Marketing an Open-Source Package (NPM / PyPI / GitHub) for SEO & GEO

If you publish a package, you’re sitting on three of the highest-authority domains on the web and probably wasting them. GitHub, NPM, and PyPI all get crawled constantly and trusted heavily — by Google and by the answer engines. The job is to make those pages point back at you, and to shape your README so an LLM quotes it. I’ll use wtcraft as the running example.

摘要:GitHub / NPM / PyPI 域名权重极高(DA 90+),且被 AI 搜索高频抓取。两件事:(1) 用包元数据 + README 回链把权重导回个人站;(2) 把 README 重构成 LLM 爱提炼的形状。

Why registries are leverage

GitHub, NPM, and PyPI carry domain authority in the 90s and — more importantly for GEO — answer engines treat them as trusted, frequently-recrawled sources. When someone asks Perplexity “what’s a good git-native multi-agent tool,” it’s far more likely to retrieve your NPM page or README than your three-week-old personal blog. So the registry page is both your fastest backlink and often the chunk that actually gets cited. Treat it as a primary surface, not an afterthought.

The goal is to make every registry surface declare your site as home, so the equity flows to your domain instead of pooling on the registry.

NPM — package.json: set homepage to the canonical article, not just the repo, and fill in keywords with the phrases people actually search.

{
  "name": "wtcraft",
  "homepage": "https://zywkloo.github.io/blog/dont-build-another-agent-codex-claude-code-wtcraft/",
  "repository": { "type": "git", "url": "git+https://github.com/zywkloo/wtcraft.git" },
  "keywords": ["claude-code", "agentic-workflow", "git-worktree", "openai-codex", "devtool"]
}

PyPI — pyproject.toml: same idea, via [project.urls].

[project]
name = "wtcraft"
keywords = ["claude-code", "git-worktree", "agentic-workflow", "vibe-coding"]

[project.urls]
"Homepage" = "https://zywkloo.github.io/blog/dont-build-another-agent-codex-claude-code-wtcraft/"
"Bug Tracker" = "https://github.com/zywkloo/wtcraft/issues"
"Source Code" = "https://github.com/zywkloo/wtcraft"

README footer backlink: end the GitHub README.md with a deep link to the canonical write-up on your site. GitHub’s authority is high; one honest link from it meaningfully accelerates how fast your own page gets discovered.

关键术语:homepage / [project.urls] 指向正文链接(canonical article),不是只丢个 repo 地址 — 这样回链落点是有内容、能被引用的页面。

GEO: rewrite the README for the model, not just the human

When an answer engine retrieves a package, the LLM summarizes the README. Most READMEs open with a badge wall and a vague tagline — useless for extraction. Three fixes:

  • BLUF in the first screen. One or two sentences that define the tool concretely, before any install instructions or badges:

    wtcraft is a lightweight CLI that uses git worktrees to isolate multi-agent workspaces, preventing code conflicts and preserving task-handoff context.

    That single line is what gets lifted into an answer. Make it precise and claim-free of marketing fluff.

  • A comparison matrix. Models strongly prefer tables when answering “how is X different from Y.” A short Markdown table contrasting your tool with the obvious alternatives is the thing Perplexity will render directly into its answer card.

  • An FAQ section. End with question-shaped H3s — ### Why does wtcraft use git worktrees? — that mirror how people phrase prompts. High-density, self-contained Q&A is exactly what the retriever wants to slice out.

Verify it (验收)

After a few days, ask Perplexity “lightweight git-native tooling for multi-agent coding?” and check the Sources panel. Two passing signals: your README/site shows up as a citation, and the comparison table renders inside the answer. If the model describes the tool wrong, the README BLUF line is the first thing to tighten — see the field-notes overview for the full acceptance loop.

🔗Citation (BibTeX)
@misc{zhang2026,
  title = {Marketing an Open-Source Package (NPM / PyPI / GitHub) for SEO & GEO},
  author = {Victor Zhang},
  year = {2026},
  howpublished = {\url{https://zywkloo.github.io/docs/package-marketing/}}
}