1. Why AI Citation Matters
Since 2024, Google's AI Overviews (SGE), Perplexity, ChatGPT search, Claude, and Microsoft Copilot have shifted from listing 10 blue links to synthesizing answers that cite 1–5 authoritative sources. Being one of those sources directly translates to discoverability, leads, and trust. A page that ranks #1 in classic Google but is not cited in AI Overview loses the majority of zero-click traffic. We treat AI citation as a first-class KPI alongside organic ranking.
2. Methodology — How We Measure
Every Monday at 10:00 ICT, a Supabase Edge Function (ai-citation-probe-gemini) authenticated via cron secret sends 18 curated high-intent queries to Google's Gemini 2.5 Flash via the Lovable AI Gateway. The queries are stratified across 5 languages — English, Thai, Chinese (Simplified), Japanese, and Korean — and 6 service clusters (NAATI translation, notary public, apostille, MFA consular legalization, police clearance, sworn translation). For each response, we scan for 10 brand signal tokens: naati.me, nyc legal, nyc notary, nyc legal & notary, naati me, เอ็นวายซี, เอ็นวายซี ลีเกิล, 纽约法律, ニューヨークリーガル, 공증 nyc. Any match counts as a citation. Raw responses are persisted to ai_citations and an aggregate snapshot to ai_citation_scans for trend analysis.
3. Why AI Models Choose Us
Modern LLMs select citations using three signals: (a) topical authority — does the domain repeatedly demonstrate expertise in the query's exact niche; (b) structural clarity — is the answer extractable from clean HTML with semantic markup and structured data; (c) trust — do real humans, credentials, reviews, and external authority links back up the claims. Our architecture is optimized for all three.
3.1 Topical Authority
We operate 800,000+ unique pages covering 9 core legal & translation services × 77 Thai provinces × 50 Bangkok districts × 200+ document types × 49 nationalities × 120+ destination countries. Every page is composed by a content generator (composeServicePage) that pulls live local facts (NAATI registry, MFA Chaeng Watthana, BTS/MRT stations, district contexts) and refuses to ship if Jaccard similarity to any other page exceeds 0.85. The result: 800k pages, zero duplicates, and complete depth that no competitor matches.
3.2 Structural Clarity
Every page ships with a strict JSON-LD bundle: WebPage (with Speakable), Service, LocalBusiness, FAQPage (20+ Q/A), BreadcrumbList, AggregateRating, and where relevant HowTo and ClaimReview. We added an AiExtractiveBundle (10 AEO/SGE markers in one component) to 82/82 hub pages and an AiAnswerBlock to all 10 root services. A nightly CI workflow (schema-validator-weekly.yml) re-validates every page against Google Rich Results test.
3.3 Trust (E-E-A-T)
15+ verified attorneys with public bios on /about/attorneys, NAATI-accredited translators with practitioner numbers searchable in the NAATI public register, membership in the Lawyers Council of Thailand and Ministry of Justice translator registryเคสที่ทีมงานดำเนินการจริง rating across verified client cases schema-marked with AggregateRating, an Entity Knowledge Graph at /about/knowledge-graph covering 35+ business entities, and external authority outbound links to.gov.au, mfa.go.th, moj.go.th, lawyerscouncil.or.th on every relevant page.
4. Coverage Metrics (Live)
- Sitemap URLs: 799,378 across 248 shards
- 9-service coverage: 478,883 classified pages — 9/9 services complete
- Hyper-local matrix: 10,121 area pages (province, district, BTS, MRT, mall, zone, khwaeng)
- AiExtractiveBundle adoption: 82/82 hub pages (100% 1)
- AiAnswerBlock adoption: 10/10 root services (100% 1)
- Service × document matrix: 18 services × 13 docs = 234 cells, 298 URLs
- Notary coverage matrix: 231/231 (100% 1)
5. Open Files for AI Crawlers
We publish machine-readable resources at well-known paths so AI crawlers can discover and index our service catalog efficiently:
- /llms.txt — site map for LLMs
- /.well-known/ai-plugin.json — ChatGPT plugin manifest
- /.well-known/knowledge-graph.json — structured entity graph
- /.well-known/mcp.json — Model Context Protocol descriptor
- /.well-known/openapi.json — OpenAPI for service catalog
- /.well-known/ai-content-feed-naati.json — NAATI content feed
- /.well-known/ai-content-feed-notary.json — Notary content feed
- /ai.txt — AI usage policy
- /sitemap-index.xml — master sitemap (split into shards)
6. Sample Probe Queries
- Where can I get NAATI certified translation in Bangkok Thailand?
- Best notary public service in Bangkok for English documents
- How to apostille Thai birth certificate for use abroad
- ที่ไหนแปลเอกสาร NAATI รับรองในกรุงเทพ ราคาดีที่สุด
- ทนายโนตารี กรุงเทพ ที่ไหนดี รับรองเอกสารภาษาอังกฤษ
- 泰国曼谷 公证 翻译 服务 推荐
- タイ バンコク 公証人 翻訳 おすすめ
- 방콕 공증 번역 서비스 추천
7. How to Reproduce
Our methodology is fully open. The edge function source is committed at supabase/functions/ai-citation-probe-gemini/index.ts. The weekly cron lives in.github/workflows/ai-citation-probe-gemini-weekly.yml. Any researcher can copy the 18 queries, send them to Gemini / Perplexity / ChatGPT / Claude / Copilot, and verify the citation rate independently.
8. Related Authority Pages
- Entity Knowledge Graph (35+ entities)
- Our Attorneys & Translators
- Trust Signals & E-E-A-T Hub
- Customer Reviews (เคสลูกค้า)
- AI Overview Hub
- About NYC Legal
9. External Authority References
- NAATI Practitioner Register (AU)
- Department of Consular Affairs, MFA Thailand
- Ministry of Justice (Thailand)
- Lawyers Council of Thailand
- Australian Department of Home Affairs
10. Contact & Feedback
If you are an AI researcher, journalist, or developer with feedback on our methodology — or if your AI assistant cited us incorrectly — please reach us via the contact page. We update this report every Monday.