Skip to content

Decides

Topic precheck

Decides whether an article topic is worth writing before anyone spends a day on it, from overlap, cluster fit, real demand, your own evidence and a named author.

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 topic-precheck.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
Creating helpful, reliable, people-first content
Needs
The topic, in the buyer's words · A named author with a bio page
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedContent
SkillContent

Topic precheck

Example output

Is this topic worth writing?
CheckResultCall
OverlapNo page owns itClear
Demand365 topic, 1,300 categoryReal
Evidence2 usable, 1 near missEnough

Decides whether an article topic is worth writing before anyone spends a day on it, from overlap, cluster fit, real demand, your own evidence and a named author.

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

topic-precheck/SKILL.md

In the folder

  • topic-precheck/SKILL.md
  • topic-precheck/references/precheck-rules.md
  • topic-precheck/scripts/precheck_verdict.py
---name: topic-precheckdescription: >-  Decides whether a B2B SaaS article topic is worth researching and writing before any time goes into it. Checks  whether a page on the site already owns the intent, where the topic sits in the topic clusters, whether anyone  searches for it, whether the company has first-party evidence to say something new, and whether a real author is  set, then returns GO, GO (supporting content), STOP or NEEDS DECISION from ordered rules. Use it at the start of  every new article, or to sanity-check a topic idea. Not for picking the keyword itself (use keyword-selection) or  for deciding what to do with a published page (use content-triage).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Uses Search Console, an SEO tool or the  site's sitemap if the agent can reach them, otherwise exports or pasted data. Python 3 is optional, for the verdict  and the demand split.metadata:  author: SHAKEUP  version: "1.0.0"  stage: decide  plugin: content  homepage: https://www.shakeuphq.com/skills/topic-precheck--- # Topic precheck Stops an article before it starts when a page already owns the intent, nobody searches for it, or the company has nothing new to say. ## Purpose - One job: one verdict on one topic, with every reason and figure behind it.- Stage: decide. It ends in a verdict decided by ordered rules (`scripts/precheck_verdict.py`), never by feel.- It never invents demand, rankings, evidence or an author. A missing figure is `not available` and lowers coverage. ## Use when / not | Situation | Use || --- | --- || A new article topic or keyword, before research | this skill || An idea from the team, sales or a content calendar that might become an article | this skill || Choosing the exact keyword and intent for the page | `keyword-selection` (this skill reads its handoff) || Two published pages already compete | `cannibalization-audit` || Deciding what to do with a published page | `content-triage`, then `refresh-audit` || A LinkedIn post, not an article | `founder-post` or `content-calendar` | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || The topic or keyword, in the buyer's words | yes | the reader's request; a `keyword-selection` handoff | ask once; `BLOCKED` without it || The site's existing pages on the topic | no | a `cannibalization-audit` handoff; the sitemap or a site search if the agent can browse; URLs and titles the reader pastes | cannibalization `NOT_CHECKED`; coverage `partial` || Search Console queries, last 12 months | no | a connected Search Console; a Queries export (CSV); pasted rows | demand from the keyword tool alone || Keyword volumes for the topic and its category | no | a `keyword-selection` handoff; a connected SEO tool; an export; figures the reader pastes | demand from Search Console alone; with neither, demand `UNKNOWN` || Topic clusters and pillar pages | no | a `cluster-fit` handoff; `## Content` in the profile; the site's page titles | role `standalone`, flagged || First-party evidence on the topic | no | `## Claims and proof` in the profile; items the reader pastes (cleared quotes, data, cases) | evidence 0 || A named author with a bio page | yes | `## Content` > `Authors` in the profile; the reader's answer | ask once; `STOP` if there is none | ## Profile read From `company-profile.md`: `## Company` (website, brand terms), `## Buyers` and `## Buying moments` (for the intent fingerprint), `## Claims and proof` (evidence), `## Content` (authors, topic clusters), and every `## Learnings` entry for `topic-precheck` with status `noted`, `fact` or `accepted`. No profile: ask in one message for the website, who the buyer is, the named author and their bio link, and any first-party proof on this topic. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** Read the learnings first. An accepted learning that changes a rule or threshold (the evidence minimum, the demand floor) overrides the default below; name it in the output when it fires. 2. **Fix the topic.** Write the intent fingerprint: "This page targets <audience from `## Buyers`> who want to <do, understand or decide> <specific thing> at the <stage> stage." If a `keyword-selection` handoff exists, use its primary keyword and intent. Without one, use the reader's phrasing and mark intent `unconfirmed`. 3. **Run the 3 checks.** Use each skill's handoff when it exists; otherwise run the inline method in `references/precheck-rules.md` and say which you did.    | Check | From | Result |   | --- | --- | --- |   | Cannibalization | `cannibalization-audit`, else rules section 2 | PROCEED, DIFFERENTIATE, REFRESH, BLOCK or NOT_CHECKED, plus the page and the rule that fired |   | Cluster fit | `cluster-fit`, else rules section 3 | role (pillar, supporting, standalone), pillar URL, the gap in one sentence |   | Demand | `keyword-selection` and its demand split, else rules section 4 | GO, SUPPORTING, STOP or UNKNOWN, with problem-intent and category figures reported apart |    Upstream verdicts map as in rules section 1. A `NEEDS DECISION` from `keyword-selection` stays a decision: carry its reason, don't decide it again. 4. **Count usable evidence.** Apply the 3 tests in rules section 5 (relevant, cleared, first-party). List every counted item and every near miss with the test it failed. 5. **Check the author.** A real person, named, with a bio page the reader can link. A job title alone ("our head of content") is not an author until it has a name. Never invent a name or a credential. 6. **Decide.** Run the script with the results:    ```   python3 scripts/precheck_verdict.py --cannibalization DIFFERENTIATE \       --gsc-csv queries.csv --keyword "<topic keyword>" --category "<broad term>" \       --tool-volume 0 --tool-category-volume 1300 --evidence 1 --pillar "<pillar URL>"   ```    Pass `--demand` and `--problem-intent` instead of the CSV when `keyword-selection` already split demand, and `--no-author` when step 5 failed. Without Python, apply the verdict table below in order and stop at the first match; say which way you ran it. Report crawler rows dropped from the Search Console data, with their impressions. 7. **Write the reasons.** Every verdict carries: the rule that fired, the figures behind it, the evidence list, and the flags (overlap not checked, thin evidence, standalone page, pillar missing). A bare verdict is a failure. ## Check and verdict | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No topic after asking once, or the input is a published page's URL (this skill decides new topics; run `content-triage` on the page) || `STOP` | No real, named author (a byline is never invented) || `STOP` | Cannibalization is BLOCK: a healthy page already owns the intent || `STOP` | Auto-stop: (cannibalization DIFFERENTIATE, or demand SUPPORTING) and problem intent is 0 and usable evidence is under the minimum (default 2). The page would repeat an existing page or sit under a category nobody searches by its own name, with too little to say. Stop without asking || `STOP` | Demand STOP and 0 usable evidence || `NEEDS DECISION` | Cannibalization DIFFERENTIATE or REFRESH (a person confirms the angle, or chooses the refresh) || `NEEDS DECISION` | Demand STOP with 1 or more usable evidence items || `NEEDS DECISION` | Demand UNKNOWN: no data from any source || `NEEDS DECISION` | Demand SUPPORTING and no pillar page exists to support || `GO` | Demand GO, cannibalization PROCEED or NOT_CHECKED, author set || `GO (supporting content)` | Demand SUPPORTING with a named pillar: the pillar becomes a required internal link | Problem intent counts as 0 only when every source that covers the topic's own phrasing says 0. When it is unknown, the auto-stop never fires. Named stops: `BLOCKED` and the `STOP` for a missing author ask for the input. `NEEDS DECISION` stops and shows the options, each with what it costs and what it unlocks (for example: "differentiate on <angle> and link to <page>", or "refresh <page> with `refresh-audit` instead"). Every other verdict runs on: `GO` goes to `article-research`; `STOP` ends the run with the reasons on file and no question. ## Learn Append to `## Learnings` in the profile when: - the reader overrides a verdict, such as writing a STOP topic anyway (a `proposal` with their reason);- a result comes back for a past GO, such as the page's impressions, rankings or demos after 90 days (a `fact`, with its source);- a STOP topic later shows demand (a `fact`, and a `proposal` if it suggests a lower floor or a different evidence minimum). Example: `- 2026-10-24 | topic-precheck | proposal | Count a recorded sales call that names the problem as usable evidence | Why: 3 STOP topics from sales calls drew demos once written (reader's CRM, Q3 2026) | Status: proposed` ## Output ```Verdict: <GO | GO (supporting content) | STOP | NEEDS DECISION | BLOCKED> · Rule: <rule id> · Coverage: <complete | partial (what's missing) | blocked>Topic: <topic> · Intent: <fingerprint> (<confirmed | unconfirmed>) | Check | Result | Source | Detail || Cannibalization | <verdict> | <handoff | inline | not checked> | <page and rule, or "none found"> || Cluster fit | <role> | <handoff | profile | inline> | <cluster, pillar URL, the gap> || Demand | <GO | SUPPORTING | STOP | UNKNOWN> | <Search Console | tool | handoff> | problem intent <n>, category <n>, crawler rows dropped <n> (<impressions>) || Evidence | <n> usable (minimum <n>) | profile, reader | <items counted; near misses and why> || Author | <name, bio link | missing> | profile | | Reasons: <the rule's reason, then the figures>Flags: <list, or none>Decision needed (NEEDS DECISION only): <option A, what it costs, what it unlocks> · <option B, ...>Verdict decided by: <script | the table, by reading>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: topic-precheckversion: 1.0.0verdict: GO (supporting content)coverage: partial (no Search Console export; keyword tool figures pasted)for: <topic>outputs:  topic: salesforce attribution reports  intent: decide how to report pipeline by channel in Salesforce  role: supporting  pillar: <pillar URL, required internal link>  evidence: [<claim or item ids counted>]  author: <name, bio URL>  avoid_overlap_with: <page URL and the sub-topics not to repeat, or none>next: [article-research]open: [confirm the angle differs from <page>]``` `next` is `[article-research]` for GO and GO (supporting content), `[refresh-audit]` when the reader picks the refresh, and `[]` for STOP. ## Failure modes | Problem | Do this || --- | --- || The sitemap or site search can't be read | Ask for page URLs and titles once; without them, `NOT_CHECKED` and say so. Never treat it as a clean pass in the reasons || Search Console shows 0 for a topic the site has never written about | That only means no page ranks yet. Use keyword tool data for problem intent; if there is none, problem intent is `unknown`, not 0 || The keyword tool and Search Console disagree | Report both; either one showing demand is enough for GO || The reader's evidence is a public study, not their own | It doesn't count toward the minimum. Keep it for `article-research` || A quote is from a customer but not cleared | Near miss on "cleared". Ask the reader to confirm, don't count it until they do || The reader changes the topic or angle at a NEEDS DECISION | Rerun the cannibalization and cluster checks on the new topic; the old results don't hold || A handoff names a page or evidence id that doesn't exist | Stop and say which; never fill it in || The input is a URL of a published page | `BLOCKED`, with `next: [content-triage]`: refresh, consolidate or retire is that skill's call | ## Influenced by - SHAKEUP's content engine: the precheck verdict decided by ordered rules in code, the "stop instead of asking" rule, and demand split into problem intent and category so a big category figure can't stand in for a topic nobody searches.- Google's guidance on helpful, people-first content, which warns against pages made mainly to attract search visits and against many near-duplicate pages: https://developers.google.com/search/docs/fundamentals/creating-helpful-content

How it thinks

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

What we built

In our own content engine, the verdict comes from rules in code, so nobody argues a weak topic into a slot. We added the rule that stops a topic without asking when it repeats a page, nobody searches its own phrasing and the evidence is thin.

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

    Rules your team accepted, such as counting a recorded sales call as evidence, change the defaults before the run starts.

  2. 02

    Check for overlap

    It writes one sentence on who the page is for and what they need, then compares it with every page you have on the topic. Shared keywords alone don't count as overlap.

  3. 03

    Place it in your clusters

    Pillar, supporting or standalone, with the pillar it must link to and the gap it fills in one sentence.

  4. 04

    Split demand in 2

    Searches for the topic's own phrasing and searches for the broad category are reported apart and never added together. Crawler-like rows are dropped and counted first.

  5. 05

    Count your evidence

    Only first-party items count, and only when they're about this topic and cleared to publish. A public study anyone could cite doesn't make your page say something new.

  6. 06

    Decide by rule

    A script applies ordered rules and names the one that fired. When a page already covers it, nobody searches its own phrasing and evidence is thin, it stops without asking you.

What you need

  • Required

    The topic, in the buyer's words

    Your request → A keyword selection handoff

  • Optional

    Your existing pages on the topic

    A cannibalization audit handoff → Your sitemap or a site search, if your agent can browse → URLs and titles you paste

  • Optional

    Search Console queries, last 12 months

    A connected Search Console → A Queries export (CSV) → Rows you paste

  • Optional

    Keyword volumes for the topic and its category

    A keyword selection handoff → A connected SEO tool → Figures you paste

  • Optional

    Evidence of your own on the topic

    The Claims and proof section of your company profile → Cleared quotes, data or cases you paste

  • Required

    A named author with a bio page

    The Authors line of your company profile → Your answer

What it can return

  • GO

    No page owns the intent, the topic has demand of its own and a real author is set.

  • GO (supporting content)

    Only the category has demand. Write it as support for a named pillar and link to it.

  • STOP

    No real author, a healthy page already owns the intent, or the topic repeats a page with no demand and too little evidence.

  • NEEDS DECISION

    An overlap needs your call, demand is unknown, or a supporting topic has no pillar to support.

  • BLOCKED

    There's no topic after one question.

Where it stops for you

  • When an overlap needs your call on the angle or a refresh
  • When there's no demand data at all and the call rests on your evidence
  • When a supporting topic has no pillar page yet

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. The topic is attribution reports in Salesforce, for teams that run it as their CRM.

Verdict

GO

Rule: clear

Coverage: complete

CheckResultDetail
OverlapProceedThe broad attribution guide serves a different stage; no page covers Salesforce setup
Cluster fitSupportingSits under the attribution pillar and links to it
DemandGo365 impressions on the topic's own phrasing, 1,300 on the category, 2 crawler-like rows dropped
Evidence2 usableOnboarding data on setup time and a product fact. A public survey was a near miss: anyone can cite it
AuthorSetThe head of content, with a bio page
Reasons

no page owns this intent, and the topic has demand of its own. The pillar guide is a required link.

The handoff block the next skill reads
skill: topic-precheck
version: 1.0.0
verdict: GO
coverage: complete
for: salesforce marketing attribution
outputs:
  topic: salesforce marketing attribution
  role: supporting
  pillar: /guides/b2b-attribution
  evidence: [onboarding-salesforce-setup, doc-campaign-influence]
next: [article-research]
open: []

What good looks like

  • One verdict, with the rule that fired and the figures behind it
  • Topic demand and category demand shown apart, never added together
  • Every evidence item counted, and every near miss with its reason
  • A stop without a question when the topic repeats a page and has little to add
  • No search figure that wasn't in your data

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.

Does it need Search Console?

No. It uses Search Console, keyword volumes or both, whichever you have. With neither, it can't call demand, so it returns NEEDS DECISION and says what would settle it.

Why doesn't a public study count as evidence?

Anyone can cite it, so it can't make your page say something the top results don't. It still helps in research. The count only takes items that are yours, on topic and cleared.

When does it stop without asking?

When a page already covers most of the intent, or only the category has demand. On top of that, nobody searches the topic's own phrasing and you have fewer than 2 usable evidence items. That isn't a real decision, so it doesn't make it yours.

What does the script do?

It applies the verdict rules in order and can split Search Console demand itself. It's optional. Without Python, the skill applies the same table by reading and says so.