Skills/SEO
5 skills · 0 measured · no API keys, no scraping, no folklore

Claude SEO skills, without the folklore

Search work built from primary sources: what the engines actually document, what they explicitly tell you to ignore, and the numbers everyone publishes that have no source at all. How we test →

All 5 shown

5 skills we have not measured yet

Written, reviewed and free to take. No run behind them, so no claim about what they do to an output. What that means.

SEONot yet measured Internal link map builder Produces the actual link map, direction by direction, with anchor text held to Google's documented standard and changes sequenced into dated batches you can verify one at a time. skill · 3,248 words · MIT Read the write-up
SEONot yet measured Keyword cluster map builder Clusters keywords by SERP result overlap rather than semantic similarity, then rules on which pages to build, merge or drop. skill · 3,256 words · MIT Read the write-up
SEONot yet measured SERP-first SEO article writer Classifies what a keyword's results page rewards, matches an article's format, depth and structure to it, then checks the draft against the three real competitors before it can pass. skill · 3,183 words · MIT Read the write-up
SEONot yet measured SERP title and meta description writer Replaces the unsourced 50-to-60-character rule with Google's documented title rewrite triggers, a width check needing no invented constant, and a branch for when a description is worth writing. skill · 3,120 words · MIT Read the write-up
SEONot yet measured Traffic decay content refresher Separates a real ranking decline from a Search Console measurement artefact or a retired rich result, then rewrites, consolidates, leaves alone or removes the page, with a record of why. skill · 3,330 words · MIT Read the write-up
no API keys, no scraping, no folklore

The list

Most SEO skills in circulation are a few hundred words of advice that stopped being true some time ago, and the tell is always a confident number. If one tells you to hold keyword density between one and two per cent, to add LSI keywords, to keep a title under sixty characters, or to add FAQ markup for rich results, it was written from folklore rather than from anything current, and the rest of its advice deserves the same suspicion.

This is the most crowded topic in the library, so each of these files has to earn its place on a specific angle rather than on covering the subject. The angle is primary sources, and it turns out to be a much bigger differentiator than it sounds, because the gap between what search engines document and what the field repeats is enormous.

Three examples, all verified against the engines' own documentation in August 2026. Neither Google nor Bing publishes a character limit for a title or a meta description. Not one. Google says truncation fits the device width, which makes it variable by definition, and the famous Bing figures trace to a 2009 blog post that is no longer at a live URL. So the fifty to sixty characters in every SEO tool on the market has no primary source on either engine.

Google publishes a list of things it says you should not focus on, in its own starter guide, under that heading. It disposes of word count, keyword-rich domains, the choice between a subdomain and a subfolder, heading-tag rules, and the duplicate content penalty. The same document answers the question of whether E-E-A-T is a ranking factor with the words "No, it's not." Almost nobody quotes this, because the field prefers staff comments on social media to the documentation sitting on the same domain.

FAQ rich results are gone. Not restricted, gone: they stopped appearing on 7 May 2026 and the documentation was removed on 15 June 2026, so FAQPage markup now produces nothing at all. HowTo was fully deprecated in September 2023, and six further types were retired between September 2025 and January 2026. Any guide still recommending FAQ markup for visibility is describing a feature that does not exist, and any page that lost traffic on those dates lost clicks without losing a single position, which looks identical in a chart and needs the opposite response.

There is one genuine asymmetry worth knowing, and these files use it rather than papering over it. Bing publishes its ranking parameters in stated order of importance and openly documents that user engagement is among them. Google publishes no such list. So where a claim about engagement signals can be sourced, it is sourced to Bing and labelled as Bing, rather than being asserted about Google on the strength of inference.

None of these skills can crawl your site, query an API, or see your analytics. They work on material you supply and on rules that hold regardless of the account, and every figure they state carries the date it was checked, because a fact with no timestamp rots silently in this subject faster than in almost any other.

What the two labels mean

Some of these carry a number. Most do not, and they say so.

Measured means the skill was given a realistic task on real material, then the identical task was run again with the skill removed, five runs each way. Each output was graded alone, against a rubric written by someone who had never seen the skill, by a session that was not told the other arm existed. Whether it passed was decided by a rule written down before any run executed. Those pages carry the worst case, the median, the p value, and what the skill costs you as well as what it buys.

Not yet measured means exactly that. It is written, it has been read, it is free to take, and we have run no experiment on it, so we make no claim about what it does to an output. It is not a skill that failed. Skills that failed are not published at all, in either state, and their numbers are in the results table.

Measuring one skill properly costs roughly twenty model sessions. We are working down the queue and moving skills from the second group into the first. Read the full method, or go back to all skills.