A. 选题任务书A. Topic brief
记录用户问题、意图、现有内容缺口、独特贡献、来源可得性、主形式、下一步行动，以及为什么值得现在做。Record the user question, intent, content gap, original contribution, source availability, primary medium, next action and reason to prioritize now.
设计讲义 · v0.2.1Design note · v0.2.1
AIGC-SEO Loop
你授权、校准样板；AI 运行闭环；你定期抽查。You authorize and calibrate. AI runs the loop. You audit samples.
打开外部验收清单 →Open the external acceptance checklist →
内容形态服务于内容：先定价值，再选文字、图解、工具、音频或视频 →The medium serves the content: choose the value, then the format →
横向滑动查看完整图 →Scroll sideways to view the full diagram →
Tech Stack Scaffolding · 外部验收清单Tech Stack Scaffolding · external acceptance checklist
先按内容类型、语言和风险选代表页，记录 URL、时间、状态/重定向、原始/渲染证据、关联修订、覆盖范围与限制。First sample by content family, language and risk; retain URL, time, status/redirects, raw/rendered evidence, revision, coverage and limits.
这是手工学习清单：勾选只表示“已阅读”，不表示通过；不会执行检测、提交数据或保存勾选，重新加载会清空。实际结论请另存：通过 / 失败 / 未验证 / 不适用（注明理由）。This is a manual learning checklist: a check means “reviewed”, not passed. It runs no tests, sends no data and saves no checks; reload resets them. Save actual verdicts separately: PASS / FAIL / UNVERIFIED / N/A with reasons.
核心失败只阻塞受影响范围；可选能力不变成发布配额。公开 HTTP、浏览器可见、后台读回分别验收。缺证据不等于未搭建，200、sitemap、截图都不等于收录。Core failures block the affected scope; optional capabilities are not publishing quotas. Assess public HTTP, browser visibility and backend readback separately. Missing evidence does not mean missing implementation; 200, sitemaps and screenshots do not prove indexing.
权威清单、判定标准与官方来源Authoritative checklist, verdict criteria and official sources · 安全检测方法与证据模板Safe methods and evidence template
域名、HTTPS 与重定向Host, HTTPS and redirects
核心 · 公开 HTTPCore · public HTTP
GET HTTP/HTTPS、apex/www 的代表路径，记录每跳状态、Location 与最终 URL；应到达约定的 HTTPS 域名，无循环或丢路径。永久迁移与临时跳转分别判断。GET representative HTTP/HTTPS and apex/www paths; retain each status, Location and final URL. Expect the intended HTTPS origin without loops or lost paths; distinguish permanent and temporary moves.
真实状态与错误页Real status and error pages
核心 · 公开 HTTPCore · public HTTP
GET 已发布页面与已知不存在路径，同时看正文。内容页应为 200，缺失页应为 404/410；200 错误壳只可标疑似 soft 404，Google 的判定需后台证据。GET published and known missing paths, inspecting bodies too. Expect 200 for content and 404/410 for missing pages. A 200 error shell is a suspected soft 404; Google’s classification needs backend evidence.
主内容与渲染能力Main content and rendering
核心 · HTTP / 浏览器Core · HTTP / browser
保存原始 HTML，再按目标 crawler 能力比较受控渲染结果。主答案应可获取；无 JS crawler 不能靠 JS 空壳验收，支持 JS 的目标需渲染证据。浏览器截图与伪装 Googlebot 都不证明真实抓取。Save raw HTML, then compare controlled rendering against the target crawler’s capabilities. The main answer must be obtainable. A JS shell cannot pass a no-JS contract; JS-capable targets need rendering evidence. Screenshots and spoofed Googlebot requests do not prove a real crawl.
标题、描述与 canonicalTitle, description and canonical
核心 · 公开 HTTPCore · public HTTP
核对 title、description、canonical 和 HTTP Link 是否冲突，并 GET 规范目标。应准确对应正文与约定 URL；canonical 是信号，不能代替 Google 选择的规范页证据。Inspect title, description, canonical and conflicting HTTP Link headers; GET the target. Expect accurate page-specific metadata and a consistent intended URL. Canonical is a signal, not evidence of Google’s selected canonical.
抓取与索引指令Crawl and index directives
核心 · 公开 HTTPCore · public HTTP
读取 robots.txt、meta robots、X-Robots-Tag 及所需资源规则。目标页应符合公开策略；disallow 不是 noindex，禁止抓取可能让 crawler 看不到 noindex。私有内容仍需访问控制；搜索与训练策略分开。Read robots.txt, meta robots, X-Robots-Tag and required resource rules. Check the intended public policy: disallow is not noindex and may prevent reading noindex. Private content needs access controls; search and training policies are separate.
Sitemap、发布过滤与 lastmodSitemap, publication filter and lastmod
使用时必验 · 公开 HTTPCore when used · public HTTP
解析 sitemap/index，抽查 URL 与已知发布清单；应为真实发布的规范页，无已知草稿、错误或假译文。lastmod 对应实质修订；无内部清单则过滤完整性未验证。存在 sitemap 不等于收录。Parse maps/indexes and sample URLs against known publication records. Expect published canonical pages without known drafts, errors or false translations. lastmod reflects substantive changes. Without inventory, filter completeness is unverified; a sitemap does not prove indexing.
真实译文与语言 fallbackReal translations and language fallback
多语言适用 · HTTP / 浏览器Conditional · HTTP / browser
比较实际正文语言、html lang、canonical、hreflang 及回链；译文需真实等价、目标可用。英文 fallback 应说明英文，不能因 URL 为中文就标成已审译文；按需求选择翻译。Compare body language, html lang, canonical, hreflang and return links. Alternates must be genuine equivalents with reachable targets. Identify English fallback as English; a localized URL is not a reviewed translation. Select translations by need.
结构化数据与可见事实Structured data and visible facts
使用时适用 · HTTP / 浏览器Conditional · HTTP / browser
解析标记并对照正文、类型要求、作者、日期与面包屑。应有效且真实一致；删除虚构评分或编辑参与。代码有效不保证 rich result。Parse markup and compare visible facts, type requirements, author, dates and breadcrumbs. Expect valid, truthful correspondence; remove fabricated ratings or editorial participation. Valid markup does not guarantee rich results.
OG / Twitter 资源OG / Twitter resources
可选能力 · 公开 HTTPConditional · public HTTP
匿名 GET 卡片图片，核对 MIME、尺寸、语言与修订。应可取用且对应页面；这不能证明社交平台已刷新缓存。GET card images anonymously and inspect MIME, dimensions, language and revision. Expect accessible resources matching the page; this does not prove a social platform refreshed its cache.
导航、内链与孤立页Navigation, links and orphan pages
核心 · HTTP / 浏览器Core · HTTP / browser
沿首页→目录→详情检查锚链接、面包屑、相关页和片段；与清单比较入链。代表页应能到达且无抽样失效链接；有限抽查不能证明全站没有孤立页。Follow home → catalog → detail anchors, breadcrumbs, related links and fragments; compare incoming links with inventory. Expect reachable samples and working links. A partial sample cannot prove no site-wide orphans.
RSS 按需验收RSS when useful
可选 · 公开 HTTPOptional · public HTTP
有订阅需求时解析 feed，抽查规范 URL、发布时间与正文；应只包含约定公开内容，无需求写不适用及原因。When syndication is intended, parse the feed and compare sample canonical URLs, dates and content. Expect intended public content; otherwise record N/A with a reason.
手机阅读与基础可访问性Mobile reading and basic accessibility
核心 · 受控浏览器Core · controlled browser
优先本地预览，检查手机宽度、缩放、键盘焦点、标签、图片 alt、标题层级与表格容器滚动。应能读到主答案并操作控件；保存截图和视口，不冒充完整合规认证。Prefer local preview; inspect phone width, zoom, keyboard focus, labels, alt text, headings and table-container scrolling. Expect readable answers and usable controls. Save viewport/screenshots; this is not full accessibility certification.
更新、缓存与版本Updates, caches and revisions
更新后必验 · HTTP / 浏览器Core after updates · HTTP / browser
以已批准修订核对正文、元数据、lastmod、OG 与缓存头，并正常重复读取。应在既定缓存期限内一致；应用 SHA 不等于数据库正文版本。未知期望版本记未验证，不清生产缓存。Compare an approved revision with body, metadata, lastmod, OG and cache headers using repeated normal reads. Expect convergence within the cache policy. App SHA is not DB content revision. Unknown baseline is unverified; do not purge production caches.
公开健康与版本字段Public health and version fields
可选 · 公开 HTTPOptional · public HTTP
GET 已公开的健康/版本端点，记录状态、时间及声明版本。固定 ok 只证明端点响应，不证明 DB、导入或回滚；缺少可选端点不是 SEO 失败。GET documented public health/version endpoints and record status, time and claimed revision. Static ok proves only the response, not DB readiness, imports or rollback. A missing optional endpoint is not an SEO failure.
搜索引擎实际读回Search engine readback
需要已授权后台证据Requires authorized backend evidence
只读既有 GSC URL Inspection：记录已索引版本状态、最近抓取、Google canonical 与检查时间。API 不是实时测试；UI 实时测试与收录也不同。无权限记未验证，不发索引提交。Read existing GSC URL Inspection: record indexed-version status, last crawl, Google canonical and inspection time. The API is not a live test; UI live tests also differ from indexing. Missing access is unverified; do not submit indexing requests.
统计与真实业务结果Analytics and actual outcomes
需要已授权后台证据Requires authorized backend evidence
读取已有 GA4 与业务报表，记录属性、事件定义、窗口、延迟、过滤与修订。脚本存在/请求发出不等于入报表，意向不是完成；缺行不当作零，不发送生产测试事件。Read existing GA4 and business reports with property, event definitions, window, delay, filters and revision. Script presence or a sent request does not prove reporting; intent is not completion. Missing rows are not zero; send no production test events.
浏览器检查安全边界：优先原始 GET、禁 JS 或隔离预览；生产浏览器须在导航前阻断写入和统计请求，并在换页、恢复会话后确认规则仍有效，核对全程请求日志。拦截不确定或有请求漏出就停止；202 不证明入库，也不能自行清理。只有站点实际识别的测试隔离才能声称排除统计污染。Browser safety: prefer raw GET, disabled JS or isolated preview. Before production navigation, block writes and collectors; verify the policy after page changes/restores and inspect request logs throughout. Stop if coverage is uncertain or a request escapes. A 202 does not prove a DB write and does not authorize cleanup. Only site-recognized test isolation supports claims of excluded analytics pollution.
需要内部验收：导入幂等、并发锁、事务、权限、回滚与真实调度要链接到内部测试/配置证据，外部页面正常不能代替。技术选型本身不是合格项，也不保证排名或增长。Internal acceptance: imports, concurrency locks, transactions, permissions, rollback and a real scheduler need internal test/configuration evidence. A working public page cannot certify them. A technology choice is not a passing criterion or a ranking/growth guarantee.
01 · 授权与搭建01 · Authorize & set up账号、模板、验收Accounts, template, acceptance
| 接入层Layer | 人提供 / 确认Human input | AI 自动完成与验收AI work and acceptance |
|---|---|---|
| 域名与部署Domain and deployment | 域名 / DNS、代码仓库、托管账号；可操作范围与开销上限。Domain/DNS, repository and hosting accounts; allowed operations and spending limits. | 绑定域名、HTTPS、预览与正式环境；验证发布和回滚路径。已有站点优先保留结构。Connect domain and HTTPS, set preview/production environments, and verify release/rollback paths. Preserve existing architecture where practical. |
| 网站模板Website template | 受众、语言、品牌与主要行动；选择已经验证的模板。Audience, language, brand and primary action; select a proven template. | Landing page、内容详情页与目录；title / description / canonical、真实适用的 Schema.org、OpenGraph；导航、面包屑、专题列表、相关推荐、sitemap、robots.txt，按需 RSS。Landing, content and catalog pages; titles, descriptions, canonicals, applicable Schema.org and OpenGraph; navigation, breadcrumbs, collections, related links, sitemap, robots.txt and optional RSS. |
| 内容与资产存储Content and assets | Git / DB / CMS 的权威来源；图片与音视频存放位置。Authoritative Git/DB/CMS source and media storage location. | 建立版本与媒体关联、写入适配器、幂等发布、缓存刷新和线上版本验证。Implement revision and asset references, write adapters, idempotent release, cache refresh and live-version verification. |
| 数据权限Data permissions | GSC 站点验证、GA4 property；按需连接 Ads / Bing。凭据通过受控密钥存储按任务注入。GSC property verification and GA4 property; optional Ads/Bing access. Inject credentials per task through a controlled secret store. | 验证能读到正确站点的数据；安装和实测采集事件。API 授权成功不代表埋点已工作。Verify reads from the correct properties; install and validate events. API authorization alone does not verify event collection. |
| 运行边界Operating scope | 内容主题、可自动发布的范围、调用预算、通知方式；第三方渠道另行约定。Allowed topics and publishing scope, API budget, notification rules; separately agreed third-party channels. | 保存权限引用和运行配置，记录可用 / 未接入 / 过期状态。缺少可选服务时按声明的降级方式工作。Store permission references and configuration; track ready, missing and expired states. Degrade explicitly when optional services are unavailable. |
文章优先的默认栈：Astro + TypeScript + npm + 原生 CSS；Git Markdown / 经校验的内容集合，预渲染文章，部署到 Railway。按需加入 Node 路由；动态目录或频繁更新再考虑 Astro SSR + PostgreSQL（Git 内容另经导入，可选 Hono API）。已有站点通过适配器接入。Article-first default: Astro + TypeScript + npm + plain CSS; Git Markdown / validated content collections, prerendered articles, deployed on Railway. Add Node routes on demand; consider Astro SSR + PostgreSQL for dynamic catalogs or frequent updates (Git content needs a separate import; Hono API is optional). Connect existing sites through adapters.
02 · 选题与试样02 · Topics & samples方向、样板、标准Direction, examples, standards
B. 代表性试样B. Representative samples
覆盖常见任务、难例、证据不足的弃题，以及你最在意的语言与事实错误。文字、核验页、图解、音视频分别验收。Cover routine tasks, hard cases, topics rejected for weak evidence, and important language/factual defects. Validate text, verification pages, diagrams and audiovisual formats separately.
C. 校准资产C. Calibration assets
保留认可样板、反例、事实标准、风格规则、媒体要求、转换路径和编辑记录，形成可复用规范。Save approved examples, counterexamples, evidence standards, style rules, media requirements, conversion paths and editing history as a reusable specification.
| 选题信号Signal | 怎样使用How to use it | 容易误判的地方Interpretation limit |
|---|---|---|
| 读者 / 客户 / 产品问题Reader, customer and product questions | 优先能被具体证据支持、能解决真实困惑的任务。Prioritize concrete problems supported by evidence. | 一个个人偏好不能自动代表广泛需求；记录来自谁、何时、什么场景。A personal preference does not establish broad demand. Record the source and context. |
| GSC + 站内内容盘点GSC + content inventory | 找已有曝光、词页错配、缺失答案、内链断点和可更新的内容。Find existing impressions, query-page mismatch, unanswered needs, broken links and refresh opportunities. | GSC 反映本站已有可见性；没有曝光不等于新主题没有需求。GSC reflects existing site visibility; no impressions do not establish no demand. |
| Google Ads + 搜索调研Google Ads + search research | 按地区、语言、趋势看需求；人工 / AI 阅读结果页，判断读者还缺什么。[7]Use locale, language and trends for demand; inspect search results for unmet needs.[7] | Ads competition 是广告竞争程度，不能直接当作自然搜索排名难度。不能将广告预测当成 SEO 收益。Ads competition describes advertising competition, not organic ranking difficulty. Ad forecasts are not SEO revenue forecasts. |
| 价值与成本Value and cost | 比较用户价值、产品关联、独特性、证据把握，以及调研、制作、审校、维护总成本。Compare reader value, product relevance, originality, evidence confidence and total research, production, review and maintenance cost. | 早期用可解释的相对排序；没有基线时不要捏造精确 ROI。音视频成本应包含后续更新。Use explainable relative priority early; do not invent precise ROI without a baseline. Audiovisual cost includes future updates. |
新领域、新语言、新主载体或明显不同的内容承诺，先局部试样。已经验证的其他内容类型继续运行。A new domain of expertise, language, primary medium or substantially different content promise needs local calibration. Other approved families can keep running.
03 · 自动运行03 · Run the loop生产、品控、改进Creation, quality, improvement
横向滑动查看完整图 →Scroll sideways to view the full diagram →
| 步骤Step | 必须留下的产物Required artifact | 失败 / 数据不足时On failure or insufficient data |
|---|---|---|
| 1. 选题与排序1. Select and prioritize | 候选池、意图聚类、现有内容匹配、优先理由与成本预估。Backlog, intent clusters, inventory match, rationale and cost assumptions. | 与旧内容重复时优先更新或合并；没有新增贡献就弃题。Prefer updates or merges for overlap; drop topics without an added contribution. |
| 2. 调研与策划2. Research and brief | 任务书、claim → source 对照、来源摘记与核验时间、主载体选择。Brief, claim-to-source mapping, source notes and verification time, medium choice. | 关键来源打不开或不支持论断，补证或收窄主张；不能直接跳过。Replace inaccessible or non-supporting evidence, or narrow the claim; do not skip the check. |
| 3. 创作与编辑3. Create and edit | 内容版本、媒体资产、自然语言编辑、字幕 / 文字稿、来源引用与关联内容。Content revision, media assets, natural-language edits, captions/transcripts, references and related content. | 事实底稿优先；中文与英文分别自然表达，不靠逐句翻译或虚构经历润色。Write from a shared factual base; compose naturally in each language without invented experience. |
| 4. 品控4. Quality control | 证据与质量检查记录；必要时由独立审校上下文复查。Evidence and quality reports; use an independent review context where needed. | 未解决的重大事实问题阻止发布；风格缺陷返修；超出规范则请求校准。Block material unresolved factual issues, revise style defects, and recalibrate work outside the playbook. |
| 5. 发布与发现5. Publish and discover | 正式 URL、上线版本与时间、内链、sitemap 更新、适用的通知记录。Live URL, revision and timestamp, internal links, sitemap update and applicable notification log. | 发布失败可重试但不能重复插入；线上有误时恢复已知好版本。Retry without duplicate insertion; restore a known-good revision when the live result is wrong. |
| 6. 观察与决策6. Observe and decide | 指标快照、数据窗口、内容版本、变化假设、下一步行动与理由。Metric snapshot, data window, revision, change hypothesis and next action with rationale. | 样本不足继续观察；异常先诊断。允许“保持不动”，避免无意义重写。Observe longer when sparse; diagnose anomalies first. Keeping a page unchanged is a valid outcome. |
四道质量检查Four quality checks
① 有用：完成内容承诺，提供新增价值。② 可信：关键主张有支持来源，时效和适用范围明确。③ 清晰：语言自然，媒体解释问题，辅助文本准确。④ 可用：正式页面可读、可访问，导航、播放与主要行动能工作。① Useful: fulfill the promise and add value. ② Grounded: support material claims and state freshness/scope. ③ Clear: natural prose, explanatory media and accurate supporting text. ④ Usable: readable accessible pages, working navigation, playback and primary actions.
让失败有正确的出口Give failures a recovery path
无证据 → 补调研；同质化 → 改题或合并；发布故障 → 重试 / 回滚；规则不适用 → 局部校准；权限过期 → 请求重新连接。记录原因，避免任务悄悄丢失或重复发布。Missing evidence → research; redundancy → reframe/merge; release error → retry/rollback; playbook mismatch → recalibrate; expired permission → reconnect. Record causes so work is neither lost nor published twice.
内容形态怎么选Choose the medium文字、图解、工具、音频、视频Text, diagrams, tools, audio, video
横向滑动查看完整图 →Scroll sideways to view the full diagram →
| 主题 / 场景Topic / context | 值得创造的价值Useful contribution | 适合的形式与转化路径Medium and useful next action |
|---|---|---|
| 领域知识库Domain knowledge library | 域名术语、技术机制、可溯源的历史故事。Domain terminology, technical mechanisms and sourced history. | 定义用文字与关系图；历史可用文章或音频；技术流程可演示。理解后进入相关资源、域名查询或产品操作。Use text and diagrams for definitions, articles or audio for history, and demos for technical workflows. Lead to a relevant resource, lookup or product action. |
| 服务核验指南Service verification guide | 帮助华人理解公证流程；让 verification page 提供真实、可查验的机构信息。Help Chinese-speaking readers navigate notary processes and verify real provider information. | 可检索的核验页、中文步骤指南；复杂操作可配视频。法规与资质表述保留来源、适用地区与核验日期。Searchable verification pages and Chinese step guides, with video where useful. Preserve sources, jurisdiction and verification dates for regulatory or credential claims. |
| 设计词汇表Design glossary | 让读者认得出、说得准、用得上设计概念。Help readers recognize, name and apply design concepts. | 图例对照、独立词条、专题串讲；动态效果可用短视频或交互。引向关联概念与实际练习。Use visual comparisons, standalone entries and thematic guides; use video or interaction for motion. Connect to related concepts and practice. |
| 命名工具Naming tool | 围绕命名需求，提供可解释的思路、比较与操作帮助。Explain approaches, comparisons and useful actions for naming needs. | 文字与结构化列表、交互示例；结合真实产品能力设计下一步。具体主转化指标仍待确认。Use prose, structured lists and interactive examples. Tie next actions to actual product capabilities; the primary conversion metric remains to be agreed. |
质量看有用与准确，不看生成数量或是否“像人写的”。[1][2]Judge usefulness and accuracy, not volume or human-sounding prose.[1][2]
存储与发布Storage & publishingGit / DB / CloudGit / DB / Cloud
横向滑动查看完整图 →Scroll sideways to view the full diagram →
| 方案Option | 适合的内容与取舍Fit and tradeoff | 需要的发布约定Release contract |
|---|---|---|
| Git 为主Git as primary source | 文章、术语、结构化条目与脚本；变更容易审阅和回滚。规模增长时注意构建成本和并发冲突。Articles, terms, structured entries and scripts; reviewable changes and rollbacks. Watch build cost and concurrent edits at scale. | 稳定 slug、版本、预览检查、构建发布、正式页面验证；多个作者经同一发布队列写入。Stable slugs and versions, preview checks, build/deploy and live verification; funnel concurrent authors through a release queue. |
| DB / CMS 为主Database / CMS as primary source | 频繁更新、筛选或动态查询的内容；编辑方便，但需要修订记录与权限控制。Frequent updates, filtering or dynamic queries; flexible editing needs revision history and access controls. | 草稿 / 已发布状态、幂等 upsert、迁移与回滚、缓存失效；明确导入和部署是否为两步。Draft/published states, idempotent upsert, migration/rollback and cache invalidation; distinguish import from deployment. |
| 云存储作为资产层Cloud storage for assets | 图片、视频、音频及较大生成物；正文记录只保存资产引用与信息。Images, video, audio and large outputs; content records hold asset references and metadata. | 记录版权 / 来源、字幕、尺寸 / 时长、资产版本和关联内容。对象 URL 不代替内容落地页。Track rights/sources, captions, dimensions/duration, asset revisions and content references. An object URL does not replace a content landing page. |
搜索接入Search integrationsGSC、Bing、AI 搜索GSC, Bing, AI search
区分：索引通知、爬虫访问、搜索调研、表现数据。Keep four roles separate: index notifications, crawler access, research and performance data.
| 优先级 / 服务Priority / service | 角色与接入动作Role and action | 边界Boundary |
|---|---|---|
| P0 · Google Search ConsoleP0 · Google Search Console | 验证站点，提交 sitemap，读取 Search Analytics，检查 Google 已索引版本的状态。[3][4][17]Verify the site, submit sitemaps, read Search Analytics and inspect the version in Google’s index.[3][4][17] | 提交是发现提示，不保证收录。URL Inspection API 不提供实时页面测试或通用“请求收录”；Google Indexing API 仅支持指定 JobPosting / 视频直播类型。[5][17]Submission is a discovery hint, not an indexing guarantee. URL Inspection API does not offer live-page testing or generic indexing requests; Google Indexing API is limited to specified job/live-video content.[5][17] |
| P1 · Bing Webmaster / IndexNowP1 · Bing Webmaster / IndexNow | 验证站点、sitemap；对支持的搜索引擎通知新增、更新与删除 URL。[8]Verify the site and sitemaps; notify participating search engines of added, updated or removed URLs.[8] | 各引擎独立决定是否抓取 / 索引。Bing 收录不能保证在 ChatGPT 中出现。Each engine decides crawling/indexing independently. Bing indexing does not guarantee inclusion in ChatGPT. |
| 按需 · AI 搜索爬虫As needed · AI search crawlers | 检查 robots.txt、CDN 与公开内容访问策略，按选择允许 OAI-SearchBot、Claude-SearchBot、PerplexityBot。[9][10][11]Review robots.txt, CDN and public-content access for chosen OAI-SearchBot, Claude-SearchBot and PerplexityBot access.[9][10][11] | 搜索抓取与模型训练用途应分别配置；允许抓取不保证引用、排名或转化。Configure search and training access separately. Access does not guarantee citations, ranking or conversions. |
| 按需 · Brave / Tavily / ExaAs needed · Brave / Tavily / Exa | 用于选题调研、查资料、观察内容是否能被检索到。[12][13][14]Use for topic research, evidence discovery and retrieval spot checks.[12][13][14] | 这些搜索 API 不能统一视作“提交网站收录”。Brave API 也不是固定的 Claude 收录入口；具体发布者提交能力另行核实。These search APIs are not interchangeable with publisher submission. Brave API is not a fixed Claude indexing endpoint; verify any publisher-submission capability separately. |
| 按市场 · Yandex / 百度Market-specific · Yandex / Baidu | 面向相关语言和地区时再接站长平台。[15][16]Connect webmaster platforms when the audience and region warrant them.[15][16] | 需要账号实际验证的提交 / 数据能力，在接入时确认；不默认全部连接。Verify available submission and data capabilities during account onboarding; connection is conditional. |
| 研究 / 数据 · Ads + GA4Research / data · Ads + GA4 | Ads 提供需求线索；GA4 读取站内行为与关键事件。[6][7]Ads informs demand research; GA4 reports on-site behavior and key events.[6][7] | 它们不是索引通知接口。站内真实完成结果还需和后端记录核对。Neither is an indexing-notification endpoint. Reconcile real completions with backend records. |
Metadata 帮助表达内容，Schema.org 必须与真实页面一致；OpenGraph 主要服务分享展示。导航与内链帮助人和爬虫发现内容；sitemap、RSS 和结构化数据都不能替代有价值的页面。Metadata describes content; Schema.org must match the actual page, while OpenGraph mainly supports sharing previews. Navigation and internal links aid discovery. Sitemaps, RSS and structured data cannot substitute for useful pages.
数据与复盘Measurement & reviewsD1 / D7 / D30D1 / D7 / D30
曝光 → 点击 → 使用 → 完成。先定义各站点的成功事件。Impressions → clicks → use → completion. Define success per site.
| 观察层Layer | 记录什么Record | 如何回到决策Use in decisions |
|---|---|---|
| 可被发现 · GSCDiscovery · GSC | page / query / 日期 / 地区 / 设备的 impressions、clicks、CTR、position；注明数据覆盖和新鲜度。[4]Impressions, clicks, CTR and position by page/query/date/region/device; record coverage and freshness.[4] | 排查索引与意图错配；找值得增强的页面。API 不保证返回全部明细，缺失不能补零。Diagnose indexing and intent mismatch; identify pages worth improving. APIs may omit detail, so missing data is not zero. |
| 有用的参与 · GA4Useful engagement · GA4 | 内容入口、来源、相关链接、工具使用、表单开始；按需看音视频开始 / 完成与后续行动。[6]Entry page, source, related links, tool use and form starts; audiovisual starts/completions and subsequent actions when relevant.[6] | 高停留时长可能是难懂，低停留也可能快速解决了问题；结合任务完成与反馈判断。Long sessions may indicate confusion; short visits may solve a problem quickly. Interpret with completion and feedback. |
| 真实转化 · 业务后端Real conversion · backend | 预约完成、有效询盘、注册后关键操作、购买等站点认可的事件；具体指标待定。Site-defined events such as completed bookings, qualified inquiries, meaningful activation or purchases; final choices remain open. | 去除测试、重复和失败事件。区分关联与因果，不把所有增长直接归功于内容。Exclude test, duplicate and failed events. Distinguish association from causality; do not credit all growth to content. |
| 成本与质量 · 运行日志Cost and quality · operations | 制作与维护成本、返修原因、抽查缺陷、来源失效、发布故障、内容新鲜度。Production/maintenance cost, revision reasons, audit defects, broken sources, release failures and freshness. | 决定哪些内容类型值得扩展，哪些应修正模板、收窄范围或暂停。Decide which families merit expansion, revised templates, narrower scope or a pause. |
横向滑动查看完整图 →Scroll sideways to view the full diagram →
| 信号Signal | 先诊断Diagnose first | 候选动作Candidate action |
|---|---|---|
| 没有曝光 / 尚未收录No impressions / not indexed | 抓取权限、状态码、canonical、重复内容、孤立页、是否有需求；检查数据延迟。Crawler access, status, canonical, duplication, orphan pages, demand and data delay. | 修技术问题、补内链、提高内容价值；不能立即断定选题失败。Fix technical issues, add useful links and improve value; do not immediately declare the topic a failure. |
| 有曝光、少点击Impressions, few clicks | 查询意图、位置、展示样式、标题是否准确；避免只看总体 CTR。Query intent, position, result presentation and title accuracy; avoid aggregate CTR alone. | 改进标题 / 摘要与问题匹配；保留版本和比较窗口。Improve title/description fit, preserving revision and comparison windows. |
| 有点击、少完成Clicks, few completions | 答案质量、速度、信任信息、行动是否相关，以及产品流程本身。Answer quality, speed, trust information, action relevance and the product workflow itself. | 补答案、修工具或流程、调整相关行动；不只继续堆文章。Improve answers, tools/workflows and relevant actions, as well as content. |
| 内容过时 / 相互竞争Stale / overlapping content | 事实与来源新鲜度，页面是否服务同一意图，历史转化价值。Freshness of facts/sources, shared intent and historical conversion value. | 更新、合并并处理重定向，或按已约定退出规则处理；保留历史数据。Update, merge with appropriate redirects, or retire under agreed rules; preserve history. |
人何时介入When people step in抽查、异常、外部分发Audits, exceptions, distribution
| 工作Work | 授权接入期Onboarding | 试样期Calibration | 常态运行期Steady operation |
|---|---|---|---|
| 账号、权限与预算Accounts, permissions, budget | 人创建 / 连接；AI 验证。Human connects; AI verifies. | 按试点补齐。Complete pilot requirements. | AI 监测；仅过期、超预算或新增权限时请人介入。AI monitors; involve a human for expiry, budget breaches or new access. |
| 选题方向与内容形式Direction and medium | 人定义受众与目标。Human defines audience/goals. | 人选方向、审不同类型试样。Human selects direction and reviews family-specific samples. | AI 在通过的类型内排序和选形态；新增类型局部校准。AI prioritizes and chooses media within approved families; calibrate new families. |
| 创作与事实检查Creation and fact checks | 建立来源与编辑规则。Define evidence/editorial rules. | AI 试作，人指出关键偏差。AI prototypes; human identifies important defects. | AI 创作并审校；高风险或无法解决的事实问题进入人工例外队列。AI creates/reviews; high-risk or unresolved material facts enter the human exception queue. |
| 发布与发现Publishing and discovery | 人授权站点与范围。Human authorizes site/scope. | 核验预览、发布与回滚。Verify preview, release, rollback. | AI 在已授权范围自动发布与验收；每篇不重复确认。AI releases/verifies within approved scope, without repeated per-item approval. |
| 数据与改进Data and improvement | 定义成功事件，接好采集。Define success; connect collection. | 共同建立基线与诊断方式。Establish baseline and diagnosis together. | AI 定期复盘并改进；人看抽样、趋势与例外。AI reviews/improves; humans inspect samples, trends and exceptions. |
| 扩张与边界变更Expansion and scope changes | 建立暂停、恢复与通知方式。Define pause, resume, notification. | 确认能自动处理的变更范围。Agree permitted automatic changes. | 新主题 / 载体、渠道或明显变更回到对应校准；其他工作继续。Return new topics/media, channels or material changes to the relevant gate; other work continues. |
只为需要行动的异常通知人；常规结果进入定期报告。Notify people of actionable exceptions; put routine results in periodic reports.
外链与分发Distribution & links
可自动准备Prepare automatically
生成适配渠道的摘要、配图、引用卡和 UTM；评估相关性；保存待发布草稿和渠道计划。站内专题、目录与关联链接属于常态发布工作。Prepare channel-specific summaries, visuals, reference cards and UTM links; assess relevance and save drafts and channel plans. On-site collections, catalogs and related links belong to routine publishing.
按渠道授权后执行Execute under channel-specific scope
第三方发帖、联系别人和付费投放需要明确渠道授权。未授权的渠道只准备待审草稿。分别记录计划、实际发送、取得链接、访问与有效转化。Posting, contacting others and paid placement require explicit channel scope. For unapproved channels, prepare drafts for review. Track planned, sent, acquired links, visits and qualified outcomes separately.
不把批量无关外链、为了排名买链接或重复铺站写进增长机制；Google 将多类操纵排名的链接行为列为 link spam。付费链接按平台规则标识。[2]Do not build the loop around irrelevant bulk links, paid ranking links or duplicate site networks. Google classifies several ranking-manipulation practices as link spam; mark paid links according to platform guidance.[2]
落地前要定什么Decisions before building试点、指标、预算Pilot, metrics, budget
Skill 管流程；适配器管读写；调度器、状态存储和监控负责持续运行。The Skill organizes work, adapters read and write, and a scheduler, persistent state and monitoring sustain the loop.
输入：域名 + 已连接环境Input: domain + connected environment
domain、站点 / 存储适配器、账号权限引用、受众与目标、内容类型规范、来源策略、发布边界、预算与抽查配置。只给域名可以开始发现和诊断，完成接入后才能写入。Domain, site/storage adapter, permission references, audience/goals, content playbooks, source policy, publishing scope, budget and audit settings. A domain alone enables discovery and diagnosis; writes require completed onboarding.
持续状态Persistent state
站点档案、选题池、事实与来源表、样板、内容版本、发布记录、指标快照、实验与决策、例外队列。失败与恢复都可追踪。Site profile, topic backlog, claims/sources, samples, revisions, releases, metric snapshots, experiments/decisions and exception queue. Failures and recovery remain traceable.
输出：可验收的结果Output: verifiable results
上线内容与对应版本、质量报告、数据回顾、下轮行动、待人工处理事项。收入和增长只有在有真实证据时才报告。Live content and revisions, quality reports, performance reviews, next actions and human exceptions. Report revenue and growth only when supported by real evidence.
- 先选哪个站点试点？Which site should pilot first?
选择内容价值清楚、发布链路可控、反馈可观察的站点；最终选择待定。Choose a site with clear reader value, manageable publishing and observable feedback; final selection is open.
- 模板与接入契约如何验收？How are templates and adapters accepted?
用上面的文章优先默认栈起步；确认存储、导入、发布、缓存与回滚接口。界面语言与文章发布语言分别配置，不要求同步扩展。Start with the article-first default above; verify storage, import, publishing, cache and rollback interfaces. Configure interface and article languages separately; they need not expand together.
- 各站点真正的成功是什么？What is success for each site?
为每站确认一个主要完成事件，并保留曝光、点击、使用作为解释指标。Agree one primary completion event per site; retain impressions, clicks and use as diagnostic metrics.
- 哪些内容类型先自动化？Which content families go first?
分别确认受众、主题、形式、证据要求与样板。不能从一种文章通过推导出所有音视频都通过。Agree audience, topic, medium, evidence and samples for each family. Approval of one article family does not validate all audiovisual content.
- 如何验收、抽查与暂停？How do gates, audits and pauses work?
约定试样覆盖、质量阈值、抽查频率、敏感主题规则和恢复条件。Agree pilot coverage, quality thresholds, audit cadence, sensitive-topic rules and recovery conditions.
- 自动操作的范围与成本？What are the operation and cost limits?
确定发布频率、预算、并发、可自动更新 / 合并 / 退出的范围；分发渠道单独约定。Agree publishing cadence, budget, concurrency and permitted updates/merges/retirements; define distribution channels separately.
历史经验与资料History & sources流程教训与引用Process lessons & references
以下是从历史工作中提炼的流程经验，不是新的线上审计，也不证明流量或收益增长。These are general process lessons distilled from past work, not a fresh production audit or evidence of traffic or revenue growth.
引用与断言Claims and citations
引用存在不代表支持具体断言；逐条核对证据、适用范围与核验日期。A citation can exist without supporting a claim. Check the evidence, scope and verification date for each material claim.
发布链路Release path
源文件合并后可能仍需导入或刷新缓存；以正式 URL 的内容版本验收。A merged source may still need import or cache refresh. Verify the content revision at its live URL.
内容与派生版本Content and derivatives
为不同语言、图文和音视频保留版本关联；原始事实变化后检查派生内容。Track relationships between languages and media; check derivatives when underlying facts change.
编辑与并发Editing and concurrency
把修改理由、正反样板和写入顺序记录下来，减少重复纠正与发布冲突。Record edit reasons, positive and negative examples, and write order to reduce repeated corrections and publishing conflicts.
官方资料Official sources
官方资料核对：2026-09-17；接入时复核。本讲义不报告任何站点的实际流量或 ROI。Official references checked on 2026-09-17; recheck during integration. This note reports no site traffic or ROI measurements.
- Google · Creating helpful, reliable, people-first content有用、原创、面向读者的内容标准。Reader-first usefulness and original contribution.
- Google · Spam policies规模化低价值内容与链接垃圾政策。Scaled content abuse and link spam.
- Google · Build and submit a sitemapsitemap 发现提示与提交方式。Sitemap discovery hints and submission.
- Google · Search Analytics query表现维度、指标与返回数据限制。Performance dimensions, metrics and data limits.
- Google · Indexing API prerequisites仅指定招聘 / 直播页面类型适用。Restricted eligible job/live-video page types.
- Google · GA4 Data API basics读取已采集的分析数据。Reporting on collected analytics data.
- Google Ads · Generate historical metrics历史需求与广告竞争，不是 SEO 难度。Historical demand and ad competition, not SEO difficulty.
- IndexNow · FAQ通知、引擎独立处理及不保证收录。Notifications, independent processing, no indexing guarantee.
- OpenAI · Publishers and developers FAQ搜索发现与爬虫配置。Search discovery and crawler configuration.
- Anthropic · Web crawlers搜索、用户访问与训练爬虫的用途区别。Distinct search, user-access and training crawlers.
- Perplexity · Crawlers搜索发现与用户访问。Search discovery and user-directed access.
- Brave · Search API搜索与资料检索 API。Search and retrieval API.
- Tavily · Search API面向研究与检索的搜索接口。Search endpoint for research and retrieval.
- Exa · Search API内容搜索接口。Content search endpoint.
- Yandex · Webmaster站长平台入口；按市场选择。Webmaster platform; choose by market.
- 百度 · 搜索资源平台 / Baidu webmaster platform平台入口；具体能力在账号接入时确认。Platform entry point; verify account-specific capabilities during onboarding.
- Google · URL Inspection API只读 Google 已索引版本的状态，不做实时页面测试。Reads the version in Google’s index, without live-page testing.