citlyze

aeo-page-audit Skill

审计任意网页的答案引擎优化(AEO)就绪度——AI 爬虫能否访问它,AI 引擎能否从中提取答案。当用户问"这个页面符合 AEO 标准吗"、"为 AI 搜索审计这个页面"、"为什么 AI 引擎不引用我的页面"或"检查 AI 爬虫访问权限"时使用。可独立使用,无需账号。

安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。

查看源码

技能指令原文(SKILL.md)

AEO page audit

Grade a page on the two things that decide whether AI engines use it: crawlability (can AI bots fetch it) and extractability (can an engine lift a clean answer from it). Scored out of 100 using the same public checklist as the free Citlyze AEO grader.

What you need

Only fetch tools (WebFetch, curl, or equivalent). Fetch raw HTML and judge it as served: AI crawlers generally do not execute JavaScript, so content that only appears after hydration does not count.

Checks and scoring

Fetch https:///robots.txt, the page HTML, and probe https:///llms.txt, then score:

| # | Check | Max | Pass condition |
| --- | --- | --- | --- |
| 1 | AI crawler access | 25 | robots.txt does not block major AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bingbot, CCBot, and similar) for this path. Any blocked major crawler scores 0. |
| 2 | Reachable and indexable | 15 | HTTP status below 400 and no noindex in meta robots or X-Robots-Tag. |
| 3 | One descriptive H1 | 10 | Exactly one H1 with at least 3 words. Multiple H1s: half credit. None: 0. |
| 4 | Question-shaped headings | 12 | At least 2 subheadings phrased as questions users would ask. Exactly 1: half credit. |
| 5 | Extractable answer blocks | 15 | At least 2 headings immediately followed by a direct 25-90 word answer. Exactly 1: half credit. |
| 6 | Entity structured data | 10 | JSON-LD declaring who/what the page is about (Organization, Product, Article, or similar entity types). |
| 7 | Tables and lists | 8 | Content uses tables or lists AI engines can lift verbatim, where the content type calls for them. |
| 8 | Title and meta description | 5 | Title at least 15 characters and meta description at least 50, both describing the page honestly. |
| - | llms.txt | 0 | Informational only: note whether it exists. There is no strong evidence engines consume it yet, so it earns no points. |

Report format

  1. Score: N/100 with a one-sentence verdict.
  2. Check table: each check with pass/warn/fail, points earned, and a one-line finding quoting the evidence (the robots rule matched, the H1 text, the failing heading).
  3. Fix list, ordered by points recoverable: concrete edits ("add a 40-word direct answer under 'How does X work?'", "unblock GPTBot for /blog/*"), not generalities.
  4. If several pages were audited, add a comparison table.

Judgment calls

  • Blocking AI crawlers may be a deliberate policy; if the site clearly intends it, say so instead of calling it a defect.
  • Question headings and answer blocks are about genuine user questions and honest answers, not keyword stuffing; flag content that games the shape but says nothing.
  • For sites where content loads only via JavaScript, report it as the primary finding: fix that before polishing anything else.

Prefer running this online? The hosted AEO grader runs the same checks on any URL for free. To track how your pages actually perform inside AI answers over time, that's what Citlyze does.