agentleFS
fabio-bergmannfabio-bergmann/seo-geo-skills/skills/seo-page-writer/SKILL.md

seo-page-writer

Skill1 starsChanged 13 days agoMITView on GitHub

Decides whether a page is worth writing, then writes it to rank on Google and be cited in AI answers. Runs five gates (fit, search intent from the live results, a duplicate on the site, the page's information edge, worth it now) with a skeptic go, hold or kill, then drafts answer-first with sourced, dated facts and checks the draft before it ships. Covers blog posts, comparison (X vs Y), alternatives, best-of lists, integration and use-case pages. Use when the user asks to write an SEO article, blog post, comparison page, alternatives page or listicle, plan content for a keyword, write a content brief, or create a page for AI search or GEO. To change a page that's already live, use seo-page-refresh.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read references/ground-rules.md first.

    line 14
  2. Very specific:

    Run the topic through these in order.

    line 18
  3. Very specific:

    Before you write, in seo/briefs/{slug}.md or the ledger row:

    line 30
  4. Very specific:

    Nothing invented: Every number, quote and claim about another company comes from a source you read, with its date.

    line 45
  5. Distinctive:

    Ask the owner before you publish, unless they've said you may.

    line 59
  6. Distinctive:

    Add at least 3 internal links in from related pages, with anchors that say where they go.

    line 63
  7. Distinctive:

    Ask for indexing in Search Console, and send IndexNow for Bing.

    line 65
  8. Distinctive:

    Leave it alone until its kind of change is due (seo-page-refresh).

    line 66

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SEO Page Writer

Writes one page that can rank on Google and be cited in AI answers, after deciding it deserves to exist. Most pages that fail were never worth writing, so the gates come first.

Read references/ground-rules.md first. Page types and the evidence each needs: references/page-types.md. The keys in brackets are in references/sources.md.

1. The gates

Run the topic through these in order. The first one it fails is the reason. Write the outcome to seo/ledger.md.

  1. Fit. Would someone searching this ever buy the product? Read the top 10, not just the keyword: a shared word isn't a shared field. Grade it 0–3: 3 the product is the answer, 2 the searcher has the problem, 1 adjacent, 0 another field. Drop 0, hold 1.
  2. Intent. Read the live top 10 (the type, format and angle of each page) and the AI answer for the same question. Pick the page type from what ranks and what gets cited. A list only when the query asks for one: explicit "best" or "ideas" searches had a list in the top 3 54.5% of the time, implicit ones 10.2% (PPC-LISTICLES). If the top 10 is all service or product pages, an article rarely ranks there (CHATSEO-TEAM).
  3. Duplicate. Does a page on the site already answer this question, in any wording? Check seo/pages.md, the sitemap and Search Console's queries per page. Then it's a section of that page or a refresh (seo-page-refresh), never a twin. Two keywords belong on one page when Google ranks several of the same URLs for both (KWI-CLUSTERING). A review, an alternatives page and a comparison for one rival are three different questions.
  4. Edge. Name, in one line, what this page will say that the top 10 doesn't: the brand's own data, a dated primary source, a real comparison, first-hand product detail. No edge, no page. Many pages that add little to no value are what Google's scaled content abuse policy targets, however they're made (GSP), and Google's raters judge value against other pages on the same topic (G-QRG).
  5. Worth it now. Demand (keyword volume as a rank, impressions, a buyer prompt where the brand is missing), winnability against the site's authority (a young or weak site goes for long-tail and comparison searches first, CHATSEO-TEAM), and focus: cover at least 3 of a topic's 5 buyer questions before you start the next topic (SEMRUSH-TOPICAL). Comparison, alternatives, pricing and "best X" pages come before definitions.

Then Go, Hold or Kill, with the reason. Be the skeptic: if more than a third of what you look at gets Go, you're cheering, not deciding (CHATSEO-TEAM).

2. The brief

Before you write, in seo/briefs/{slug}.md or the ledger row:

  • The one question, the main keyword, and the other wordings that become sections.
  • What the top 5 results and the AI answer all say, and what they all miss.
  • The sub-questions an AI model would search on the way to its answer, not only the keyword: 95% of Gemini 3's fan-out queries had no search volume (SEER-FANOUT).
  • The edge, with its source.
  • The evidence for every claim about another company (references/page-types.md, Evidence).
  • An outline that isn't the shape of the last page you wrote. Practitioners read the 2026 spam updates as hitting pages built from one template, though Google named no target (LILYRAY-SELFLISTS, CHATSEO-TEAM).
  • At least 3 pages on the site that will link in, and where this page links out to next.

3. Write

  • Answer first. Under every question heading, 1–3 sentences that answer it and make sense lifted out alone, then the detail. Cited pages opened with a direct answer 20% of the time, against 5% for top-10 pages the AI passed over (TJR).
  • Name things. The brand, the tools and the rivals by name, and the entity named in the first sentence of each section, not "it". Show a visible date and one hard number or new claim in each section. Passages AI answers quoted had these more often than passages they used without credit, in one small sample (AWR-PASSAGES).
  • The strongest evidence early. Put the brand's own data and the freshest facts in the first 30% of the page: 44% of ChatGPT citations come from there (INDIG-FIRST3RD). Define each number where it first appears, and give the method in one line ("90 answers from three AI models, 1–30 October 2026").
  • Nothing invented. Every number, quote and claim about another company comes from a source you read, with its date. Write [SOURCE NEEDED] where you have none, then find it or cut the claim.
  • The product, once. A blog post pitches the product in one place at most, worded freshly. Comparison, alternatives, integration and use-case pages are about the product by nature, and say so in a disclosure line.
  • Real freshness. The year goes in the title only when the content is current and the results show years. Adding the month and year cost clicks in one split test (SEMRUSH-TITLEYEAR).
  • Substance over format. Formatting-only edits barely change what gets cited (SPRINKLR).
  • Plain semantic HTML: headings, paragraphs, real tables and lists, with the content, prices and dates in the HTML, never loaded by a script: not every bot runs JavaScript (G-JS).

4. Check before it ships

  • Every claim has its source, and every [SOURCE NEEDED] is resolved or cut.
  • Against the brief: the question answered in the first lines, the edge on the page, the evidence a page type needs.
  • What changed: when you rewrote a draft or started from a template, list every element you removed or changed, and put back what you weren't asked to change. In one practitioner's account, an agent silently removed a page's tables, and another dropped its internal-linking step (CHATSEO-TEAM).
  • The basics: one title, one H1, a meta description, the canonical, the date, the disclosure where the page needs one.
  • For the owner: what you're unsure about, every claim you couldn't source and the three sentences you're least sure of. Never empty.

Ask the owner before you publish, unless they've said you may.

5. After it goes live

  • Add at least 3 internal links in from related pages, with anchors that say where they go. Google asks for at least one (G-LINKS); 3 is our floor.
  • Record the page in seo/pages.md and the change in seo/log.md.
  • Ask for indexing in Search Console, and send IndexNow for Bing.
  • Leave it alone until its kind of change is due (seo-page-refresh).

With Sightkick

With Sightkick connected, the duplicate check, the results page, the draft's checklist against these rules and the publishing schedule are one call each. The sightkick skill has them.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.