Multilingual SEO ในปี 2026: คู่มือฉบับสมบูรณ์เพื่อการมองเห็นทั่วโลก (พร้อมเวิร์กโฟลว์ AI Agent)

---
name: multilingual-market-scorer
description: Analyze GA4 and GSC data to score and rank target markets for multilingual SEO expansion
---

# Multilingual Market Scorer

Score potential target markets using existing analytics data. This skill helps you prioritize which languages and countries to target first for multilingual SEO.

## Prerequisites
- User has shared GA4 and GSC data (CSV exports from each platform)
- User has defined their top 5 English-language target keywords
- User knows their business constraints (shipping regions, supported currencies, languages of customer support)

## Input
1. GA4 country-level organic traffic export (CSV)
2. GSC country-level performance export (CSV)
3. Top 5 target keywords in English
4. List of countries/regions where the business currently operates

## Workflow

### Phase 1: Extract demand signals
- Parse the GA4 CSV to extract: country, monthly organic sessions, conversion rate by country
- Parse the GSC CSV to extract: country, clicks, impressions, average CTR, average position
- Merge the two datasets on country name

### Phase 2: Score each market
For each country with measurable traffic:
- **Demand Score (1-5):** Based on monthly organic sessions. <100 = 1, 100-500 = 2, 500-2000 = 3, 2000-5000 = 4, 5000+ = 5
- **Opportunity Score (1-5):** Based on average CTR. Below 1% = 5 (high opportunity — users find you but can't read you), 1-2% = 4, 2-4% = 3, 4-7% = 2, 7%+ = 1
- **Business Fit Score (1-5):** User must provide this based on whether they operate in that country. Default to 3 if unknown.

### Phase 3: Rank and recommend
- Multiply Demand × Opportunity × Business Fit for a composite score (max 125)
- Tier 1 (60+): immediate priority — start localization for these markets
- Tier 2 (30–59): second-wave candidates — plan for next quarter
- Tier 3 (<30): monitor — revisit when Tier 1 and 2 are live
- Output a ranked table with: country, primary language, composite score, demand score, opportunity score, business fit score, recommended URL structure, estimated page count to localize

### Phase 4: Output a priority action plan
- Top 3 markets to target first, with recommended languages and URL structure
- List of pages to localize first (based on GSC top pages for each country)
- Estimated word count and translation budget (using current AI translation API pricing)
- Risk: list countries where high demand + low CTR exists but business doesn't operate — flag as strategic decisions

## Output
A structured report with:
1. Market prioritization table (all countries scored and tiered)
2. Top 3 recommended markets with rationale
3. First-wave localization page list (max 20 pages per market)
4. Estimated budget and timeline
5. Flagged strategic gaps

## Limitations
- All scores are estimates based on available data — real-world performance will vary
- Does not replace native-language market research or local competitive analysis
- GA4 and GSC data reflects current English-language performance, not latent non-English demand
- Does not access paid APIs; relies on user-provided CSV exports

---
name: multilingual-keyword-research
description: Generate and validate localized keyword lists for any target language and country, using AI translation plus SERP validation
---

# Multilingual Keyword Research Agent

Generate a localized keyword research report for any target language and market. Combines AI translation with SERP validation steps to produce keyword lists that reflect how real users search.

## Prerequisites
- User has defined target country (ISO code) and language
- User has provided 10-20 seed keywords in English (or their source language)
- User has access to Google Keyword Planner, DataForSEO, or is comfortable with manual SERP checks
- No API keys needed for the basic workflow; DataForSEO integration is optional for automated volume data

## Input
1. Target country (e.g., `DE`, `ES`, `JP`) and language (e.g., `de`, `es`, `ja`)
2. 10-20 seed keywords in source language
3. Business category or industry (for context)
4. Optional: DataForSEO API credentials (for automated search volume data)

## Workflow

### Phase 1: Translate and expand seed keywords
For each seed keyword:
- Generate a first-pass translation into the target language using AI (note which model was used)
- Identify 3-5 natural variants: synonyms, longer-tail phrasings, question forms, local terminology
- Flag any keywords where the direct translation is likely different from how locals search

### Phase 2: SERP validation (manual or automated)
For each translated keyword:
- Check Google Autocomplete on the target country's Google domain — record top 5 suggestions
- Check "Related searches" at the bottom of the SERP — record all relevant terms
- If DataForSEO is available: query search volume, CPC, and competition for each keyword
- If DataForSEO is NOT available: note this and provide instructions for manual Keyword Planner lookup

### Phase 3: Cluster by search intent
Group keywords into:
- **Informational:** "what is X," "how to Y," guides, definitions
- **Commercial:** "best X," "X vs Y," reviews, comparisons
- **Transactional:** "buy X," "X price," "X near me," product names
- **Navigational:** brand names, specific site searches

### Phase 4: Prioritize
Score each keyword cluster on:
- Relevance to business (1-5)
- Estimated volume tier (Low/Medium/High — do not fabricate exact numbers if API data is unavailable)
- Competition level based on SERP analysis (number of well-optimized pages on page 1, ad density)
- Content gap: does the user already have content targeting this keyword in this language?

### Phase 5: Output a keyword map
For the top 20 keywords:
- Keyword in target language
- English translation (for the user's reference)
- Search intent category
- Volume tier (L/M/H)
- Recommended content type (landing page, blog post, product page, glossary entry)
- Existing URL if content already exists in any language that could be adapted

## Output
A structured keyword research report with:
1. Market overview: total keyword opportunities found, intent distribution, volume summary
2. Top 20 prioritized keywords with full metadata
3. Content mapping: which keywords map to which existing or new pages
4. Native-speaker validation checklist: top 10 keywords to send to a native speaker for review, with specific questions ("Does [keyword] sound natural? What would a local say instead?")
5. Data sources used, retrieval dates, and gaps (keywords where volume data was unavailable)

## Limitations
- AI translation of keywords is a starting point, not a final answer — always validate with a native speaker
- Search volume data is an estimate from the data provider; actual volumes vary by season and market conditions
- Does not access paid keyword APIs unless user has configured them; manual fallback instructions are provided
- Cannot capture ultra-niche local slang or emerging terms without native-speaker input

---
name: translation-quality-check
description: Audit AI-translated pages for common quality issues — missing translations, glossary violations, text expansion, metadata gaps, and localization consistency
---

# Translation Quality Check Agent

Review AI-translated or human-translated pages for common multilingual SEO quality issues. Produces a prioritized fix list.

## Prerequisites
- User provides URLs or HTML files of both source and translated pages
- User optionally provides a glossary file (CSV: source_term, target_term, notes)
- User specifies source language and target language

## Input
1. Source page URL or HTML file path
2. Translated page URL or HTML file path (can be multiple target languages)
3. Source language code (ISO 639-1)
4. Target language code(s) (ISO 639-1)
5. Optional: glossary CSV for terminology consistency
6. Optional: brand voice guide or translation memory notes

## Workflow

### Phase 1: Structural check
For each translated page:
- Verify the page has a unique, translated URL (not the same URL as the source)
- Verify the HTML `lang` attribute matches the target language
- Verify the page serves real HTML (not client-side JS translation) — check if translated text is present in the raw HTML source
- Check that all meta tags are translated: `<title>`, `<meta name="description">`, `<meta property="og:title">`, `<meta property="og:description">`

### Phase 2: Content coverage check
Compare source and translated pages:
- Count `<h1>` through `<h4>` headings — verify all are translated
- Check image `alt` attributes — flag any still in the source language
- Check button text, form labels, error messages, footer links — these are frequently missed
- Check structured data (JSON-LD) — flag if schema content is in the source language

### Phase 3: Translation quality indicators
Flag potential quality issues:
- **Untranslated segments:** Text blocks in the source language appearing on the translated page
- **Text expansion/truncation:** Compare character counts of key elements (titles, CTAs, navigation items). Flag elements where the target language is >40% longer (risk of truncation in SERPs and UI)
- **Glossary violations:** If a glossary is provided, check that defined terms use the approved translation
- **Inconsistent terminology:** The same source term translated differently across the page (e.g., "checkout" translated as both "Kasse" and "Zur Kasse gehen" on the same German page)

### Phase 4: Localization depth check
Score the page on localization quality (not just translation accuracy):
- **Examples and references:** Are case studies, statistics, and examples localized or still US/UK-centric?
- **Currency, dates, measurements:** Are these in local format?
- **Cultural markers:** Currency symbols, address formats, phone number formats
- **Images:** Do images contain text in the target language, or are they still in the source language?

### Phase 5: Output a prioritized fix list
Categorize issues:
- **Critical (fix before launch):** Missing hreflang, untranslated title/meta, wrong lang attribute, client-side-only rendering
- **High (fix within 1 week):** Untranslated headings, missing alt text, schema in wrong language
- **Medium (fix within 1 month):** Inconsistent terminology, untranslated UI text, text expansion risks
- **Low (improvement backlog):** Non-localized examples, source-language images, formats

## Output
A structured quality report with:
1. Overall quality score (A–F) and summary
2. Critical issues list (with exact element locations)
3. High-priority issues list
4. Comparison table: source vs. translated page element-by-element
5. Native-speaker review checklist: specific sections to prioritize if budget is limited
6. Estimated fix time per priority tier

## Limitations
- Automated check cannot assess naturalness, idiomatic quality, or cultural nuance — a native speaker is still needed for final review
- Glossary matching is exact-string only; won't catch inflected forms without a comprehensive glossary
- Does not evaluate translation accuracy against a reference translation — this is a coverage and consistency check, not a fluency evaluation

<link rel="alternate" href="https://example.com/blog/" hreflang="x-default">
<link rel="alternate" href="https://example.com/blog/" hreflang="en">
<link rel="alternate" href="https://example.com/blog/de/" hreflang="de">
<link rel="alternate" href="https://example.com/blog/es/" hreflang="es">
<link rel="alternate" href="https://example.com/blog/fr/" hreflang="fr">

<url>
  <loc>https://example.com/blog/</loc>
  <xhtml:link rel="alternate" hreflang="en" href="https://example.com/blog/"/>
  <xhtml:link rel="alternate" hreflang="de" href="https://example.com/blog/de/"/>
  <xhtml:link rel="alternate" hreflang="es" href="https://example.com/blog/es/"/>
  <xhtml:link rel="alternate" hreflang="x-default" href="https://example.com/blog/"/>
</url>

Link: <https://example.com/brochure.pdf>; rel="alternate"; hreflang="en"
Link: <https://example.com/brochure-de.pdf>; rel="alternate"; hreflang="de"

---
name: hreflang-auditor
description: Crawl and audit hreflang implementation across an entire multilingual site — catch broken links, missing return tags, invalid codes, canonical conflicts, and generate fix-ready reports
---

# Hreflang Audit Agent

Crawl a multilingual website and audit every hreflang tag against Google's requirements. Produces a fix-ready report with exact URLs, error types, and severity ratings.

## Prerequisites
- User provides the base URL of the site (any language version — the agent will discover others via hreflang links)
- User confirms which URL structure is in use (subdirectory, subdomain, or ccTLD)
- No API keys required — uses HTTP requests and HTML parsing

## Input
1. Base URL of the website (e.g., `https://example.com/` or `https://example.com/de/`)
2. Known language codes if not all are discoverable (e.g., `["en", "de", "es", "fr", "ja"]`)
3. Optional: sitemap URL if hreflang is implemented via XML sitemaps
4. Optional: ignore list — URL patterns to skip (e.g., `/tag/`, `/author/`, `/page/`)

## Workflow

### Phase 1: Discover all language versions
- Crawl the provided base URL
- Extract all hreflang links from `<link rel="alternate" hreflang="...">` tags in the HTML `<head>`
- If XML sitemap is provided, also extract hreflang clusters from the sitemap
- Build a language-page matrix: every URL × every language version

### Phase 2: Validate each page in the cluster
For every page in the matrix, check these 8 rules:

1. **Self-referencing:** The page's own hreflang value points to its own canonical URL
2. **Bidirectionality:** For every pair (A→B), verify B→A exists
3. **Valid codes:** Language code matches ISO 639-1; region code matches ISO 3166-1 Alpha-2
4. **x-default present:** At least one page in the cluster has `hreflang="x-default"`
5. **Canonical alignment:** Each hreflang URL is the canonical version (not a parameterized or alternate URL)
6. **HTTP status:** Every hreflang URL returns 200 (not 301, 302, 404, or 500)
7. **No cross-language canonicals:** Each page's canonical points to a URL in the same language
8. **HTML lang match:** The `<html lang="...">` attribute value is consistent with the page's hreflang value

### Phase 3: Check for structural issues
- **Inconsistent clusters:** Pages that don't have the same set of hreflang tags as others in their cluster
- **Orphan pages:** Translated pages that exist but aren't referenced by any hreflang cluster
- **Chained redirects:** Hreflang URLs that redirect (301/302) — these should point directly to the final URL
- **Protocol mismatches:** HTTP vs HTTPS inconsistencies across the hreflang set

### Phase 4: Generate the fix report
For each issue found, output:
- Error type (from the 8 rules above)
- Severity: **Critical** (breaks the entire cluster), **High** (wrong page may be served), **Medium** (compliance issue), **Low** (best-practice deviation)
- Source URL (where the error was found)
- Target URL (the problematic hreflang link)
- Fix instruction: exact code or configuration change needed

### Phase 5: Generate corrected hreflang tags
For clusters with fixable errors:
- Output the corrected set of `<link>` tags for each page
- Output the corrected XML sitemap entries if applicable
- Flag clusters that cannot be auto-fixed (e.g., missing pages that need to be created first)

## Output
A structured audit report with:
1. Executive summary: total pages crawled, language versions found, errors by severity, overall health score (A–F)
2. Error table: each error with type, severity, source URL, target URL, and fix instruction
3. Cluster-by-cluster health: each page cluster with pass/fail per rule
4. Auto-generated fix code: corrected hreflang tags for every broken page
5. Prioritized action plan: which errors to fix first and why

## Limitations
- Crawls only linked pages within the hreflang clusters it discovers; pages without hreflang tags that should have them won't be found
- Cannot fix pages server-side — output is advisory only
- For JavaScript-rendered hreflang tags, the raw HTML approach won't work — use the XML sitemap method instead
- Does not check if the *content* of each language version is actually translated (use the translation quality agent for that)

---
name: international-visibility-monitor
description: Track multilingual search visibility across traditional SERPs and AI answer surfaces — generate weekly or monthly visibility reports per language and market
---

# International Visibility Monitor

Monitor how your site performs across languages in both traditional search and AI answer surfaces. Generates structured reports that track rankings, traffic, and AI citation presence per market.

## Prerequisites
- Google Search Console access (user provides CSV exports or grants view access)
- GA4 access (user provides CSV exports of language-segmented traffic)
- Optional: Ahrefs/Semrush/DataForSEO API access for automated rank tracking
- No API keys needed for manual data imports

## Input
1. Target languages and countries to monitor
2. Top 10-20 keywords per language
3. GSC performance export (CSV, filtered by country)
4. GA4 traffic-by-language export (CSV)
5. Optional: rank tracking API credentials
6. Previous monitoring report (for trend comparison)

## Workflow

### Phase 1: Collect traditional search data
- Parse GSC CSV: extract clicks, impressions, CTR, and average position per country and per language subdirectory
- Parse GA4 CSV: extract sessions, conversions, and engagement rate per language version
- If rank tracking API is available: pull current positions for tracked keywords per country
- If rank tracking API is NOT available: flag this and provide manual lookup instructions

### Phase 2: AI visibility check (manual or automated)
For each target language and its top 5 keywords:
- Document whether the brand appears in Google AI Overviews for those queries (search from the target country's Google domain)
- If tool access permits: check Perplexity and ChatGPT for the same queries
- Record: cited or not cited, which URL was cited, whether the citation is accurate
- Note any cross-language contamination (e.g., English URL cited for a Spanish query)

### Phase 3: Competitor visibility snapshot
For the top 3 competitors in each target market:
- Record their ranking positions for your shared keywords
- Check their AI Overview citation presence for the same keywords
- Flag competitors gaining or losing visibility month-over-month

### Phase 4: Trend analysis
Compare current data against the previous period:
- Traffic change per language (%)
- Ranking change per tracked keyword (positions gained/lost)
- AI citation presence change (new citations won, citations lost)
- Competitor movement (major gains or losses)

### Phase 5: Generate the report
Output a structured visibility report with:
1. **Executive dashboard:** One table with all languages and their key metrics (traffic, avg position, AI citations, trend arrows)
2. **Language deep-dives:** Per-language breakdown of top keywords, ranking changes, AI visibility status, and competitor activity
3. **Alert section:** Red-flag items needing immediate attention — traffic drops >20%, lost AI citations, hreflang errors found in GSC, competitors gaining on top keywords
4. **Action items:** Specific, prioritized tasks based on findings (e.g., "German blog post at position 11 for 'beste Laufschuhe' — optimize and add internal links to push into top 10")

## Output
A structured monitoring report with:
1. Multi-language dashboard (all languages, key metrics, trend arrows)
2. Per-language detailed section with keyword rankings, AI citations, and competitor snapshots
3. Red-flag alerts
4. Prioritized action items with expected impact
5. Data freshness: when each data source was last updated, and gaps where data is missing

## Limitations
- AI visibility checks are point-in-time snapshots — AI answers change frequently and may differ between queries minutes apart
- Rank tracking without an API requires manual lookups; automated rank data depends on third-party API availability
- GSC and GA4 data have inherent lag (24-48 hours for GSC, up to 48 hours for GA4)
- AI citation tracking is observational, not exhaustive — no tool currently provides complete AI citation coverage across all platforms