SEO's New Unit of Distribution Is a Markdown File

Illyism/seo-skills packages an entire SEO practice into SKILL.md files that any coding agent can read — which says as much about where search is heading as it does about the repo.
There is a particular genre of GitHub repository that would have made no sense eighteen months ago: no source code, no dependencies, no build step. Just folders of Markdown. Illyism/seo-skills is one of these, and it is worth taking seriously precisely because of that — it’s a bet that the deliverable unit of an entire professional discipline can now be a text file an agent reads before it starts working.
The repo, by Ilias Ism (the builder behind LinkDR, GenPPT, and AI SEO Tracker), contains eleven skills organized into four categories — SEO, content, marketing, and copywriting. Each skill is a folder with a SKILL.md file following the Agent Skills spec, and the whole thing installs into Claude Code, Cursor, Codex, or anything else that speaks that format. Once installed, the pitch goes, your agent can run daily SEO checks, rewrite title tags, refresh posts that slipped in the rankings, map search intent, draft comparison pages, and — the one that stops people mid-scroll — buy backlinks.
The interesting part is the boring part
Strip away the AI framing and what you have is a set of opinionated playbooks. The flagship, seo-autopilot, is a daily routine: site health checks, fixes, a keyword map, article drafts, free tools, link drafts, and tracking of how AI systems answer questions about your site. blitz-seo is a 30-day sprint to rank one page for one money keyword — a scope reduction that most SEO consultants would endorse through gritted teeth, because it’s the honest version of what actually works. blog-refresher does a SERP gap analysis on an existing post so it can outrank the pages above it. The copywriting skills encode structures like Hook → Spike → Payoff.
None of this is software. There’s no crawler, no rank tracker, no API integration in the repo itself. The skills are methodology rendered as prose — and that’s the point. The value of an SEO agency was never really the tooling; the tooling was Semrush subscriptions and a spreadsheet. The value was knowing what to do on Monday morning. A SKILL.md file is that Monday-morning knowledge, written down in a form an agent will actually follow.
This is the same insight driving the broader skills ecosystem, and it’s why the boring part deserves explanation: the distribution mechanism is the product. A skill file is small, versioned, composable, and portable across agents. It competes not on features but on the quality of the judgment baked into the text. That’s an unusual competition for software to be in, and it favors people who’ve done the work over people who’ve written the framework.
A crowded shelf, quickly
Illyism is early but not alone, and the pace of the land grab is itself a signal. Claude SEO Skill, built by Alex Labat, runs full SEO, GEO, and AEO audits from inside a Claude conversation — crawling a site, scoring it, and producing DOCX and PDF reports with a priority matrix. The claude-seo.md project has escalated to 26 skills, 34 user commands, 19 specialist agents, and support for up to 17 parallel agents, with a test suite that reportedly passes at every commit — a level of engineering ceremony around prompt packages that would have seemed absurd a year ago. Corey Haines’ marketingskills repo takes a different shape: a single 500-line ai-seo skill with a reference-routing system, an explicit evidence hierarchy (official docs first, then controlled experiments, then observational data), and a warning against turning study percentages into client forecasts. That last touch — epistemic hygiene inside a prompt file — is arguably more sophisticated than anything in the larger packages.
Meanwhile, the commercial world is converging on the same territory from the other direction. WorkAgentic sells AI agents for keyword research with deployment processes, pilot phases, and testimonials from manufacturers and healthcare firms. The pattern is clear: SEO is becoming something you delegate to an agent, and the open question is whether the winning form is a hosted product, a services layer, or a Markdown file you got from GitHub.
The Markdown file has one structural advantage: it’s inspectable. You can read exactly what your agent will do before it does it, edit the parts you disagree with, and diff your changes against upstream. Whether that advantage survives contact with buyers who don’t want to read anything is a separate matter.
Why now: the GEO land rush
The timing of all this isn’t accidental. The discipline being automated is itself mid-mutation. What the industry now calls GEO — generative engine optimization, getting cited in AI-generated answers rather than ranking on a results page — has gone from conference talk to Coursera course with ten thousand enrolled learners. Semrush’s practical guide draws the distinction cleanly: SEO optimizes for crawlability, keywords, and backlinks, measured in rankings and traffic; GEO optimizes for clarity, extractability, and credible mentions, measured in AI visibility and share of voice. Built In’s coverage frames the two as complementary, while noting that GEO measurement tools haven’t reached the granularity of SEO’s mature metrics.
Google, for its part, has published official guidance on optimizing for AI Overviews and AI Mode, and its position is notably deflationary: the generative features are built on core Search ranking systems, so traditional best practices still apply. The guide describes retrieval-augmented generation and query fan-out — the model firing off related queries to gather more sources — and lands on the least exciting advice possible: create unique, valuable content with first-hand perspective, because that will matter more than anything else for long-term AI visibility.
Illyism’s skills sit squarely in this shift. The autopilot skill tracks AI answers, and the customer-journey skill maps the prompts people type into Google and ChatGPT at each stage of awareness — a small syntactic detail that captures the whole transition. When the unit of discovery becomes the answer rather than the link, the work becomes making your content the thing an agent extracts, summarizes, and cites. Skills files are, fittingly, written for exactly that kind of reader.
The caveats, stated plainly
Three things deserve scrutiny.
First, this is glue — or rather, it isn’t even glue; it’s instructions. The repo contains no verification machinery of its own. Whether the autopilot’s daily checks actually catch what matters depends entirely on the underlying agent and whatever tools it can reach. Compare that to claude-seo.md’s claimed test suite or Labat’s structured crawl-and-score pipeline, and the difference in engineering weight is visible. For a solo operator who already trusts their agent, that may not matter. For an agency, it might.
Second, the commercial threading is not subtle. The linkdr skill buys managed backlinks from LinkDR — Ism’s own product — and sends a checkout link. The scribiz skill routes video transcription through Scribiz, also his. The README’s free checklist funnels into a newsletter about those same products. This is affiliate marketing wearing the clothes of open source, and it’s worth saying clearly: MIT-licensed or not, the repo is also a distribution channel. That doesn’t make the skills bad — the LinkDR skill’s idea of planning a domain-rating-tier mix against a budget before spending anything is genuinely sensible — but a reader should know the incentives. And buying backlinks at scale sits in the territory Google’s spam policies have never been friendly toward, a tension the repo doesn’t address.
Third, the scaled-content question hangs over the entire category, not just this repo. Google’s own guidance warns that generating content targeting every query variation to manipulate rankings or AI responses violates its scaled content abuse policy. An agent that runs daily, writes articles, drafts comparison pages, and builds free tools is precisely the machinery that policy is aimed at. The skills that encode restraint — blitz-seo’s one-page focus, blog-refresher’s gap analysis before rewriting — are the ones that age well. The ones that encode volume won’t.
Outlook
The open question for seo-skills and its many siblings is whether SEO judgment compresses well into prose. The early evidence is mixed but interesting: the market is simultaneously rewarding tiny opinionated files (this repo), sprawling multi-agent frameworks (claude-seo.md), and hosted products (WorkAgentic). That’s the signature of a discipline whose form factor hasn’t settled.
What seems durable is the underlying move. SEO was always a race between people who understood the algorithm and people who executed at scale, and agents collapse the execution side to nearly zero. What’s left is the understanding — and a SKILL.md file is currently the cheapest way to ship understanding to someone else’s agent. Ism’s repo may or may not be the one that wins; it’s a small, honest, commercially motivated package in a field rapidly filling with larger ones. But the bet it makes — that a professional practice can live in eleven Markdown files — is the bet the whole industry is now testing.
Sources
- Claude SEO Skill — AI-Powered SEO, GEO & AEO Audits
- Google's Guide to Optimizing for Generative AI Features on ...
- AI SEO Audits Help Him Manage 3.3X More Agency Clients
- 26 Free AI SEO Skills
- AI SEO: Mastering Generative Engine Optimization (GEO)
- AI Agents for SEO Keyword Research
- My New Claude Skill - SEO consultant - 13 sub-agents, 17 ...
- Generative Engine Optimization (GEO): A Practical Guide
- AI agent for SEO: Automate your SEO tasks
- marketingskills/skills/ai-seo/SKILL.md at main
- Generative Engine Optimization (GEO): Is It the New SEO?
- AI Agents for SEO: What They Are, How They Work, and ...