11,302 Stars, No Verdict

Sorting GitHub by stars when you pick an SEO tool selects for the author's marketing skill, not the tool's engineering. Our directory data makes the separation unusually clean.

Here are two data points from the SEO tooling space.

The first is technical-seo-checker, by aaron-he-zhu. It scores 9.3 in our six-dimension review — the highest of any SEO skill we have assessed — with a perfect 10 on structure. Crawlability, indexability, sitemap validity, canonical consistency, redirect chains: the mechanical work that decides whether a site can rank at all. Its star count on GitHub is unremarkable. A few hundred, at best.

The second is geo-seo-claude, which sits at 11,302 stars. It is not in our directory, and not because we missed it. It is a collection repo — a grab bag of prompts and scripts assembled under one banner — and when we looked at what the stars were actually attached to, we could not verify that there was a coherent tool underneath. We don't review what we can't assess, so it stays out.

One of these repos has eleven thousand stars. The other is the better tool.

What stars measure

Star counts are a distribution outcome. They track launch timing, Hacker News and Reddit placement, a well-designed README badge row, and the "awesome-list effect" — collection repos accrue stars because they look like resources worth bookmarking, the way a link directory accumulates bookmarks regardless of what's behind the links.

None of that is quality signal. It is audience-acquisition signal, and those are different disciplines done by different people. A collection repo that ships nothing verifiable can out-star a rigorous single-purpose tool by two orders of magnitude, and on GitHub it does, routinely.

What our numbers say

We ran the correlation across our own directory, using downloads as the behavioral outcome — the closest proxy we have for "a working engineer kept this installed."

• Star count vs. downloads: r = 0.02. Nothing. You cannot predict whether a skill gets used from how starred it is.
• Review score vs. downloads: r = 0.47. A real relationship.
• The relationship is sharper at the top: skills scoring above 9.5 show a strong score-to-download correlation, while below that threshold the correlation disappears entirely.

That last point is worth sitting with. Below roughly 9.5, users can't tell the difference between an 8.2 and an 8.9 from the outside, so score stops driving behavior. Above it, the difference is visible — the tool does something others don't — and downloads follow. technical-seo-checker at 9.3 sits just under that line and still has the highest download velocity in our SEO category (⬇14), which tells you the effect starts before the statistics say it must.

SEO tooling is the worst case for star-sorting

Every tool category has a star-quality gap. SEO has the widest one, for a structural reason: the authors of SEO tools are, professionally, marketers. A person who writes an SEO skill knows how to run a launch — what subreddit, what title, what time of day, which badge goes first in the README. The star count is the output of their SEO skill applied to their own repo.

So when you sort GitHub's SEO category by stars, you are running a query that selects on exactly the confounder you're trying to exclude. It's an observational study where the treatment assigns itself.

The same author field cuts the other way, too. ai-seo by coreyhaines31 scores 9.2 with a 10 on security — a dimension where collection repos basically never survive review, because nobody has read all 11,302 stars' worth of inherited code. on-page-seo-checker (9.2) and seranking's seo-google (9.1) and seo-api (9.0) round out the top of our SEO table, and none of them are star leaders.

What to sort by instead

Structured review, with the criteria published. Our six dimensions — trigger, structure, workflow, content, engineering, security — exist precisely because the aggregate signals on the platform are noise. r = 0.47 versus r = 0.02 is not a subtle difference; it's the difference between a signal and its absence.

Stars tell you a repo is good at getting starred. Scores tell you what the tool does when you run it. In SEO tooling, where the authors are marketers by trade, conflating the two isn't a rounding error — it's the whole mistake.