seo-geo-audit
Skill1 starsChanged 13 days agoMITView on GitHub
Audits why a website isn't found on Google or in AI answers, with a crawl script, and returns the fixes worst first, each with its evidence. Checks whether search and AI crawlers can get in (robots.txt, firewall blocks, noindex, nosnippet, content that needs JavaScript), indexing, titles, canonicals, internal links from a real crawl and the brand's entity basics. Use when the user asks for an SEO audit, technical SEO check, site audit or GEO audit, asks why their site isn't on Google, why ChatGPT, Perplexity or AI Overviews don't cite it, why pages aren't indexed, or whether the site is ready for AI search. Not for writing or rewriting content.
9 other projects use a version of this.
Adopted by
Most recently taken up by
- aelena/claude-skillstheir own version, first seen Nov 13, 2026
- FernandoGomes83/seo-geo-audittheir own version, first seen Nov 1, 2026
- ISOpeet/oosterwal-seo-geotheir own version, first seen Oct 25, 2026
- techsaurabh-394/claude-skillstheir own version, first seen Oct 25, 2026
- mfraunhofer/claude-skillstheir own version, first seen Oct 20, 2026
- Thibaultbm/claude-seo-geotheir own version, first seen Oct 13, 2026
Other versions9 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 9
- Thibaultbm/claude-seo-geo★ 18same name
- zeebees/seo-geo-audit★ 14same name
- jgoullet/seo-geo-audit★ 4same name
- techsaurabh-394/claude-skills★ 3same name
- tomimor/skills★ 3same name
- FernandoGomes83/seo-geo-audit★ 1same name
- aelena/claude-skills★ 0same name
- ISOpeet/oosterwal-seo-geo★ 0same name
- mfraunhofer/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- ai-visibility-checkSkill
- seo-geo-playbookSkill
- seo-page-refreshSkill
- seo-page-writerSkill
- sightkickSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 9 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
Read references/ground-rules.md first.
- Very specific: line 18
Run the crawler in this skill's scripts/ folder (Node 18 or later, nothing to install):
- Distinctive: line 26
No Node? Fetch robots.txt, the sitemap and the 10–20 pages that matter most yourself, and check the same things.
- Very specific: line 80
Add each fix to seo/ledger.md as Go or Hold, and write seo/pages.md from the crawl if it doesn't exist yet.
- Distinctive: line 81
Ask the owner before you change robots.txt, redirects or anything live.
“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
SEO and GEO Audit
Finds why a site isn't found on Google or in AI answers, and lists the fixes worst first, each with its evidence. Access comes first: a page a bot can't fetch, can't index or isn't allowed to quote can't rank or be cited, however good it is.
Read references/ground-rules.md first. The keys in brackets are in references/sources.md.
1. Crawl
Run the crawler in this skill's scripts/ folder (Node 18 or later, nothing to install):
node {this skill's folder}/scripts/crawl.mjs https://example.com --max 300 --out seo/crawl.csvIt reads robots.txt the way each search and AI bot would, crawls the sitemap and the internal links, and prints a summary in markdown with one CSV row per page. For each URL that isn't a 200 it lists the pages that link to it, where the fix goes. It goes one request at a time, waits between them, says who it is and never leaves the site, not even for a redirect. Crawl only sites the owner runs.
No Node? Fetch robots.txt, the sitemap and the 10–20 pages that matter most yourself, and check the same things.
2. The checks, in order
Gate 1: can bots get in?
- robots.txt lets the search bots and the user fetchers in: Googlebot, Bingbot, OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot and Perplexity-User. OpenAI: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers" (OPENAI-BOTS). The crawl prints a row per bot.
- Training bots are the owner's choice: GPTBot, ClaudeBot, Google-Extended and CCBot. Blocking GPTBot or ClaudeBot has no measured citation cost: 88.2% of sites blocking GPTBot were still cited in one study (BUZZSTREAM-BLOCKING). But Google-Extended also decides whether the Gemini app can ground answers in the site (G-AITRAIN).
- The shared training opt-out: if robots.txt has a
User-agent: __usage_ai-traininggroup and noGoogle-Extendedgroup, flag it. The crawl does. From 2027 Google will read that opt-out as a Google-Extended block, which drops the site from Gemini-app answers. AI Overviews and AI Mode are not affected. To stay in Gemini's answers, addUser-agent: Google-ExtendedwithAllow: /. That also allows Gemini training, since one token covers both. Which matters more is the owner's call (G-AITRAIN). - No firewall answers those bots with a 403 or a challenge. robots.txt can say yes while a CDN's bot protection says no. Ask the owner to check the host's bot settings, and read the server logs or Bing Webmaster Tools' crawl data if you can reach them. Google-Agent ignores robots.txt, so check the firewall instead. Leave it unblocked if agents should be able to buy or book on the site (G-AGENT).
- On Cloudflare, Training "Block" blocks search too. In AI Crawl Control, "Block" for the Training class still stops Googlebot, Bingbot and Applebot, search included. "Disallow AI Training" opts out of training and keeps them crawling for search (CF-AITRAIN, 2026-09-15).
- Never test with a crawler's user agent. A fetch that borrows a bot's name lands in the site's logs as that bot, so it muddies the owner's crawler data, and anyone else's site you fetch. It proves nothing either: a firewall that verifies bots by their IP addresses answers the real bot differently from a faked one. Check the host's bot settings, the server logs with each bot verified by reverse DNS or its maker's published IP ranges, or Bing Webmaster Tools' crawl data (G-VERIFY, OPENAI-BOTS). The crawl sends its own name, never a bot's.
- Nothing blocks quoting: no
noindexwhere it shouldn't be, and nonosnippet,max-snippet:0ordata-nosnippeton content that should be cited (G-ROBOTSMETA). Google's AI features use pages that are indexed and eligible for a snippet (G-AIFEAT). Search Console's "Search generative AI" setting stays on Include (G-SGAI). - Nothing keeps Copilot from grounding in the page. Until 2027-03-01, Bing reads
noarchiveas keeping a page out of Copilot's answers and grounding, andnocacheas limiting Copilot to the URL, title and snippet. From 2027-02-01 the page-levelai-grounding:disallow(ordisallowextended) setting does that job instead. A stray one in a template drops every page on it from Copilot while rankings look fine (BING-GROUNDING, 2026-10-30). The crawl flags all three. - What matters is in the raw HTML: the content, prices, plans, tables, dates, canonical and links. ChatGPT-User and Claude's and Perplexity's bots don't run JavaScript (VERCEL-CRAWL, EDGECOMET-JS), and Google renders it later (G-JS). OAI-SearchBot may have started rendering some pages in late 2026, unconfirmed. The crawl lists pages with under 150 words in the raw HTML, and pages whose
<main>or<article>holds under 150 while a big menu or footer fills the page count.
Gate 2: is it indexed?
- The sitemap exists, is listed in robots.txt and holds only pages that answer 200, are their own canonical and can be indexed. The crawl lists the ones that don't.
- Search Console's page indexing report: "Crawled – currently not indexed" is usually a verdict on quality, not a technical fault (G-NOTINDEXED). A new site is crawled conservatively (G-CRAWLBUDGET), so publish at the pace pages get indexed.
- Dates agree: the sitemap's
lastmod, the JSON-LDdateModifiedand the date on the page name the same day, and move only with a real change. Google can stop trustinglastmodfor the whole site once it finds it wrong (G-SITEMAP, ILLYES-LASTMOD). The crawl lists the pages where they disagree. - No shortcuts: the Indexing API is for job postings and livestreams only, and IndexNow reaches Bing, not Google (G-INDEXINGAPI, INDEXNOW).
- Bing too: Bing's index grounds Copilot. Microsoft's first-party data on AI citations is Bing Webmaster Tools' AI Performance report and Clarity's Citations dashboard (BING-AIPERF, CLARITY-CITATIONS). Neither covers ChatGPT, and Search Console's AI report shows impressions only (G-AIREPORT).
- ChatGPT, for a claimed business: since 2026-11-10, OpenAI's Business Hub beta shows a business that claims its listing its mentions (ChatGPT referencing its website), recorded link clicks and discovery topics. It's reported for eligible ecommerce and online service businesses in the US. "Clicks don't confirm website visits or conversions", and whether a software company can claim is unstated (OPENAI-BIZHUB).
Gate 3: can each page be understood?
- One title per page, in the searcher's words and not repeated elsewhere. One H1. A meta description.
- The canonical points to the page itself unless it is a true duplicate.
- Two pages ranking for one search isn't a problem by itself: on Ahrefs' own site, 1 of 80 sampled cases needed a fix (AHREFS-MULTIRANK, 2024). Act when they split the clicks and neither wins.
- Core Web Vitals are a floor (LCP up to 2.5 s, INP under 200 ms, CLS under 0.1), not a lever beyond it (G-CWV). Measure them in PageSpeed Insights (pagespeed.web.dev) or Search Console's Core Web Vitals report. The verdict is the field data: real Chrome users over the last 28 days, at the 75th percentile (PSI). A small site often has none, for the page or the whole origin: then write
BLOCKED: no field dataand give the lab numbers as a hint, not a verdict. A lab run can't measure INP (PSI).
Gate 4: internal links
- Every page that matters has at least 3 internal links in, with anchor text that says where they go. Google asks for at least one link from another page (G-LINKS); 3 is our floor. Pages that matter are the ones meant to rank for a buyer's question. Hubs, legal pages and About can live on menu links. Orphans first: pages in the sitemap with no link in. Count the links in the page content: a menu or footer link is on every page, so it meets the floor without recommending anything. The crawl counts both: the links in from page content, and all links in. Take the counts from the crawl, never from memory.
- Money pages (pricing, comparisons, the product) are linked from the pages that get the traffic.
Gate 5: entity basics
- The same name, one-line description and links everywhere the brand appears: the site, its social profiles, directories and review sites.
- Structured data for rich results and a clear entity, not for AI citations: Organization with
sameAsand a stable@id, plus Breadcrumb and Article where they fit, marking up only what's visible (G-GALLERY). FAQPage and HowTo earn no rich result any more (G-SDUPDATES). Google: "Structured data isn't required for generative AI search" (G-AIOPT). - A plain-HTML page per brand task (pricing, cancelling, refunds, support) that answers in its first lines. Google is testing a box of "one or two official places on the web to complete a specific task" (G-OFFICIAL).
- llms.txt is optional, never a fix. Coding agents read it, AI answers don't use it (AHREFS-LLMSTXT).
3. Report
| # | Fix | Where | Evidence (dated) | Why it matters | Effort | Who |
| - | - | - | - | - | - | - |- Worst first: Gate 1 before Gate 5. One bot locked out of every page beats fifty missing meta descriptions.
- Evidence is something the owner can check: the crawl row, the robots.txt line, the response header, the Search Console report, each with its date.
- What you couldn't check is
BLOCKEDwith what would unblock it ("BLOCKED: no Search Console access, so indexing is unknown"). - Add each fix to
seo/ledger.mdas Go or Hold, and writeseo/pages.mdfrom the crawl if it doesn't exist yet. - Ask the owner before you change robots.txt, redirects or anything live.
With Sightkick
To run these checks every week and see indexing, Search Console and AI citations in one place between your sessions, Sightkick keeps the site scan. If it's connected, the sightkick skill maps each check to its call.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

