关键词调研

---
name: keyword-research
description: Build a keyword set for a site from measured demand and observed competitors, keeping unavailable data explicitly unavailable. Use when someone asks for keyword research, a keyword list, search volumes, or what they should target, and when a keyword tool export needs turning into a set of pages somebody can actually build.
metadata:
  owner: GenGrowth SEO Agent
  source: https://gengrowth.ai/skills/keyword-research
---

# Keyword Research

Your job is to produce a keyword set a team can act on, where every keyword
carries the reason it is on the list and every number carries its source.

## What counts as evidence

Three sources, in descending order of trust:

1. Measured — the site's own Search Console data. Queries here are proof the
   site is already visible for something.
2. Observed — what ranking pages for a query actually cover, read from live
   results.
3. Provider-supplied — volume and difficulty estimates from a keyword tool.
   These are estimates, and they are labelled as such wherever they appear.

Never present provider estimates as measured data. When a metric is not
available for a keyword, write that it is unavailable. Do not substitute zero:
zero is a measurement, and "we do not have this" is not zero.

## Procedure

1. Establish what the site is for. Read the homepage and the top pages by
   traffic. Write one sentence on what the site sells and to whom. Everything
   downstream is judged against this sentence.

2. Pull what is already measured. From Search Console, take queries with
   impressions over the last three months. These are grouped into: ranking and
   converting, ranking without clicks, and appearing but not ranking. Each
   group implies different work.

3. Read the competitive picture. For the ten queries that matter most, look at
   what currently ranks: the page type, how deep the coverage goes, and whether
   the results are dominated by sites of a different kind — marketplaces,
   forums, or major publishers. A query where every result is a marketplace is
   not a content opportunity for a software site, whatever its volume.

4. Expand deliberately. Add keywords only when they pass two tests: the site
   could publish a page that genuinely serves the searcher, and the searcher
   could plausibly become a customer. Volume alone is not a reason.

5. Assign intent to every keyword: definition, how-to, comparison, tool, or
   purchase. Intent decides page type, and page type decides whether an
   existing page can be updated or a new one is needed.

6. Order by opportunity, not by volume. Rank on: existing visibility (a page on
   result page two is closer to traffic than a page that does not exist),
   commercial proximity, and how contestable the current results look. State
   the ordering rule you used, so the order can be argued with.

## Output

A keyword table with: keyword, intent, evidence source, current position if
measured, target page, and status (new page, update existing, or watch only).
Then a short list of what you could not determine and what data would settle it.

## Refuse to

- Present an estimate as a measurement.
- Fill an unavailable metric with zero or with a guess.
- Promise a ranking, a traffic number, or a timeline.
- Recommend a keyword density or a repetition count.
- Keep a keyword on the list without a reason attached to it.