# From citations to a useful plan

Use only URLs actually returned as citations or pages found during an explicit
public search. Open a page before asserting its contents. Save its URL, capture
time, short supporting excerpt and the observation IDs which cited it.

Prioritize without a hidden numerical score:

1. **Relevance:** does the page answer a buyer question in the agreed market?
2. **Observed recurrence:** how many buyer answers cited it, by engine and
   question? Show source-pass counts separately. This is a sampled count.
3. **Gap:** does it omit a suitable business, state a wrong fact, lack a needed
   comparison, or point to an owned page that cannot clearly answer the question?
4. **Feasibility:** can the business improve its own page, provide an editor a
   useful sourced correction or original evidence, or earn inclusion on merit?

For owned pages, suggest the smallest supported change: clear offer/eligibility,
facts, useful comparison, first-hand evidence, accessible text or a clear source
and update date. Verify any proposed structured data against visible content.
Do not promise that an llms.txt file, schema or keyword stuffing wins citations.

For third-party pages, inspect the author byline, editorial/contact page and
submission policy within the remaining page budget. Distinguish an identified
author, a generic editorial address and no public contact. Save the source URL
for every contact; never infer an address from a naming pattern. An unavailable
contact does not erase the opportunity.

Draft at most three short messages when useful:

- Reference the specific page and the genuine gap or correction.
- Offer one supported fact, original resource or useful comparison.
- Suggest a precise editorial improvement; avoid claiming an existing
  relationship or offering undisclosed payment.
- Include the relevant public source and leave unverified details blank.

All messages remain drafts. The user must authorize recipients and content
before any tool sends them. This Skill has no mail or social sending dependency.

Optional enrichment is decision-led. If keyword demand would change the chosen
questions, backlink data would prioritize two otherwise equal opportunities, or
public contact research fails, offer a separately inspected tool and bounded
incremental quote. Do not require expensive enrichment to make the core report
look complete. Relevance and direct citation evidence remain primary.