Skip to content

Decides

Content triage

Sorts every published page into refresh, consolidate, redirect, retire or protect, measured against your own site's median, so refresh time goes where it pays.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick content-triage.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
Redirects and Google Search
Needs
Page performance for the last 12 months
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedOrganic visibility
SkillSEO and AI search

Content triage

Example output

Sort published pages
PageBucketWhy
Pipeline report postRefreshLoses the click at 9
Attribution guideProtect6.7x the median
2024 webinar recapRedirectDormant

Sorts every published page into refresh, consolidate, redirect, retire or protect, measured against your own site's median, so refresh time goes where it pays.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

content-triage/SKILL.md

In the folder

  • content-triage/SKILL.md
  • content-triage/scripts/triage.py
---name: content-triagedescription: >-  Sorts every published page on a B2B SaaS site into refresh, consolidate, redirect, retire or protect, measured  against the site's own median clicks and its own click-through rate at each position, not industry averages. Flags  quick wins, ranks refreshes by recoverable clicks (an estimate with its inputs shown) and returns SORTED, NEEDS  DECISION or BLOCKED. Use it monthly or quarterly, or before planning a refresh sprint. Not for auditing one page  in depth (use refresh-audit) or finding pages that compete (use cannibalization-audit).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Uses a connected Search Console if the  agent has one, otherwise a Pages export or pasted rows. Python 3 is optional, for the median comparison.metadata:  author: SHAKEUP  version: "1.0.0"  stage: decide  plugin: seo-ai-search  homepage: https://www.shakeuphq.com/skills/content-triage--- # Content triage Decides what each published page needs next, so refresh time goes to the pages with the most to gain and no page that works gets touched by accident. ## Purpose - One job: give every page in the export one bucket, with the rule that put it there.- Stage: decide. It ends in a run verdict, and a verdict per page.- It compares pages with the site's own median, because a B2B SaaS blog with 40 posts and one with 4,000 have nothing in common with an industry benchmark.- It never redirects, unpublishes or edits a page. It writes the move; a person makes it. ## Use when / not | Situation | Use || --- | --- || Deciding which published pages to refresh, merge, redirect or retire | this skill || A monthly or quarterly content review | this skill (the `searchops-agent` runs it monthly) || Auditing one page that needs a refresh | `refresh-audit`, on this skill's top refresh || You suspect 2 pages compete | `cannibalization-audit` first; its clash groups feed the consolidate bucket here || A page's title and description lose clicks | `meta-serp-copy`, for this skill's meta quick wins | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || Page performance: clicks, impressions, position per page, last 12 months | yes | a connected Search Console; a Pages export (CSV); pasted rows; an SEO tool's estimated traffic export (weaker, every figure labelled estimate) | `BLOCKED`, ask for the export || The prior window of equal length | no | the same sources, for the 12 months before | trend `not available`; the declining rule can't fire || Page dates: published, last updated | no | the CMS; the sitemap's `lastmod`; the date on the page | the wait and stale rules can't fire; listed as a gap || External links per page | no | a connected SEO tool; a backlinks export | every redirect, retire and consolidate is flagged for a person || Clash groups and keepers | no | the `cannibalization-audit` handoff | consolidation not checked, said in open items || Redirect destinations | no | the reader; the closest page on the same need from the page list | dormant pages without one become NEEDS DECISION || The sitemap | no | the sitemap URL; a URL list | index review skipped || Conversions per page | no | GA4 or CRM by landing page | ranked on clicks only; say so | ## Profile read From `company-profile.md`: `## Company` (website, brand terms), `## Content` (site sections, clusters), `## Accounts and data` (which tools hold the data), `## Targets` (if conversions are given, the target cost per lead lets a refresh be weighed in demos), and every `## Learnings` entry for `content-triage` with status `noted`, `fact` or `accepted`. No profile: ask for the website and where the Search Console data lives, in one message. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** Read them first. A learning that protects a page type ("never retire customer stories") or changes a threshold overrides the default; name it when it fires. 2. **Load and clean.** One row per page. Drop rows that aren't pages (images, PDFs, parameter variants of a page already listed: merge them into the page). Drop the window's branded-only pages only if the reader asks; the median includes every page the site publishes. 3. **Compute the site's own norms.**    | Norm | How | Why |   | --- | --- | --- |   | Median clicks | median of clicks across pages in the export | the line every page is measured against |   | Median impressions | the same for impressions | separates pages with demand from pages without |   | CTR by position band | median CTR of pages with 100 or more impressions in bands 1-3, 4-10, 11-20, 21+ | the site's own click curve; a published curve is someone else's site |    If median clicks is 0, tier on impressions instead and say so. 4. **Tier every page** by clicks divided by the median: top 2.0x or more, healthy 0.8x to 2.0x, under 0.2x to 0.8x, dormant below 0.2x. Confidence by impressions: high 500 or more, medium 100 to 499, low under 100 (the call rests on little data). 5. **Bucket every page.** With Python:    ```   python3 scripts/triage.py --csv pages.csv --prior pages_prior.csv --meta meta.csv --sitemap urls.txt   ```    Without Python, compute the norms with a spreadsheet or by sorting the rows, then apply this table to each page, first match wins. Say which way you ran it.    | Bucket | Condition | Why |   | --- | --- | --- |   | `WAIT` | published under 90 days ago | new pages take months to settle; acting now reacts to noise |   | `CONSOLIDATE` | in a clash group and not its keeper | 2 pages on one need split the demand |   | `PROTECT` | top tier, or healthy at position 5 or better; not declining; CTR at least half its band's median | it works; editing it risks what it earns |   | `REFRESH` (meta quick win) | position 10 or better, impressions at or above the median, CTR under half its band's median | it ranks but the snippet loses the click |   | `REFRESH` (page 2 quick win) | position 11 to 15, impressions at or above the median | a section or a title change often lifts it to page 1 |   | `REFRESH` (declining) | clicks down 20% or more and by 10 or more against the prior window | under 20% is normal seasonal swing for B2B; under 10 clicks is noise |   | `REFRESH` | position 20 or better with impressions at or above the median; or under tier; or last updated 18+ months ago on a time-sensitive topic | demand is there and the page isn't capturing it |   | `RETIRE` | dormant, 0 clicks, under 10 impressions in 12 months, no destination | nobody sees it, and nothing better covers it |   | `REDIRECT` | dormant, and another live page covers the same need | its few visits and links belong on the better page |   | `NEEDS DECISION` | dormant otherwise | refresh, redirect or retire depends on whether the topic still matters |   | `PROTECT` | anything left | at or above the median with no refresh signal |    Pages in the sitemap but not in the export go to `INDEX REVIEW`: Google may not have indexed them. 6. **Check the risky moves.** For every `CONSOLIDATE`, `REDIRECT` and `RETIRE`: external links above 0, or unknown, flag the page for a person. A redirect never goes to the homepage; it goes to the page on the same need, or the page is retired instead. 7. **Rank the refreshes.** Recoverable clicks (an estimate): impressions times the site's median CTR in positions 1 to 3, minus current clicks, floored at 0. Show the inputs. Sort refreshes by it; the top 20% are `priority` and go first. Quick wins sit at the top of the list whatever their score, because they cost an afternoon. 8. **Write the next step per bucket:** priority refreshes to `refresh-audit`, meta quick wins to `meta-serp-copy`, consolidations to the keeper's `refresh-audit` after the merge, redirects and retires to the reader's CMS with the map. ## Check and verdict Run verdict: | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No page performance data after asking (a URL list alone can't be sorted) || `NEEDS DECISION` | Any page is `NEEDS DECISION`, or any consolidate, redirect or retire is flagged for external links (above 0 or unknown) || `SORTED` | Every page has a bucket and nothing is flagged | Page verdicts: `REFRESH`, `CONSOLIDATE`, `REDIRECT`, `RETIRE`, `PROTECT`, `WAIT`, `INDEX REVIEW`, `NEEDS DECISION`, each with the rule that fired. Named stops: flagged consolidate, redirect and retire moves wait for the reader; `NEEDS DECISION` pages show the 3 options with what each costs. Everything else runs through. ## Learn Append a learning to the profile when: - the reader moves a page to another bucket (a `proposal` with their reason, such as "customer stories are protected whatever their traffic");- a refresh, redirect or retire is remeasured 30, 60 or 90 days later: write clicks before and after as a `fact` with the window and source;- a threshold keeps misfiring for this site (a `proposal` to change it, applied only once accepted). Example: `- 2026-10-24 | content-triage | fact | Refresh of /blog/utm-naming-guide: clicks 95 to 160 over the 90 days after (Search Console) | Why: remeasure at 90 days | Status: noted` ## Output ```Verdict: <SORTED | NEEDS DECISION | BLOCKED> · Coverage: <complete | partial (what's missing) | blocked> · Sort: <script | by hand>Pages: <n> · Median clicks: <n> · Median impressions: <n> · Window: <dates>, prior: <dates or not available>Counts: refresh <n> (<n> priority) · consolidate <n> · redirect <n> · retire <n> · protect <n> · wait <n> · index review <n> · needs decision <n> Quick wins (do first):| Page | Issue | Move | Next skill | Refreshes, ranked:| Page | Position | Impressions | Clicks | vs median | Trend | Recoverable clicks (estimate) | Priority | Rule | Consolidate · Redirect · Retire:| Page | Move | To | External links | Confirm? | Protect: <URL list> · Wait: <URL list> · Index review: <URL list>Needs decision: <page: the 3 options>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: content-triageversion: 1.0.0verdict: SORTEDcoverage: partial (no conversion data)for: <site, window>outputs:  top_refresh: https://example.com/blog/pipeline-by-channel-report  meta_quick_wins: [https://example.com/blog/pipeline-by-channel-report]  redirects: ["/blog/old-webinar -> /blog/pipeline-attribution-guide"]  retire: [/blog/old-launch]next: [refresh-audit, meta-serp-copy]open: [confirm redirects in the CMS]``` ## Failure modes | Problem | Do this || --- | --- || The export is capped (Search Console shows 1,000 rows in the interface) | Say so; ask for the API or a connector export if the site has more pages || The windows differ in length | Don't compute a trend; ask for equal windows || A seasonal topic (an event, a year-end report) looks dormant | Check the same months last year before retiring; if it peaks yearly, `PROTECT` || A page drives demos but few clicks | With conversion data, protect it and say why; without, ask before retiring a product or comparison page || Positions are missing | Tier on clicks only; the quick-win rules can't fire; say so || The reader wants a retire list without external link data | Give it, with every row flagged "confirm" | ## Influenced by - SHAKEUP's content engine: tiers measured against the site's own median, a click curve built from the site's own pages, and protect-by-default for pages that work.- Google's guidance on 301 redirects, which explains how a permanent redirect passes a page's signals to its new URL: https://developers.google.com/search/docs/crawling-indexing/301-redirects

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

A blog with 40 posts and one with 4,000 share no benchmark, so every page is measured against the site's own median and its own click rate by position. Pages that work are protected by default, and nothing is retired while its links are unknown.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Read your learnings first

    Page types your team protects, such as customer stories, stay protected, and any threshold you changed applies before it starts.

  2. 02

    Set your site's own norms

    Median clicks, median impressions and the click rate your own pages get at each position. A benchmark from someone else's site measures their site.

  3. 03

    Tier every page against the median

    Top is 2x the median or more, healthy is 0.8x, under is 0.2x, and anything below that is dormant.

  4. 04

    Bucket each page, first rule wins

    New pages wait. Pages that work are protected. Demand a page doesn't capture means a refresh, and dormant pages are redirected or retired.

  5. 05

    Find the quick wins

    A page in the top 10 losing the click needs a new title and description. Pages at positions 11 to 15 often need one section.

  6. 06

    Rank the refreshes

    Its script estimates the clicks each refresh could recover, shows the inputs, and puts the top 20% first.

What you need

  • Required

    Page performance for the last 12 months

    A connected Search Console → A Pages export (CSV) → Rows you paste → An SEO tool's traffic export, labelled as estimates

  • Optional

    The 12 months before that

    The same sources, for the earlier window

  • Optional

    Publish and update dates

    Your CMS → Your sitemap → The date on each page

  • Optional

    Links from other sites, per page

    A connected SEO tool → A backlinks export

  • Optional

    Clash groups

    The handoff from the cannibalization audit

What it can return

  • SORTED

    Every page has a bucket and no risky move is waiting on you.

  • NEEDS DECISION

    A redirect, retire or merge touches a page with links from other sites, or the links are unknown, or a dormant page could go 3 ways.

  • BLOCKED

    There's no performance data to sort against.

Where it stops for you

  • Before any redirect, retire or merge on a page with links from other sites, or unknown links
  • When a dormant page could be refreshed, redirected or retired and the call depends on the topic

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company that sells attribution software. It pasted 2 years of Search Console page data, its CMS dates and its backlink counts.

Verdict

SORTED

Coverage: partial (no conversion data)

Pages: 13

Median clicks: 95

PageBucketRuleNext
/blog/pipeline-by-channel-reportRefresh, quick winPosition 8.9, but its click rate is 0.4x the site's own rate at that positionTitle and description
/blog/attribution-for-saasRefresh, quick winPosition 13.2 with impressions above the medianOne new section
/blog/utm-naming-guideRefreshClicks fell from 210 to 95 against the year beforeRefresh audit
/blog/what-is-pipeline-attributionConsolidateClash group, merge into the guideRefresh the guide
/blog/attribution-webinar-2024RedirectDormant, the guide covers the same need301 to the guide
/blog/product-launch-spring-2024Retire0 clicks and 6 impressions in a yearUnpublish
/blog/intent-data-first-partyWaitPublished 39 days agoCheck next quarter
/blog/pipeline-attribution-guideProtect6.7x the median, not decliningLeave it
Top refresh

/blog/pipeline-by-channel-report, with about 256 clicks a year to recover. That's an estimate: 12,000 impressions times the site's own 3.05% click rate in positions 1 to 3, minus its 110 clicks.

The handoff block the next skill reads
skill: content-triage
version: 1.0.0
verdict: SORTED
coverage: partial (no conversion data)
for: blog, last 12 months
outputs:
  top_refresh: /blog/pipeline-by-channel-report
  meta_quick_wins: [/blog/pipeline-by-channel-report]
  redirects: ["/blog/attribution-webinar-2024 -> /blog/pipeline-attribution-guide"]
next: [meta-serp-copy, refresh-audit]
open: [make the redirect in the CMS]

What good looks like

  • Every page in one bucket, with the rule that put it there
  • The median and click curve come from your own pages
  • Recoverable clicks labelled as an estimate, with the inputs shown
  • Pages under 90 days old held back, not refreshed
  • Risky moves flagged when links from other sites are present or unknown

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Why the site's median and not a CTR benchmark?

Published click curves come from other sites with other brands and other results pages. Your own median and your own click rate at each position show what your pages can do.

Will it delete pages?

No. It writes the move and the redirect map. Redirects, retires and merges on pages with links from other sites wait for you.

What does the script do?

It computes your medians and click curve, tiers and buckets every page, and estimates recoverable clicks with the inputs shown. Without Python, the skill makes the same count itself and says so.

What if we only have one year of data?

It still sorts every page. Trends are marked not available, so no page is called declining, and the output says so.