Skip to content

Decides

Cited pages outreach

Finds the third-party pages AI answers already cite for your own buyer prompts and ranks which to get onto first, with the one move that gets you on each.

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 cited-pages.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
Prioritized GEO
Needs
The pages cited for your prompts, per prompt and engine
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

Cited pages outreach

Example output

Rank cited pages
PageCallPlay
Best-of listGet on firstSegment pitch
Review platformFix listingReview program
Competitor blogSkipNone

Finds the third-party pages AI answers already cite for your own buyer prompts and ranks which to get onto first, with the one move that gets you on each.

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

cited-pages/SKILL.md

In the folder

  • cited-pages/SKILL.md
  • cited-pages/references/plays.md
  • cited-pages/scripts/cited_rank.py
---name: cited-pagesdescription: >-  Finds the third-party pages AI answers already cite for a B2B SaaS company's own prompt panel and ranks which to  get onto first, by the buying intent of the prompts that cite them, how many engines cite them, whether  competitors are listed and the company isn't, and how much of the outcome the company controls. Gives each page a  call (GET ON FIRST, GET ON, FIX LISTING, WATCH, SKIP) and a play, and returns RANKED, NEEDS DECISION or BLOCKED. Use  it after ai-answer-panel. Not for measuring visibility (use ai-answer-panel) or checking a tactic's risk (use  ai-search-risk).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Reads the cited pages if the agent can  browse, otherwise the reader's notes on each. Python 3 is optional, for the ranking.metadata:  author: SHAKEUP  version: "1.0.0"  stage: decide  plugin: seo-ai-search  homepage: https://www.shakeuphq.com/skills/cited-pages--- # Cited pages outreach Turns "get more AI mentions" into a short, ranked list of named pages, each with the one move that gets the company onto it. ## Purpose - One job: rank the pages cited for the company's own prompts and give each a call and a play.- Stage: decide. It ends in a run verdict and a call per page, each with the rule that fired.- It works from the company's own prompts only. A list of "the sites AI cites most" across all topics measures someone else's category.- It never recommends fake reviews, undisclosed paid mentions or posting under false identities; those go to `ai-search-risk`. ## Use when / not | Situation | Use || --- | --- || `ai-answer-panel` returned cited sources and a gap | this skill || Planning outreach, reviews or listings for AI visibility | this skill || You don't know how visible you are yet | `ai-answer-panel` first || A planned tactic might count as spam | `ai-search-risk` || Your own page is cited and may be out of date | `refresh-audit` on that page | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || Cited URLs per prompt and engine, with each prompt's family | yes | the `ai-answer-panel` results table; an AI visibility tool's citation export; URLs the reader logged | `BLOCKED`; explain how to log citations on the next run || Whether the brand and competitors are listed on each page | no | read each page if the agent can browse; the reader's notes | presence `unknown` (gap 0.8), coverage `partial`, each page marked "confirm" || Whether a page charges for inclusion | no | the page's own disclosure; the reader's knowledge | not flagged; noted as unchecked || The company's existing listings and reviews | no | the reader; the review platform pages | assume none, say so | ## Profile read From `company-profile.md`: `## Company` (website and brand terms, to spot own pages and listings), `## Competitors` (domains, to mark competitor pages; mention policy), `## Positioning` (the segment the company fits, for list pitches), `## Claims and proof` (only proof that may go in a pitch), `## Channels` (who owns outreach), and every `## Learnings` entry for `cited-pages` with status `noted`, `fact` or `accepted`. No profile: ask for the website, 3 competitors with their domains, and the buyer the company fits best, in one message. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** Read them first: pages already pitched (and the answer), listings already claimed, any page type the reader won't pursue. 2. **Load the citations.** One row per cited URL per prompt per engine. Drop the company's own pages into a separate list for `refresh-audit`. Normalize URLs (drop tracking parameters and fragments). 3. **Type each page.** From the domain and path, or by reading it:    | Type | Looks like |   | --- | --- |   | directory | a partner marketplace, an app store or integration listing |   | review | a software review platform's category or product page |   | list | a third party's best-of, alternatives or comparison page |   | publication | a trade title, newsletter, analyst or news page |   | community | a forum, a Q&A site, a subreddit thread |   | video | a video or podcast page |   | reference | an encyclopedia or knowledge base |   | competitor | a page on a competitor's own domain | 4. **Check presence on each page.** Read it: is the company listed, is the listing current and correct, how many competitors are listed. Mark `yes`, `stale` (listed but wrong or old), `no`, or `unknown` when it can't be read. 5. **Rank.** With Python:    ```   python3 scripts/cited_rank.py --csv cited.csv --own <website> --competitor <domain> --competitor <domain>   ```    Without Python, compute each part and multiply; show the parts, not just the result. Say which way you ran it.    | Part | Value | Why |   | --- | --- | --- |   | Demand | sum over the prompts citing it: category, comparison, budget 3; problem, use case, integration 2 | a shortlist is formed on high-intent prompts |   | Reach | 1.5 when 2 or more engines cite it, else 1.0 | most cited pages appear in one engine; one that travels is worth more |   | Gap | 1.0 absent with a competitor listed; 0.8 unknown; 0.6 absent; 0.5 listed but stale; 0 listed and fine | the gap is the reason to act |   | Effort | directory 1.0, review 0.7, list 0.6, publication 0.4, community 0.3, video 0.3, reference 0.1, competitor 0 | how much of the outcome the company controls | 6. **Give each page a call and a play.** Calls, first match wins:    | Call | Condition |   | --- | --- |   | `SKIP` | a competitor's own page, the company's own page, or effort 0 |   | `WATCH` | listed and fine; or cited by 1 prompt with priority under 3 |   | `FIX LISTING` | listed but stale or wrong |   | `GET ON FIRST` | the 5 highest priorities with a gap |   | `GET ON` | every other page with a gap |    Then take the play for its type from `references/plays.md`, with the proof it needs from the profile. For a list, name the segment it underserves in one sentence. 7. **Write the concentration.** Citations from the top 5 domains divided by all citations logged. A high share means a short target list; a low share (common in B2B) means the list is long and the panel should grow before outreach does. 8. **Flag what needs a person.** Any page that charges for inclusion, any play the reader proposes that is off limits in `references/plays.md` (send it to `ai-search-risk`), and any target where the mention policy in `## Competitors` matters. ## Check and verdict | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No cited URLs after asking || `NEEDS DECISION` | A target charges for inclusion, or the reader proposed an off-limits tactic || `RANKED` | Every page has a call and a play | Page calls: `GET ON FIRST`, `GET ON`, `FIX LISTING`, `WATCH`, `SKIP`, each with its priority parts. Named stops: paid inclusion waits for the reader's decision on budget and disclosure; an off-limits tactic stops and goes to `ai-search-risk`. No other stops. ## Learn Append a learning to the profile when: - a pitch or listing gets a result: listed, declined, no reply (a `fact` with the date), so the next run doesn't repeat it;- the panel shows the change: the prompts citing a page start naming the company (a `fact` from the next `ai-answer-panel` run);- the reader rules out a page type ("we don't do community posting") (a `proposal`, applied once accepted). Example: `- 2026-10-24 | cited-pages | fact | blog.example/best-attribution-software added the company after a segment pitch | Why: page updated 2026-11-12; panel run 2026-12 to confirm | Status: noted` ## Output ```Verdict: <RANKED | NEEDS DECISION | BLOCKED> · Coverage: <complete | partial (what's missing)> · Rank: <script | by hand>Pages: <n> from <n> citations across <n> prompts · Source concentration: <n>% (top 5 domains) | Call | Page | Type | Prompts | Engines | Demand x Reach x Gap x Effort = Priority | Play |(GET ON FIRST first) Segment pitches (lists): <page: the buyer it underserves, the proof to send>Your own cited pages (for refresh-audit): <list>Needs a person: <paid inclusion, off-limits tactics, mention policy>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: cited-pagesversion: 1.0.0verdict: RANKEDcoverage: partial (presence unknown on 3 pages)for: <brand>, panel v1outputs:  get_on_first: [https://blog.example/best-attribution-software]  fix_listing: [https://marketplace.example/apps/brand]  source_concentration: 0.71  remeasure: rerun the prompts that cite these pages next monthnext: [ai-answer-panel, refresh-audit]open: [confirm presence on 3 pages]``` ## Failure modes | Problem | Do this || --- | --- || The reader only has a "top cited domains" list from a study | Not usable for ranking: it isn't their prompts. Ask for the panel's citations || A page can't be read (blocked, paywalled) | Presence `unknown`; keep it ranked, mark "confirm" || Every top page is a competitor's own blog | Skip them; the play is the company's own page on the same question (`keyword-selection`, `refresh-audit`) || A list ranks the company badly or wrongly | `FIX LISTING`: send the author a correction with public proof || The reader wants to pay for reviews or post as customers | Stop; this is fake engagement. Route to `ai-search-risk` || The same domain has many cited URLs | Rank pages, not domains; a domain's other pages may need different plays | ## Influenced by - Grow and Convert's prioritized GEO, which ranks AI visibility work by the pages cited for the company's own buying prompts and puts outreach to those cited pages ahead of generic tactics: https://www.growandconvert.com/ai/prioritized-geo/- SHAKEUP's AI visibility playbook: the retrieval map, source concentration, and the list play built on the segment a list underserves.

How it thinks

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

What we built

The ranking shows its 4 parts, so you can see why a best-of list beats a forum thread for your prompts. Each page type gets one play, and fake reviews, sock puppets and undisclosed paid mentions are never on the list.

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

    Start from your own prompts

    Only pages cited for your buyer prompts count. A study of the sites AI cites most measures someone else's category.

  2. 02

    Type each page

    Directory, review platform, best-of list, publication, community, video, reference or a competitor's own page. Each type has its own play.

  3. 03

    Check if you're on it

    Listed and current, listed but stale, absent with competitors listed, or unknown.

  4. 04

    Rank by 4 parts, shown

    Its script multiplies the buying intent of the prompts citing the page, how many engines cite it, the size of your gap and how much of the outcome you control.

  5. 05

    Give each page a call and a play

    Claim a listing, run a review program, or pitch a list's author the buyer it misses, with proof they can check.

  6. 06

    Flag what needs you

    Paid placements and any tactic that counts as fake engagement stop for your decision.

What you need

  • Required

    The pages cited for your prompts, per prompt and engine

    The results table from the AI answer panel → A citation export from an AI visibility tool → URLs you logged during your runs

  • Optional

    Whether you and your competitors are listed on each page

    It reads each page if your agent can browse → Your own notes on each page

  • Optional

    Whether a page charges for inclusion

    The page's own disclosure → What you know about it

What it can return

  • RANKED

    Every cited page has a call and a play.

  • NEEDS DECISION

    A target charges for inclusion, or a proposed tactic is off limits.

  • BLOCKED

    No cited pages were logged.

Where it stops for you

  • When a page charges for inclusion
  • When a proposed tactic counts as fake reviews, fake mentions or undisclosed paid placement

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. Its prompt panel logged 14 citations of 8 pages across ChatGPT and Perplexity.

Verdict

RANKED

Coverage: complete

Source concentration: 79% from the top 5 domains

CallPageTypeDemand x Reach x Gap x EffortPlay
Get on firstA best-of list on a review blogList9 x 1.5 x 1.0 x 0.6 = 8.1Pitch the author the buyer the list misses: sales-led teams below enterprise size
Fix listingA software review platform's categoryReview6 x 1.0 x 0.5 x 0.7 = 2.1A review program: 9 reviews, the newest from 2024
Fix listingIts CRM marketplace listingDirectory2 x 1.5 x 0.5 x 1.0 = 1.5Update the 2022 description to today's product
WatchA head-to-head between 2 competitorsList3 x 1.0 x 1.0 x 0.6 = 1.8Cited by 1 prompt; check next month
SkipA competitor's alternatives postCompetitorEffort 0Answer the same question on its own page
Its own cited page

the pipeline attribution guide, sent to a refresh audit to make sure what answers quote is current.

The handoff block the next skill reads
skill: cited-pages
version: 1.0.0
verdict: RANKED
coverage: complete
for: panel v1, October 2026
outputs:
  get_on_first: [the best-of list]
  fix_listing: [review platform, CRM marketplace]
  source_concentration: 0.79
next: [ai-answer-panel, refresh-audit]
open: []

What good looks like

  • Every page comes from your own prompt citations
  • Each priority shows its 4 parts, not just a score
  • Best-of list pitches name the buyer the list underserves
  • Competitors' own pages are skipped, and your own go to a refresh
  • No fake reviews, sock puppets or undisclosed paid mentions

Install

Use it in the tool you already have.

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

  1. 01
    Download cited-pages.zip.
  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 not target the sites AI cites most in general?

Studies of the most cited sites mix every topic, often consumer ones. The pages cited for your own buyer prompts are the ones shaping the answers your buyers see.

Will it suggest buying reviews or placements?

Never fake reviews or undisclosed paid mentions. A page that sells placements is flagged for your decision on budget and disclosure.

How long does it take to see a change?

A listing you control can update in days. Lists take weeks to a quarter, publications 1 to 2 quarters. Rerun the prompts that cite each page the month after it changes.

What does the script do?

It types each page, scores intent, engines, gap and effort, gives each page a call and computes source concentration. Without Python, the skill makes the same count itself and says so.