internal-links

---
name: internal-links
description: Plans which of your pages should link to which, with the anchor text, and finds the pages nothing links to. Use on any site over about 15 pages.
---

# Point your own pages at each other properly

A link plan: which page links to which, with the exact anchor text, plus the pages nothing links to.

## Claude does

1. Asks for four inputs: your URL list (sitemap.xml is fine), the 5 to 10 pages that make money, the topic each should rank for, and any export of existing links. Without an export it plans from the URL list and says so.
2. Groups the URLs into hub-and-spoke clusters: one broad hub per topic, detailed pages as spokes. Anything fitting no cluster is listed separately, never forced in.
3. Sets direction: spokes link up to their hub, the hub links down to all of them, spokes link sideways only where a reader would follow.
4. Makes each money page a hub, or names the spokes pointing at it. A money page on zero inbound links is flagged first.
5. Writes anchors that describe the destination and vary. Never "click here", never one exact phrase sitewide.
6. Puts the link in body copy, high up. The first link to a URL in the source carries the anchor weight, so a navigation link firing first swallows yours.
7. Refuses to guess: no URL outside your list, no link claimed without evidence, no statistic it cannot source.

Output columns: source URL, target URL, anchor text, placement, reason.

## Then Codex does

Codex takes the mechanical half: fetching every page, extracting the links and anchors already there, counting inbound links per URL, finding orphans, spotting over-used anchors and checking which planned links already exist. Bulk fetching and counting, not judgement.

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

Fill in first: {SITE_ROOT} your domain, {URL_LIST_FILE} the URL list, {LINK_PLAN_CSV} Claude's plan, {MONEY_PAGES} the money pages.

## Claude checks

1. Every target is in your list and returned 200. A 404 or redirect kills the row.
2. No money page left on zero inbound internal links.
3. No exact anchor on more than three source pages.
4. The planned link is the first occurrence of that target in the body, not a repeat of a nav link.
5. Counts came from live fetches, not a template file. A file read is a lead, not a finding, and templates lie about what renders.

Failed rows go back to Codex with the reason and it runs again. A part-checked plan never reaches you.

## Rules

- Anchors describe the destination and vary. One phrase sitewide is the mistake this skill exists to stop.
- The first link in the source wins, so order beats count.
- Never quote a linking statistic from a source you could not open yourself.
- 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 crawls and counts itself, more slowly, and says so first.

## Built from

- inhouseseo/superseo-skills, https://github.com/inhouseseo/superseo-skills, 305 stars via api.github.com on 5 September 2026. Its first-link-weight-evidence.md holds the SearchPilot split-test figures and hub-and-spoke rules.
- seranking/seo-skills, https://github.com/seranking/seo-skills, 135 stars via api.github.com on 5 September 2026. Numbered workflows with an evidence folder and the anchor-diversity gate.
- Google Search Central, SEO Starter Guide, https://developers.google.com/search/docs/fundamentals/seo-starter-guide. Google's own anchor and linking guidance.

# Point your own pages at each other properly - Codex task

## You are given

- {SITE_ROOT} the domain, for example https://example.co.uk
- {URL_LIST_FILE} a sitemap.xml or a text file, one URL per line
- {LINK_PLAN_CSV} columns source_url, target_url, anchor_text, placement, reason
- {MONEY_PAGES} the URLs that must not be left on zero inbound internal links

## Produce

Four files in the working folder.

1. `crawl.csv` - one row per URL in {URL_LIST_FILE}: url, http_status, final_url, internal_links_out, nav_links_out, page_title. Fetch every URL, skip none.
2. `inbound.csv` - one row per URL: url, inbound_internal_links, the source pages linking to it, the anchor each uses. Count only links in the rendered body of a 200 response, and count each source page once per target.
3. `anchors.csv` - one row per distinct anchor already in use: anchor_text, times_used, target URLs. Sort by times_used descending.
4. `plan_checked.csv` - {LINK_PLAN_CSV} plus target_status, target_in_url_list, link_already_exists, existing_first_link_anchor (the anchor of the first link to that target already on source_url, else blank), anchor_reuse_count, and a verdict of OK, DEAD_TARGET, NOT_IN_LIST, DUPLICATE, ANCHOR_OVERUSED or NEEDS_HUMAN.

## Rules

- Fetch live pages. Never infer links from a template, a component or a CMS file. A template read is not evidence of what renders.
- Follow redirects but record both the requested and the final URL. A planned link into a redirect is DEAD_TARGET.
- Count internal links only. Other domains are out of scope.
- Keep site-wide navigation, header, footer and breadcrumb links out of the inbound counts, but record them in nav_links_out, because a nav link appearing first in the source changes which anchor counts.
- No more than two requests per second, with a plain user agent naming the site owner.
- Do not edit, publish or upload any page. Produce files only.
- Never invent a URL, an anchor or a count. Record a failed fetch as a failure rather than estimating it.
- Public information only, from the site you were given.

## Return

Print, in this order:

1. URLs fetched, and every one that did not return 200.
2. Every URL with inbound_internal_links of 0, headed ORPHANS.
3. Each URL in {MONEY_PAGES} with its inbound count.
4. The top ten rows of anchors.csv.
5. A verdict tally from plan_checked.csv, for example OK 41, DEAD_TARGET 2, DUPLICATE 6.
6. The four file paths.