界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
数据与研究 · SKILL.md

programmatic-seo

programmatic-seo

【数据与研究】Plan and produce SEO-driven pages at scale using templates and structured data.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: programmatic-seo description: "Plan and produce SEO-driven pages at scale using templates and structured data."

Programmatic SEO

You design template-driven pages at scale ("city × service", "tool × use-case", etc.). Read product-marketing-context first.

When pSEO works

  • A real, segmented search demand exists (verifiable in keyword tools).
  • You have or can generate distinctive content per page (data, examples, comparisons).
  • Each page can actually serve user intent — not be a doorway.

When pSEO fails

  • Pages are 95% boilerplate with one variable.
  • No unique value over what's already ranking.
  • No internal links connecting the set.
  • No proof-of-quality (reviews, data, images).

Build steps

  1. Demand mapping — find a 2D matrix of intent (e.g. [Topic] × [Audience]) with > 50 keyword variants and verified volume.
  2. Page template — design a single template that answers all variants well; require 3+ unique blocks per variant.
  3. Data spine — structured table (Sheets / Airtable / DB) with one row per page; columns include unique content fields.
  4. QA gates — minimum word count, originality check, unique title + meta, distinct internal links, real image asset.
  5. Hub & spoke — a category hub linking out to spokes and back; spokes linking to peers.
  6. Indexability strategy — start with a small batch; let GSC indexing prove the model before scaling.

Output

  • Demand matrix sheet.
  • Template wireframe.
  • Data spine schema.
  • Launch plan: 50 → 250 → 1000 pages with checkpoints.
  • Hand off to schema-markup and site-architecture.

Anti-patterns

  • Generating thousands of pages on day 1.
  • Identical templates with one keyword swap.
  • No human pass before publish. Adapted from coreyhaines31/marketingskills (MIT).
我的收藏