Skip to content

Universal SEO skill for Claude Code. 26 sub-skills + 19 sub-agents covering technical SEO, E-E-A-T, schema, GEO/AEO, agent readiness (Lighthouse Agentic Browsing, WebMCP, llms.txt), backlinks, local SEO, e-commerce, international SEO, Google APIs, and PDF/Excel reporting. 9 optional extensions for live SEO data.

fix(verify): findings from the pre-release verification run

Agrici Danielff87fcee0734
diff --git a/CHANGELOG.md b/CHANGELOG.md
index fc38318..03041f3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -32,17 +32,18 @@ as third-party instead of being stated as fact.
- `pagespeed_check.py` reports `lighthouse_version`, the Lighthouse build
PSI actually ran (13.5.0 on 2026-09-29).
- Tests: `test_no_em_dash.py`, `test_file_size_limits.py` (SKILL.md 500
- lines, references 200), and nine new canonical-fact guards, each failing
- on the v2.4.0 wording it replaces.
+ lines, references 200), and eight new canonical-fact guards, each failing
+ on the v2.4.0 wording it replaces. The ledger host allowlist adds
+ `developer.chrome.com` and `static.googleusercontent.com` (the full QRG PDF).
### Changed
- Skills: the Search generative AI control covers Discover gen-AI features
and is separate from `Google-Extended`; `Google-Extended` covers Gemini
training and grounding and training of the models behind Search gen-AI
- features, with no effect on Search inclusion or ranking; the multimodal search type is a UI and
- export filter with no verified API `type`; local business queries in EEA
- aggregator and supplier units reach seo-ecommerce.
+ features, with no effect on Search inclusion or ranking; the multimodal
+ search type is a UI and export filter with no verified API `type`; local
+ business queries in EEA aggregator and supplier units reach seo-ecommerce.
- Re-verified and corrected: URL Inspection `mobileUsabilityResult` is
deprecated; the AI optimization guide dates (published 2026-05-15,
llms.txt clarified 2026-06-15, last updated 2026-07-10); JavaScript
@@ -51,14 +52,17 @@ as third-party instead of being stated as fact.
beta; I/O 2026 wording and summer rollouts marked as announced; Gemini
image model names and shutdowns (the `-preview` image IDs retired
2026-06-25, `gemini-2.5-flash-image` retires 2026-10-02); AI Mode 1B+
- monthly users is Google-stated; FTC
- review penalty unchanged for 2026; ChatGPT 1B+ weekly users; Ahrefs and
- Profound study dates; caniuse AVIF and WebP figures.
+ monthly users is Google-stated; FTC review penalty unchanged for 2026;
+ ChatGPT 1B+ weekly users; Ahrefs and Profound study dates; caniuse AVIF
+ and WebP figures.
- Third-party claims are labelled as such: update-table impact text (Google
confirms rollout dates, not targets), the Sterling Sky 18-day finding (a
case example, not a rule), local AI Overview coverage (0.14% and 68% shown
together with their methods), the June 2025 manual-action wave, and the
late-2024 publisher site-reputation reports.
+- The `/seo` orchestrator routes Google update-history questions to the
+ bundled ledger (`seo_updates.py`), so answers no longer depend on whatever
+ older copy the model finds on disk.
- Six reference files over 200 lines were split at section boundaries into
sibling files, with no content removed.
- Em dashes removed from skills, agents, scripts, extensions and docs.
diff --git a/agents/seo-geo.md b/agents/seo-geo.md
index ae38012..f356220 100644
--- a/agents/seo-geo.md
+++ b/agents/seo-geo.md
@@ -32,8 +32,9 @@ GPTBot is OpenAI's *training* crawler, not the ChatGPT Search crawler -- do not
its status as evidence about ChatGPT Search citability. Likewise ClaudeBot is
Anthropic's *training* crawler, not the Claude search crawler -- Claude-SearchBot
governs Claude search citability (per Anthropic's crawler support article).
-Google-Extended governs Gemini/Vertex training and grounding only, never Google
-Search or AI Overviews inclusion (those follow Googlebot), and Applebot-Extended
+Google-Extended governs Gemini/Vertex training and grounding, plus training of the
+models behind Search gen-AI features, never Google Search or AI Overviews
+inclusion (those follow Googlebot), and Applebot-Extended
governs Apple Intelligence training only, never Siri/Spotlight/Safari discoverability
(that follows Applebot). Check and report each bot against the specific capability
it governs.
diff --git a/skills/seo-agentic/references/access-policy.md b/skills/seo-agentic/references/access-policy.md
index 6634096..c94c882 100644
--- a/skills/seo-agentic/references/access-policy.md
+++ b/skills/seo-agentic/references/access-policy.md
@@ -23,7 +23,7 @@ citability rules live in `seo-geo`; this table adds what matters for agents.
| PerplexityBot | Perplexity | search index | honoured | perplexity.com/perplexitybot.json |
| Perplexity-User | Perplexity | user-initiated fetches | generally ignored | perplexity.com/perplexity-user.json |
| Google-Agent | Google | user-triggered agent fetches | generally ignored | developers.google.com/static/crawling/ipranges/user-triggered-agents.json |
-| Google-Extended | Google | Gemini training/grounding control token, no crawler; never Google Search | honoured | n/a |
+| Google-Extended | Google | Gemini training/grounding (and Search gen-AI model training) control token, no crawler; never Search inclusion | honoured | n/a |
Anthropic's help center previously said IP ranges were not published; as of
2026-08 they are at `claude.com/crawling/bots.json`. Do not IP-block
diff --git a/skills/seo-geo/SKILL.md b/skills/seo-geo/SKILL.md
index 670cd0f..760e4e7 100644
--- a/skills/seo-geo/SKILL.md
+++ b/skills/seo-geo/SKILL.md
@@ -165,7 +165,7 @@ Check `robots.txt` for these AI crawlers:
| CCBot | Common Crawl | Training data (often blocked) | yes |
| Bytespider | ByteDance | TikTok/Douyin AI | yes |
| cohere-ai | Cohere | Cohere models | yes |
-| Google-Extended | Google | **Gemini/Vertex training & grounding only** (NOT Google Search) | yes |
+| Google-Extended | Google | **Gemini/Vertex training & grounding, and training of the models behind Search gen-AI features** (NOT Google Search inclusion or ranking) | yes |
| Google-CloudVertexBot | Google | Site-owner-requested Vertex AI Agent crawls | yes |
| Google-Agent | Google | User-triggered agent fetches (agentic browsing for a user) | **no (user-triggered)** |
| Google-GeminiNotebook | Google | Fetches individual user-added source URLs (replaced `Google-NotebookLM`, supported until August 2026) | **no (user-triggered)** |
@@ -201,8 +201,9 @@ bot's robots.txt status** -- check them separately and report them separately.
| "Content can be used for Apple Intelligence training" | `Applebot-Extended` | `Applebot` |
| "Content is discoverable via Siri, Spotlight, or Safari search" | `Applebot` | `Applebot-Extended` |
-- **`Google-Extended` governs Gemini and Vertex AI training and grounding use only.
- It does not affect inclusion in ordinary Google Search, or in AI Overviews and AI
+- **`Google-Extended` governs Gemini and Vertex AI training and grounding use, and
+ training of the models behind Search gen-AI features. It does not affect inclusion
+ in ordinary Google Search, or in AI Overviews and AI
Mode, both of which are served from the `Googlebot` index.** Never score
`Google-Extended` as a "Google Search readiness" signal, and never cite a blocked
`Google-Extended` as evidence that a site is missing from Google Search.
diff --git a/skills/seo/SKILL.md b/skills/seo/SKILL.md
index afbc2bf..90eed05 100644
--- a/skills/seo/SKILL.md
+++ b/skills/seo/SKILL.md
@@ -100,6 +100,14 @@ When the user invokes `/seo audit`, delegate to subagents in parallel:
For individual commands, load the relevant sub-skill directly.
After any analysis command completes, offer to generate a PDF report via `"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run google_report.py`.
+**Google update history questions** (core, spam, policy, product changes; "is the
+spam update finished?"): answer from the bundled primary-source ledger, not from
+memory or other copies on disk. Run
+`"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run seo_updates.py --kind spam --json`
+(or `--since <yyyy-mm>`; add `--unverified` for third-party claims that are not
+confirmed). If the output reports the ledger as stale, check
+status.search.google.com before stating that a rollout is complete.
+
## Synthesis Methodology
Audits are not just findings, they are findings synthesized into a coherent
diff --git a/tests/test_canonical_facts.py b/tests/test_canonical_facts.py
index d3a445d..8d2893b 100644
--- a/tests/test_canonical_facts.py
+++ b/tests/test_canonical_facts.py
@@ -70,6 +70,9 @@ WRONG = [
(r"^\|[^|]*\d{4} (Core|Spam)[^|]*\|[^|]*\|[^|]*\|\s*Confirmed\s*\|\s*$",
"Google confirms rollout dates, not impact descriptions; label rows 'Dates confirmed'",
"status.search.google.com/products/rGHU1u87FJnkP6W2GwMi/history"),
+ (r"(?i)Google-Extended[^\n]{0,80}grounding( use)? only|training (&|and) grounding only\*\*",
+ "Google-Extended also limits training of the models behind Search gen-AI features",
+ "support.google.com/webmasters/answer/16908024"),
(r"canonical selection, mobile usability", "URL Inspection mobileUsabilityResult is deprecated",
"developers.google.com/webmaster-tools/v1/urlInspection.index/UrlInspectionResult"),
(r"Nano Banana 2 \(Flash, previous gen\)", "gemini-2.5-flash-image is the original Nano Banana",