Build Personal Entity SEO

Create a verifiable person entity that search engines and answer engines can resolve across names, languages, websites, and profiles. Optimize for correct identification and citations, not guaranteed rankings.

Critical mental model: two visibility channels

Do not treat "LLM can answer correctly" and "Chinese web-search LLM can answer correctly" as one problem.

Channel	Who uses it	What they read	Typical failure
Entity layer	ChatGPT, Claude, some agents, crawlers that fetch llms.txt	llms.txt, JSON-LD, /about, entity.json	Missing files, deploy 404, inconsistent facts
Indexed web search	DeepSeek, 豆包, Kimi (联网), Perplexity-style retrieval	Baidu/Bing indexed HTML, Zhihu, .edu.cn, Baike	Site not indexed; homonyms outrank you

Lesson: A perfect llms.txt does not fix DeepSeek if Baidu has not indexed the authority page. Report channel readiness separately in every audit.

Choose the operating mode

• Audit/report: Inspect live and local state. Do not change files, deploy, submit URLs, or edit external profiles.
• Implement: Modify the in-scope site, validate, and deploy only when deployment is authorized.
• Monitor: Re-run search and LLM tests without memory, split by channel above.

Prefer the live site for crawl and rendering evidence. Preserve dirty worktrees and unrelated user changes.

Workflow

1. Establish verified facts

Collect only facts supported by the person or public evidence:

• canonical person ID;
• primary, native-script, Latin-script, and genuinely used names;
• stable role and short bilingual biography;
• gender and English pronouns when disambiguation matters;
• verified projects and organizations;
• official identity profiles;
• dated experience and education wording;
• contact method and update date;
• known homonyms and near-names to exclude (same Chinese characters, phonetic lookalikes, similar brands).

Mark unavailable projects as inactive. Do not expose dead links or add them to Schema. Do not turn admission into alumniOf, temporary experience into a permanent role, or age into a stable entity field.

Copy assets/entity-profile.template.json and fill it as the single fact source. Never invent biography, education, employment, awards, or independent coverage.

Gender lesson: Some English nicknames are often inferred as the wrong gender. If wrong-gender answers appear, add gender in JSON-LD (https://schema.org/Male or Female), state gender in llms.txt, use 他/她 or he/him in visible bios, and mirror the same wording on Zhihu/X/LinkedIn. Do not rely on a platform's hidden gender setting alone.

2. Select a canonical identity architecture

Use one stable person identifier, normally:

https://example.com/#person

Create one authority page per real language URL, for example /about and /en/about. Each page must have:

• HTTP 200 and server-rendered identity content;
• one H1;
• self-referencing canonical;
• reciprocal hreflang plus x-default;
• a first paragraph that directly answers who the person is;
• projects, dated background, official profiles, contact, and update date;
• visible copy consistent with JSON-LD;
• optional "common confusions (not the same person)" section when homonyms pollute search.

UI vs machine-readable lesson: It is valid to hide a homepage "About" link for brand minimalism while keeping /about indexable, in sitemap, and linked from llms.txt / cross-site JSON-LD. Never noindex the authority page to hide it.

Read references/entity-and-schema.md before defining the data model or JSON-LD.

3. Build the entity graph

Emit ProfilePage -> Person JSON-LD on authority pages. Connect a founded organization with:

Organization.founder -> Person @id

Use sameAs only for real identity profiles on other platforms. Do not put organizations, the site's own pages, llms.txt, OpenAPI files, or unrelated coverage into Person.sameAs on the personal authority page.

On a founded-organization site, linking the person's canonical profile URLs in embedded Person JSON-LD can help cross-domain resolution. Keep that graph consistent with the personal site's Person @id.

Ship a static entity.json and/or llms.txt on both personal and organization domains when possible. Prefer static files plus rewrite rules over extra serverless functions.

Deployment lesson (Vercel Hobby): Hitting the serverless function limit can fail the entire deploy while leaving stale production. After adding API routes, confirm deployment succeeded and that /llms.txt and /entity.json return 200, not 404 from a previous good build.

Keep inactive projects out of visible outbound links and structured data until their URLs return 200.

4. Handle multilingual names, misspellings, and homonyms safely

If the target includes Chinese search, homophones, transliterations, common misspellings, or same-name different people, read references/chinese-search.md.

Rules:

• Keep the correct native name first on the native-language profile page.
• Put only official or genuinely used aliases in alternateName.
• Explain a common misspelling once in useful visible prose.
• Put the correction in disambiguatingDescription, not in alternateName.
• Add a short not-the-same-as list for real homonyms that rank in search (e.g. same Chinese name on a .edu.cn admission list, a singer with a similar English stage name, a similarly named media brand). State why each is different; keep the list to a few high-impact cases.
• Repeat the same short correction on a real organization author page or another relevant owned source.
• Never create typo doorway pages, keyword-stuffed titles, or long lists of unrelated people.

Homonym lesson: Baidu often ranks .edu.cn admission PDFs and Baike for the exact Chinese name. A phonetically similar English name (e.g. [Stage Name] vs [Public Name]) can hijack queries like [domain][Native Name]. Explicit negative disambiguation in /about, llms.txt, and an indexed Zhihu answer beats hoping the model will guess.

5. Fix crawl and index hygiene

Ensure:

• robots.txt allows public content and blocks only non-content endpoints;
• sitemap contains canonical, indexable content URLs with real modification dates;
• canonical host is consistent across metadata, sitemap, redirects, and submissions;
• HTTP, www, and retired domains redirect permanently to the canonical host;
• verification files remain accessible when required (middleware may need exceptions for / and baidu_verify_*.html on the verification host);
• legacy content routes redirect to the closest relevant page;
• all major search and answer-engine bots receive 200 on authority pages;
• env-based site URLs are trimmed before sitemap generation (trailing newlines break <loc> URLs).

Treat llms.txt as a machine-readable summary for entity-capable systems, not a Baidu ranking lever. Use Markdown links. Including llms.txt in the HTML sitemap is optional; some teams add it for discovery.

SPA lesson: Client-routed article or profile views need runtime updates to document.title, canonical, and OG tags, or every shared URL will look like the homepage to crawlers and social previews.

6. Build cross-site evidence

Prioritize evidence in this order:

1. canonical personal profile;
2. founded organization site with founder in JSON-LD (front-end may de-emphasize founder; machine layer keeps facts);
3. official social and developer profiles with the same name, bio, and website;
4. one substantive Chinese Q&A or article (Zhihu answer > one-line bio for Baidu-dependent LLMs);
5. verified project attribution;
6. independent editorial or event citations.

Self-controlled sites improve consistency but do not count as independent notability. Seek at least three legitimate independent domains over 90 days. Do not create fake news, spam directories, bulk doorway pages, or premature Wikipedia/Wikidata entries.

Zhihu lesson: Profile pages may block automated browsers but the public API often exposes headline and gender. A good headline must include the Chinese name, role, canonical URL, and gender in text—not English-only one-liners. Example pattern:

示例姓名（男）· Example Public Name · [Organization] 创始人 · example.com

Nickname should lead with the native name when Chinese search is a goal.

7. Submit and monitor

Submit only final canonical URLs. If a verification host temporarily returns 200, restore its redirect after verification and submit the canonical host.

• Use Google Search Console, Bing Webmaster Tools, and Baidu Search Resource Platform where available.
• Use IndexNow for supported engines with scripts/submit_indexnow.mjs.
• For Baidu, treat API push, sitemap, and manual URL submit as separate channels. API push has a small daily quota on unregistered sites; prioritize authority pages first. www and apex are separate sites—verify both, use each host's token, and never push apex URLs with a www token.
• Keep credentials in environment variables. Never embed search-console, Baidu push, or IndexNow secrets in public files other than the required public key proof. Rotate tokens if exposed in chat or commits.

Read references/rollout-and-monitoring.md before planning submissions or recognition testing.

Validation gates

Before deployment:

1. Run the repository's lint and production build.
2. Copy assets/entity-seo.config.template.json, configure it, and run:

node scripts/verify_entity_seo.mjs --config /path/to/entity-seo.json --base http://localhost:3000

3. Parse every JSON-LD block as JSON.
4. Confirm visible text and Schema state the same facts.
5. Run Lighthouse SEO on each authority-page template.
6. Inspect the rendered mobile page and language switch.
7. curl production /llms.txt, /entity.json, and authority pages for 200 after deploy.

After deployment, rerun the verifier against production and confirm redirects, certificate coverage, and deployment status. Record indexing and LLM results; do not treat successful URL submission as successful indexing.

Recognition testing (split by channel)

Test in fresh, memory-free sessions. Record platform, date, query, cited URLs, wrong person, wrong gender, and homonym confusion.

Entity-layer prompts (should work once files are live):

Who is [Native Name] / [Public Name]?
What is [Public Name]'s Chinese name?
Who founded [Organization]?

Baidu-dependent prompts (may fail until indexed + Zhihu evidence):

[Native Name]是谁
[Public Name][Native Name]
[Native Name] [Organization] 创始人

Disambiguation prompts (run when homonyms are known):

[Native Name] 是男是女
[Native Name] 和 [Homonym] 是同一个人吗
[Public Name][Native Name]
[domain-slug][Native Name]

Do not promise first position or universal LLM recognition. Report technical readiness, Baidu indexing, off-site evidence, entity-layer LLM results, and web-search LLM results separately.

Public resources

• scripts/verify_entity_seo.mjs: deterministic endpoint, metadata, Schema, sitemap, redirect, and bot checks.
• scripts/submit_indexnow.mjs: verify an IndexNow key and submit sitemap or explicit URLs.
• references/entity-and-schema.md: entity model, JSON-LD graph, sameAs rules, and content patterns.
• references/chinese-search.md: Chinese names, typo correction, homonyms, Baidu host/submission, Zhihu, and anti-spam guidance.
• references/rollout-and-monitoring.md: 30/90-day rollout, metrics, evidence, and LLM testing.
• assets/: editable fact, verifier, monitoring, and evidence templates.