page-brief

---
name: page-brief
description: Turns one search term into a writing brief that makes a single page rank for it, deciding the page type, sections and facts to cite before anyone writes.
---

# A brief that makes one page rank for one thing

One term, one page, the sections in order and the facts to cite. A writer follows it without questions.

## Claude does

1. Ask four things: the search term (one only), the page URL or "new page", what the business sells, the town or country served. Refuse a list of terms.
2. Open the live results. Read the top 3 non-ad pages that are not the client. Record each one's type (guide, service, product, comparison, listicle), H1, headings in order, word count.
3. Set the page type from what already ranks: if all top 3 are service pages, a blog post will not rank.
4. Build the sections from the headings all three share, then add the questions all three leave unanswered. That gap is the reason to publish. Google says heading order "doesn't matter" and there is "no magical, ideal amount of headings": order for the reader.
5. Write one title link: unique, term near the front, no boilerplate, no repeated terms. Google calls "Foobar, foo bar, foobars" spam-like.
6. List 3 to 6 facts to cite, each with a source URL and a date. The GEO paper (KDD '24) measured AI-answer visibility rising "up to 40%" for pages carrying statistics, quotes and citations. With no real figure, write "no verified figure found".
7. Write one self-contained answer block of 134 to 167 words, near the top, standing alone if quoted.
8. Never guess search volume, competitor traffic, conversion rate or an unopened statistic. Mark those "user to supply".

## Then Codex does

Codex takes the mechanical half: section word budgets from the competitor counts, the banned-vocabulary sweep, the internal-link shortlist from the sitemap, and assembly into one Markdown file. Bulk text handling, not judgement.

```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```

Fill in first: {SEARCH_TERM}, {PAGE_TYPE}, {SECTION_LIST}, {COMPETITOR_WORD_COUNTS}, {FACTS_WITH_SOURCES}, {TITLE_LINK}, {ANSWER_BLOCK}, {SITEMAP_URL_OR_PAGE_LIST}.

## Claude checks

1. Count the answer block. Outside 134 to 167 words, reject and rerun.
2. Open every source URL. A dead link or a figure not on the page is a reject. This pack's research lost a source to a network block; an unopened citation is not one.
3. Title link unique against the page list, no term repeated, page type matching what ranks rather than what the client prefers.
4. No density target or word-count minimum framed as a ranking claim. Google: "The length of the content alone doesn't matter for ranking purposes."
5. Nothing asks the writer to assert expertise to rank. Google, on E-E-A-T as a ranking factor: "No, it's not."

## Rules

- One term, one page. Two terms means two briefs.
- Word counts budget the writer, never claim a ranking effect.
- No keywords meta tag. Google does not use it.
- Every facts row carries a URL and a date, or it goes.
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.

## If Codex is not installed

Claude assembles the file itself and says so.

## Built from

- inhouseseo/superseo-skills, https://github.com/inhouseseo/superseo-skills, 305 stars via api.github.com: banned vocabulary.
- AgriciDaniel/claude-seo, https://github.com/AgriciDaniel/claude-seo, 16,344 stars via api.github.com: the 134 to 167 word block.
- Google Search Central, https://developers.google.com/search/docs/appearance/title-link: titles, headings, length.
- GEO, https://arxiv.org/abs/2311.09735, ACM SIGKDD 2024: citing statistics.

# A brief that makes one page rank for one thing - Codex task

## You are given

- {SEARCH_TERM} - one search term. The brief serves this and nothing else.
- {PAGE_TYPE} - guide, service, product, comparison or listicle, already decided.
- {SECTION_LIST} - the headings in reading order.
- {COMPETITOR_WORD_COUNTS} - three numbers, the top 3 ranking pages.
- {FACTS_WITH_SOURCES} - 3 to 6 rows: fact, source URL, date.
- {TITLE_LINK} - one title, already written.
- {ANSWER_BLOCK} - one paragraph, already written.
- {SITEMAP_URL_OR_PAGE_LIST} - the site's own pages, for internal links.

## Produce

One file, `brief.md`, in this exact order:

1. `# Brief: {SEARCH_TERM}` then a table: Page type, Title link, Word budget.
2. Word budget: the median of {COMPETITOR_WORD_COUNTS} rounded to the nearest 50, with the three inputs printed beside it. Label it "budget for the writer, not a ranking target".
3. `## Answer block` - {ANSWER_BLOCK} verbatim, with `words: N` underneath.
4. `## Sections` - each heading from {SECTION_LIST} with a per-section budget, split in proportion to the sub-points each carries. Give the answer-block section at least 150.
5. `## Facts to cite` - a table of {FACTS_WITH_SOURCES}: Fact | Source URL | Date. Add no rows.
6. `## Internal links` - up to 5 pages from {SITEMAP_URL_OR_PAGE_LIST} whose slug shares a word with {SEARCH_TERM}. Print the URL and a descriptive anchor for each. Never "click here" or "read more". Never repeat an anchor.
7. `## Language flags` - scan {SECTION_LIST}, {TITLE_LINK} and {ANSWER_BLOCK} for these and list every hit with its location: delve, leverage, robust, seamless, elevate, unlock, harness, embark, navigate the landscape, in today's fast-paced world, it's important to note, when it comes to, game-changer, cutting-edge, revolutionise, empower, tapestry, testament, realm, foster, pivotal. Also flag any heading starting "The Importance of" or "Understanding".

## Rules

- Copy {ANSWER_BLOCK}, {TITLE_LINK} and {FACTS_WITH_SOURCES} exactly. Change no wording.
- Invent no fact, figure, source or extra section.
- Open no URL. Work only from the text handed to you.
- If an input is missing or empty, print `MISSING: <name>` in that section and continue. Do not fill the gap.
- No keyword density target, no minimum word count, no keywords meta tag anywhere.
- British English. No em dash characters.

## Return

Print to stdout after writing the file:

- the absolute path of `brief.md`
- `answer block words: N`
- `word budget: N (from A, B, C)`
- `sections: N`
- `facts rows: N`
- `internal links: N`
- `language flags: N`, then one line per hit as `word - location`
- any `MISSING:` lines