggregate-international.sh
0 4 * * 1       deploy cd /var/www/sites && pa11y-ci --config audit/international/pa11y-ci.json
0 2 1 * *       deploy /var/www/scripts/monthly-international-audit.sh
0 5 1 1,4,7,10 * deploy /var/www/scripts/quarterly-market-scorecard.sh

14.7 The Audit Rubric

Score: 15. World class: 13 plus / 15.

End of Framework Document

Document version: 2.0. Last updated: 2026-05-14.

Companion documents: framework-hreflang.md (operational hreflang reference), framework-localseo.md (within country local SEO), framework-multiengine-tradeoffs.md (per engine tradeoff matrix), framework-aicitations.md (cross engine citation surface), framework-aioverviews.md (Google AI Overview and AI Mode), framework-searchgpt.md (ChatGPT Search), framework-perplexityspaces.md (Perplexity Search, Pages, Spaces, Comet), framework-technicalseo.md (technical foundation), framework-schema.md (per market schema variants), framework-contentbriefs.md (per market brief generation), framework-brandvoice.md (per market voice), framework-eeat.md (per country trust signals).