RankRentAI Classroom
Lessons (8)

0/8 complete

  1. What actually makes a local site rank (and why "just write 50 articles" is wrong)
  2. Getting found without Search Console (and why we skip it)
  3. Citations: buy them, but check every one before you pay
  4. IndexNow — re-indexing when you add pages (and when it fails)
  5. Mining expert YouTube channels for ranking and selling playbooks
  6. Folding a distilled playbook into the skill so it actually gets used
  7. Folding in a source whose tactics are mostly rejected
  8. Building your own SEO skill — blue links, AI Overviews, and ChatGPT answers in one playbook
← All modules

Building your own SEO skill — blue links, AI Overviews, and ChatGPT answers in one playbook

  • C:\Users\61479\.claude\skills\general-seo\
  • skills-dist/general-seo.skill
  • buildinpublic-youtube/transcripts/
  • indexsy-youtube/transcripts/

What you're doing and why

Google rankings are no longer the whole game. Your customers also ask ChatGPT and Gemini who to hire, and Google itself now answers with AI Overviews above the normal results. We had thousands of expert videos already downloaded and transcribed, so we distilled them into one reusable "skill" — a playbook document set that any agent (or you) can load to do modern SEO: classic ranking, AI Overviews, and getting ChatGPT to cite and recommend a site. Because it's a separate skill from the rank-and-rent one, you can use it for ANY website or client, not just our sites.

The exact steps

  1. Ask your agent: "Use the general-seo skill — I want to [rank this page / recover traffic lost to AI Overviews / get ChatGPT to recommend my business]." The skill routes to the right playbook automatically.
  2. To see the actual searches ChatGPT runs behind your question (the "query fan-out" — the trick that makes the whole AI playbook work): ask your question in ChatGPT, wait for "Searching the web", then right-click → Inspect → Network tab → paste the part of the URL after /c/ into the filter box → refresh → open the request → Response tab → Ctrl+F search_model_queries. Those are the REAL queries — build your page titles and headings around them. (On the newest default model, switch the model picker to GPT-4o/o3 first — the data is hidden on the default.)
  3. Free first-party data before buying any tool: Bing Webmaster Tools → AI Performance report shows the real "grounding queries" AI ran against YOUR site. Submit every site to Bing via IndexNow after every publish — ChatGPT's searches draw from Bing's index.
  4. Monthly habit: ask ChatGPT/Perplexity about your niche — are you cited, or only competitors? Paste a paragraph of your content in — does the AI treat it as source material? Log the answers.

What it looked like for us

  • Skill lives at C:\Users\61479\.claude\skills\general-seo\ with 5 playbook files; a shareable copy is packaged at skills-dist/general-seo.skill (validated).
  • We verified the fan-out trick twice in our own transcript vault — two different experts demonstrate it on camera (Build in Public episodes ZXR1HvUU1kI and zKdjde0JYwk) — then confirmed the 2026 field names and the GPT-5.3 workaround against current web sources.
  • Real test results captured in the skill: an article with a genuine expert quote was cited by AI Overviews within ~2 hours; a controlled 6-month study showed LocalBusiness schema moved a business up ~3 spots in ChatGPT's local answers (the ONLY schema tactic with real evidence); one site lost 50% of its Google clicks to AI Overviews in 3 days — and recovered via YouTube.
  • The skill includes a "contradiction ledger" for the 8 topics where our expert sources disagree (schema, indexing services, Reddit, AI content, PBN listicles) so you get the reconciled position, not whichever guru spoke last.

Gotchas

  • The newest ChatGPT model hides the fan-out data. The DevTools trick stops working on the default model — switch to GPT-4o/o3 in the model picker, or use the API. This stuff changes constantly; re-verify before relying on it.
  • Fan-outs drift. Run the same prompt twice and you get different sub-queries. One capture is a sample, not truth — capture 5–10 times over several days and keep the combined list.
  • Never ask ChatGPT what it searched. It doesn't know and will confidently make something up. The only honest answer is in the network traffic or the API response.
  • Gray-hat AI tactics (PBN listicle citations, fake "research reports") work right now — citations within 24 hours, #1 answers in 10 days in one demo — but they're boom-and-bust. They're documented in the skill WITH risk labels; don't attach them to a domain you can't lose.
  • Watch the vanity metrics. "AI visibility score: 3%" means nothing by itself. Tie AI visibility to leads and revenue ("another lead from ChatGPT") or you're buying the grift.

Update (same day): internal linking added

The first version only mentioned internal links in passing. A full playbook section is now in the skill (traditional-seo.md §5), distilled from five more expert episodes. The short version students should remember:

  • Google judges a page's importance by the links pointing at it — including your own. "If your own site doesn't think a page is important, why should Google?"
  • Every page you want ranked should be reachable within 3 clicks of the homepage, and every new page should get links from older pages the day it publishes (Forbes' process).
  • New sites can manufacture authority without backlinks: rank an easy page, link it to the next new page, and keep aiming the accumulated authority at harder keywords (the "colony" method) — then point those ranking pages' links at your money pages.
  • Don't turn pages into a sea of blue links — relevant beats many. And changing every internal link won't move an established ranking to a new URL; Google locks rankings to the page with history (retire/redirect the old page instead).

The skill's two biggest holes are now filled. Two new reference files shipped, distilled from two new channel scrapes (Matt Diggity — 256 transcripts; Crawling Mondays/Aleyda Solís — 127 transcripts) plus the vault's unmined link episodes.

link-building.md — the off-page authority playbook. What to remember:

  • A link only counts if the page it's on can rank. Vet every placement: ≥1,000 visits/mo, real traffic from YOUR country (judge absolute numbers, not percentages), followed in-content links, and the page itself must have internal links pointing at it (an orphaned placement page is a link-farm tell). That's the core of Diggity's 15-point checklist (P_8TFXHIDzY) — run it before paying anyone for anything.
  • The single best link target is the page already ranking #1 for your keyword.
  • Anchor text: forget "% rules". Average the anchor spreads of the top 5 rankers for your keyword and sit inside that. Never use the same target anchor twice. "Branded" only counts if there's zero keyword overlap.
  • New sites: social profiles first, then 1–3 guest-post links/week across homepage + money + supporting pages. 1,000 links in month one = penalty bait. Keep link insertions under ~30–40% of monthly gains.
  • Parasite SEO and PBNs are documented WITH risk labels — disposable-asset plays, never on a domain you can't lose.

technical-seo.md — the infrastructure layer, and the reason "AI optimization" fails on broken sites: if you're not indexable, you're invisible to chatbots too. What to remember:

  • Googlebot never clicks. Anything behind a "load more" button, a tab, or click-to-paginate is invisible. Test your own site: disable JavaScript in Chrome and see what survives.
  • Rendering JavaScript spends YOUR crawl budget — heavy JS sites get less of their site indexed, full stop.
  • Migrations fail because SEO is called in two weeks before launch. Triage = pull the money URLs (GSC/analytics top pages), redirect those first, and SAVE a full crawl with HTML of the old site before touching anything. Verify every Search Console variant including the CDN subdomain serving your images/JS — "your www is ~10% of what Google fetches".
  • Faceted navigation (filters): sort orders never get indexed; attribute filters (color/size/material) get their own indexable page ONLY if the query has search volume AND the page shows at least 3–5 products. Amazon beats Macy's on "black leather backpack" for exactly this reason.
  • hreflang: en-gb with a dash, gb not uk, every page declares itself AND all its variants, and every variant self-canonicalizes. Hunt misalignments in Search Console's country tab (France page pulling Canadian traffic = fix it).
  • Free CRO win for any store: take your top internal site-search queries and check the right products actually come back. Days, not months.

Gotchas from the scrape itself (for the next channel pull): yt-dlp's --print writes a literal \t (backslash-t), not a tab — it silently breaks pipe-delimited parsing and, on Windows, the backslash becomes a folder separator that nests caption files inside per-video directories (and long titles truncate files to 0 bytes). Parse with Python, flatten before resuming. 4 of 134 Crawling Mondays videos have no captions at all (including the pagination and WordPress SEO episodes) — they're noted as gaps in sources.md.