Jonathon's AI Wiki

AI SEO Pre-Publish Checklist (20-Point, 5-Phase Gate)

9 min read

Source: raw/The_AI_SEO_Checklist_I_Run_Before_I_Publish_Any_Page_2026.md

A YouTube creator’s 20-point checklist run as a gate before publishing any page, organized into five phases: agentic readiness, page architecture, content, technical, and submit/measure (plus a maintenance step). Most of the checklist restates AEO/GEO and technical-SEO advice already covered elsewhere in this wiki — the reason to file it is the agentic-readiness phase, which adds a concrete, reusable verification tool the wiki did not previously have: Google’s new Lighthouse “agentic browsing” audit (in Chrome Canary), which scores a page 0-4 on agent-readiness by checking for registered Web MCP tools and an llm.txt file. Treat the checklist as an operational gate; treat the agentic-readiness phase as early and partly unproven (the creator’s own caveat).

Creator: not named in transcript (references a paid SEO/content community) · URL: https://www.youtube.com/watch?v=rssboMxtgW0 · Platform: YouTube · fetched 2026-06-12

Key Takeaways

  • The novel kernel is verification, not advice. The one thing here that is concrete, reusable, and absent from the rest of the wiki is the Lighthouse agentic-browsing audit: Chrome Canary → DevTools → Lighthouse → enable the agentic browsing category → analyze page load → aim for 4/4 agentic-browsing readiness. It checks two things the wiki only discussed conceptually before — whether the site registers Web MCP tools and whether an llm.txt file is present. This turns “the web is becoming agent-readable” (agent-readable-web) into a self-serve, checkable pre-publish step.
  • Web MCP as a site-owner move. Registering typed tools (e.g., list_topic, an OpenAI-RAG endpoint) lets search agents submit a form or fetch data without screen-scraping the page. The creator did this on an Astro site via Claude Code; the method is CMS-dependent and very new. The wiki’s webmcp-directory catalogs who exposes such tools; this source is the first to frame verifying your own page’s agent-readiness.
  • Most of the list is already covered. The content-side items — H2-as-question “content capsule,” front-loading the answer, backing every claim with a number and an inline source link, author bio + Person schema — are restatements of the Answer Capsule / fact-density / E-E-A-T pattern already in seo-patterns-learned and the Google generative-AI-search guide. The technical half (page speed, WebP images, noindex hygiene, title/meta, internal linking, GSC/Bing submission, content refresh) is generic SEO covered operationally by seo-audit-skill and gsc-autonomous-seo.
  • Page architecture: one page, one intent. For local-business/service sites, split a single “services” page into one hub page plus an individual page per service (e.g., sewer-line repair, emergency plumbing) so each target keyword has “breathing room.” Schema is page-specific (organization, local business, website, breadcrumbs, service) and only needed on the most important pages, not every page.
  • Bing submission is an AI-visibility move. The creator submits to Bing Webmaster Tools specifically because ChatGPT draws on the Bing index — so Bing indexing raises the odds of being surfaced in ChatGPT. ^[inferred — the source asserts the Bing→ChatGPT link without citing it]
  • Caveat the source flags itself: agentic readiness is “still very, very early.” The llm.txt recommendation initially looked like it conflicted with this wiki’s empirically-grounded skeptical position — resolved 2026-07-02 (see callout below): both are true on different axes. No study anywhere has found llms.txt affects AI-citation ranking (Google’s official guide now says so explicitly too), but Google’s own Chrome Lighthouse “agentic browsing” audit independently confirms a real, narrow, non-ranking benefit — mainly for AI coding agents reading documentation, not for AI search visibility.

llm.txt / LLMs.txt — required pre-publish vs. overhyped tactic

Existing claim: (from ai-citation-ranking-factors-zyppy and conductor-2026-aeo-content-marketing-trends) — Zyppy’s meta-analysis ranks LLMs.txt #23 of 23 ranking factors (2.0/10, the most-overhyped 2025 tactic); Conductor advises “skip the LLMs.txt push for the next two quarters.” New source says: (from raw/The_AI_SEO_Checklist_I_Run_Before_I_Publish_Any_Page_2026.md) — “this file… must be present on your website”; the argument that it’s useless “can stop because they actually do help.” Resolution: different axis, not a live dispute — confirmed by dedicated research after the original ingest, not just a re-read of the two sources.

  • The ranking/citation axis is more settled now, not less — and both sides agree. Google’s official AI-search guide was updated (2026-06-15) to state outright that llms.txt “won’t harm (nor help) your visibility or rankings in Google Search, as Google Search ignores them” — matching Zyppy’s 2.0/10 exactly (ai-research/searchengineland-google-llmstxt-wont-help-rankings-2026-07-02.md). Asked about the tension directly in a later interview, Zyppy author Cyrus Shepard reaffirmed “no study [in the meta-analysis]… found any relationship between llms.txt and AI citation” (ai-research/buzzstream-cyrus-shepard-llmstxt-followup-2026-07-02.md). The existing claim stands, now with Google’s own guidance behind it too.
  • The new source’s narrower claim is also real, and explains where “must be present” comes from. Google’s Chrome team shipped an llms.txt check as part of Lighthouse 13.3’s “Agentic Browsing” category (May 2026), explicitly documented as an agent-discoverability signal, not a ranking factor (ai-research/chrome-devtools-llmstxt-lighthouse-audit-2026-07-02.md) — independently confirmed by Search Engine Land, Semrush, DebugBear, and accessiBe, not just the checklist creator’s say-so. Search Engine Land’s own framing: the two Google guidances “don’t directly conflict… because these audits focus on AI agents and browser tools, not Google Search rankings” (ai-research/searchengineland-google-adds-llmstxt-lighthouse-2026-07-02.md).
  • “Must be present” overreaches even on its own agent-readability terms. Ahrefs’s 137K-domain server-log study (published 2026-06-15, the same day as Google’s ranking clarification) found 97% of published llms.txt files get zero fetches ever, and the Lighthouse audit itself accounts for only ~1-in-1,000 of that already-tiny traffic (ai-research/ahrefs-llmstxt-137k-site-study-2026-07-02.md). The realistic beneficiary is AI coding agents reading developer docs (Claude Code was the #2 most frequent fetcher, after GPTBot) — not AI search/citation retrieval bots (1.1% of fetches) or shopping/booking agents. Google’s John Mueller made the identical point directly to Lily Ray: llms.txt is “not done for search,” more a “temporary crutch” for AI coding tools parsing documentation, and “for non-developer sites… if you check your logs, you’re not getting a lot of that [agentic] traffic at the moment.”
  • Net: both claims are true as narrowly stated, and are complementary, not conflicting. Skip LLMs.txt for citation/ranking ROI — unchanged consensus, now better-sourced. The ~2-minute CMS-plugin-toggle version is defensible “why not” hygiene for the Lighthouse agentic-browsing check and for documentation-heavy pages, but the checklist’s blanket “must be present” overstates the case for a typical local-business marketing page (this article’s own Phase 2 example), where the realistic llms.txt audience is the Lighthouse auditor and coding agents — not AI search engines or end-user agents. ^[inferred] Status: resolved (2026-07-02)

Try It

The full 20-point gate, by phase. Run it before a page goes live.

Phase 1 — Agentic readiness (the new part)

  1. Web MCP — register typed tools on your site so agents can act (submit a form, fetch data) without screen-scraping. CMS-dependent and new; the creator registered tools on an Astro build via Claude Code.
  2. llm.txt — add a machine-readable summary at the root domain. WordPress: toggle it on in Rank Math or Yoast. Cheap (~2 minutes) and harmless either way, but temper expectations: Ahrefs found 97% of published llm.txt files get zero fetches ever, and the realistic audience is AI coding agents and the Lighthouse auditor itself — not AI search/citation engines (resolved contradiction above; do it for the Lighthouse check and documentation-heavy pages, not for AI-citation ranking).
  3. Schema, done well — page-specific types (organization, local business, website, breadcrumbs, service) on your most important pages only. Verify with Google’s Rich Results Test; shortcut for which schema you need = ask a web-browsing LLM “what schema should this page have?” + the URL.
  4. Verify the phase — open the page in Chrome Canary → three-dots → More tools → Developer tools → Lighthouse tab → enable the agentic browsing category → Analyze page load. Aim for 4/4 agentic-browsing readiness (it confirms Web MCP tools are registered and llm.txt follows recommendations).

Phase 2 — Page architecture

  1. One page, one intent — split a catch-all services page into a hub page + one page per individual service so each keyword has room to rank.
  2. Content capsule — make the H2 the question a real user would ask, then answer it immediately (already covered: Answer Capsule).
  3. Front-load the answer — put the direct answer in the first ~10-30% of the page.
  4. Add real experience — concrete proof (“I tried X and got Y”); E-E-A-T experience, not just a clever prompt.
  5. Back every claim — no naked statements; pair claims with a number where possible.
  6. Link the source inline — cite to a high-quality source (the creator’s example: Ahrefs) on the contextual keyword, not dumped at the end.
  7. Author bio present — who wrote it and why trust them; link to an About page, and ideally add the author to the post’s Person schema.

Phase 3 — Technical

  1. Page speed — test in GTmetrix (target grade A) and PageSpeed Insights, on desktop and mobile; aim for < 2.5s.
  2. Image SEO — serve WebP (not PNG/JPEG; convert via CloudConvert) with descriptive alt text.
  3. Don’t block the bots — in Cloudflare’s AI-crawler controls, allow GPTBot and AI crawlers (unless you deliberately want them out).
  4. Indexability — confirm no accidental noindex; check the Pages/indexing report in Google Search Console.
  5. Title tags + meta descriptions — title carries the main keyword (+ business name if it fits); write the meta description as a sales pitch with a CTA.
  6. Internal linking — add contextual links between related pages to improve crawlability and indexability.

Phase 4 — Submit & measure

  1. Google Search Console — paste the URL and Request Indexing to speed discovery (requesting ≠ guaranteed).
  2. Bing Webmaster Tools — run URL Inspection and submit; Bing indexing raises the odds of ChatGPT surfacing the page.

Phase 5 — Maintain

  1. Revive old content — rewrite posts older than ~10-16 months with fresh research or new thoughts; don’t set-and-forget.

Graph View

Backlinks