The SEO engine
you run yourself.

SEOMANTIK wires Claude Code to live search data and hands you the tested prompts. Point it at a domain and a seed keyword. It maps the whole topic, grades your coverage, plans the architecture, and writes the first drafts. Real files on your machine, $3 to $5 of data per run.

From $149one time, no subscription

  • No app to install
  • Plain-English prompts
  • Files you keep
AI changed SEO

Cover the whole topic.Write it Google's way.

Keyword tools give you a list and a guess. SEOMANTIK teaches you to run an engine that maps the whole topic with AI and live search data, finds every gap, and drafts the pages to fill them.

An engine you own, not a subscription you rent.

claude code~/your-project
Crawl  yourdomain.com and grade it against the topic universe22 pages read, 14 mapped to a subtopic. Graded against all 16 subtopics the topic spans:│   Core head termsthin, worth 321,970 searches a month│   Best-machine queriesthin, worth 301,610│   Brewing technology internalsnothing on the site│   Troubleshootingnothing on the site│   + 10 more graded in the fileWrite  data/entity-map.json1 subtopic is covered properly. 14 of the 16 are thin or missing, and each one is named with the pages it needs.
Old SEO isn't working

The old SEO loop is breakingin three places at once.

Pick a keyword, write the best page, build links, rank, get traffic. Every step of that loop is failing, and they're failing together.

01
clicks → citations

AI surfaces are eating clicks.

Informational queries resolve at the top of the SERP or inside ChatGPT, Claude, Gemini, Perplexity, Grok, and Copilot without a click. If you're not in the citation set, you're invisible on that query, and ranking position is irrelevant.

02
pages → topics

Google ranks topics, not pages.

A site that covers a topic comprehensively (every subtopic, every entity, properly interlinked) outranks a site with one "perfect" article. The winners have structural coverage; the losers have isolated articles.

03
a channel nobody runs

LLMs are a parallel traffic channel nobody you know is optimizing for.

ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot, and Google's AI mode cite sources, and those citations send traffic. The cited sites have clear answers, named entities, clean architecture. Most sites aren't even measuring it yet.

AEO and GEO are the new SEO

You rank for the question.The answer never mentions you.

The same question now gets answered twice: once in the ranked results, and once above them, where a few named sources settle it before anyone scrolls.

Being cited is the new being ranked.

What SEOMANTIK is

A course you run,not one you watch.

Ten chapters that build one thing: an SEO engine of your own, made of Claude Code, live search data, and the prompts that drive it. You follow a chapter on your own site, and what it produces stays on your disk.

After the last chapter it is yours. Point it at another site, or another SEO job, and it runs that too.

Why it wins

Real data,not a hunch.

A topic map drawn from what you assume the topic covers is a guess in a neat shape. The engine builds yours from live search data: the real subtopics, the entities the topic expects, the questions people actually ask.

And it does not stop at the map. It plans the pages and writes them.

How it works

Three moves.That's the whole thing.

No setup maze, no dashboards to learn. If you can paste a sentence into a chat box, you can run it.

01

Point it at your site.

Open Claude Code in your project and type one plain-English line: your domain and a single seed keyword. No spreadsheet to fill in, no dashboard to configure. That one sentence is the whole input the engine needs.

02

It reads your real market.

From that seed it pulls live search data and runs the tested prompts for you: mapping the full topic, grading where your site is strong or thin, ranking the questions worth answering, and ordering what to build first. Real data, not a hunch.

03

The plan lands as files.

Every phase writes a real file to a folder on your machine: your topic map, the gaps, the ranked questions, the architecture, the build order, and your first drafts. Open them, edit them, publish. They are yours to keep and re-run on any site.

claude code~/your-project
Pull related keywords for "home espresso machine" in en-US. For each one capture search volume, CPC, difficulty and intent. Save the full raw response to data/raw/. Do not summarize.Fetch  related keywordsWrite  data/raw/related_keywords.jsonSaved the full response. 225 unique keywords came back, 1,651,580 searches a month between them. The biggest: de'longhi espresso machine at 90,500.
claude code~/your-project
Pull related keywords for "home espresso machine" in en-US. For each one capture search volume, CPC, difficulty and intent. Save the full raw response to data/raw/. Do not summarize.Fetch  related keywordsWrite  data/raw/related_keywords.jsonSaved the full response. 225 unique keywords came back, 1,651,580 searches a month between them. The biggest: de'longhi espresso machine at 90,500.
01

Point it at your site.

Open Claude Code in your project and type one plain-English line: your domain and a single seed keyword. No spreadsheet to fill in, no dashboard to configure. That one sentence is the whole input the engine needs.

Every prompt is written and tested for you. You paste, it works, the file appears.

The modules

Six modules, one run. Each one writes the file the next one reads.

  1. data/universe.json
  2. data/entity-map.json
  3. data/questions.json
  4. data/architecture.json
  5. data/build-list.json
  6. output/
Module 01

The topic universe

Every subtopic, entity and question the topic spans, gathered from live search data into one file on your disk.

claude code~/your-project
Fetch  related keywords for "home espresso machine"Saved the full response to data/raw/related_keywords.json. 225 unique keywords came back. Top by volume:│   de'longhi espresso machine90,500│   home espresso machine49,500│   best at home espresso machine33,100│   + 222 more, 1,651,580 combined monthly searchesFetch  the live Google SERP for the seedSaved to data/raw/serp_seed.json. Page 1 returned 9 organic results and 8 People Also Ask questions, including "Are home espresso machines worth it?" and "What is the 30 second rule for espresso?". No AI Overview appeared on this query.Ask    the AI layer how it clusters the topicSaved to data/raw/ai_keyword_data.json. The AI clustering names 10 clusters. 3 of them appear nowhere in the SERP data: Brewing Technology, Maintenance & Reliability, Troubleshooting. Those are the opportunity subtopics.Synthesize data/universe.json from the raw files onlyWrite  data/universe.jsonThe universe holds 16 subtopics, 104 entities and 70 questions. 4 subtopics carry no keyword volume and entered on entity and AI signal alone. Every item traces back to a record in data/raw/. Read the subtopic list and tell me if any should be split or merged.
The content engine

It does not stop at a plan.
It writes the pages, behind a gate.

The plan tells you what to write. The engine collects the evidence first, writes from it, and refuses to publish a page whose claims cannot be traced back to where they came from.

  1. 1

    Assemble the evidence

    It collects what the article is allowed to stand on, and every item carries its owner, decided the moment it is collected. Who stands behind each fact is settled before a word is written.

  2. 2

    Write from the bundle

    The draft is written from that evidence, not talked into shape afterwards. There is no revise pass to rescue a draft that started from nothing.

  3. 3

    Check, and say what was not checked

    A script, not a reader. It asks whether every source id resolves, whether every price, rate, date and duration cites something, and whether anything built on a rival names that rival. No score and no rubric, and it prints the blind spots it did not look at.

  4. 4

    Humanize against real prose

    It matches the way the site already writes instead of hunting banned words. Then the same check runs again, so a style pass cannot quietly un-source a claim.

  5. 5

    Publish, or refuse

    A draft that cannot be traced does not publish. You can force it, and the record says it was forced, so there is no quiet way to ship one.

claude code~/your-project
Assemble the evidence for "How fine should I grind for espresso?"Every item in the bundle carries its owner, decided when it was collected.Write  content/how-fine-should-i-grind-for-espresso.bundle.jsonDraft   from the bundleCheck   traceability, a script, not a reader│   every source id resolves│   every price, rate, date and duration cites something│   every sentence built on a rival names that rival│   5 blind spots printed, each labelled NOT CHECKEDNo score, no rubric. Three structural questions, and an honest list of what it did not look at.Humanize against the prose the site already writesCheck   again, the same script, after the rewriteA style pass cannot un-source a claim.Write  content/how-fine-should-i-grind-for-espresso.final.mdWrite  content/how-fine-should-i-grind-for-espresso.sources.json7 sourced claims, each mapped to its sentence, its evidence, its owner and where it came from. Publish refuses a draft it cannot trace.
Centralize your SEO

You can do parts of this with Ahrefs, Semrush, Frase, Surfer. None of them do the whole pipeline.

And the parts they do are wrapped in a UI that flattens the data into charts you cannot pipe into the next step.

  1. Replace the whole stack

    Keyword research, search volume, SERP analysis, competitor gaps, on-page audits, rank tracking: the same operations you pay four tools for, run from one folder against the raw source data.

    vs. $400+/month across Ahrefs, Semrush, Frase, Surfer

  2. Wholesale, not retail

    Every keyword volume, every SERP, every question, as raw data at the price the big tools pay before the markup. A full pipeline run costs $3 to $5 of data.

    vs. $200/month for a single tool

  3. You own the pipeline

    You see the raw SERP. You add custom logic, like flagging every question where an AI answer cites someone else. No UI flattens it; the files are yours.

    vs. siloed tools you cannot chain

  4. Built to be cited

    The pipeline flags the questions AI answers already pick sources for, and structures content to be citable: clear answers, named entities, clean architecture.

    vs. ranking metrics that miss AI surfaces

  5. Structure over articles

    Every subtopic, entity and internal link planned as one architecture: the comprehensive coverage Google rewards over one perfect page.

    vs. a content calendar of disconnected articles

  6. Run it again next quarter

    Point it at a new domain, or re-run the same one in three months and see exactly what moved. No per-seat, no per-project fee.

    vs. metered tools that bill per site

One workflow to rule them all

One engine.
Four kinds of site.

Every course teaches the same engine, applied to a different kind of site. Sold separately or as a bundle.

  1. 01

    Blog

    The full pipeline for content sites: topic map, coverage audit, site architecture, drafted articles.

    • topic hubs
    • article clusters
    • internal links
    $149 one timeGet access
  2. 02

    SaaS

    Feature, comparison and use-case pages, built around how buyers search for software.

    • feature pages
    • comparison pages
    • use-case pages
    $149 one timeGet access
  3. 03

    Ecommerce

    Category, collection and buying-guide pages, organized by buying intent.

    • category pages
    • collection pages
    • buying guides
    $149 one timeGet access
  4. 04

    Local & service

    Service and area pages for businesses that depend on local search and Google Maps.

    • service pages
    • area pages
    • local questions
    $149 one timeGet access
The bundle

All four, $596 $399

Save $197versus buying the four separately

The same engine on every kind of site you will ever run, so a new project never means a new course. One payment, no renewal, and every future update to all four.

  • All four courses, start to finish
  • Every prompt, tested and ready to paste
  • Yours to keep, updates included
Get the bundle
The curriculum

Six phases, the content engine,
and the extras that keep your work.

  1. 00

    Strategy & Setup

    Claude Code installed, the data feed wired, your project folder set. Everything ready for the first run.

    ~30 min setup
  2. 01

    The Universe

    The whole topic mapped: every subtopic, entity, and question Google ties to your niche, including the gaps keyword tools never show.

    universe.json
  3. 02

    Entity Map

    Your site crawled and graded against that map. Every gap, orphan, and off-topic page named.

    entity-map.json
  4. 03

    Question Clusters + The GEO Pass

    Every question worth answering, ranked, with the GEO-critical ones flagged: where AI cites your competitors and not you.

    questions.json
  5. 04

    Architecture

    The map turned into a blueprint: hubs, spokes, and the exact internal links that make coverage compound.

    architecture.json
  6. 05

    Prioritization

    The whole build sequenced into waves, so you always know what is next: quick wins first, big clusters after.

    build-list.json
  7. 06

    Content Generation

    Per-page briefs built from every upstream file, plus your first cluster drafted and ready to publish.

    output/
  8. The Content Engine

    Evidence assembled first, the draft written from it, then a traceability check that refuses to publish a claim it cannot follow back to its source.

    sources.json
  9. +

    Bonus · WordPress automation

    The finished plan pushed straight to WordPress: drafts, meta, internal links, and redirects, in one session.

    automation
  10. §

    Annex · Session logging

    A simple Notion project log so any session resumes cleanly and recovery is one prompt.

    project log
FAQ

Questions,answered.

Do I need to know how to code?

No. You talk to Claude Code in plain English; it runs the tools and connects straight to the data for you. You give instructions, not API calls.

What does it cost to actually run?

Roughly $3 to $5 of pay-as-you-go data for the whole pipeline on one domain. The data vendor has a $50 minimum top-up, but funds never expire and unused funds are refundable, so your real cost is what you actually use. You pay wholesale, per call, with no monthly subscription.

Is this another "topical authority" theory course?

No. It is the Monday-morning version: a folder on your machine that turns a domain and a seed keyword into a complete site plan. Concrete files, not vibes.

Why not just use Ahrefs or Semrush?

You can do parts of it. None do the whole pipeline, and their UIs flatten the data so you cannot pipe one step into the next or add your own logic.

Will this work for my niche?

If your seed keyword is broad enough to have a topic around it, yes. Every course runs its worked example on a real site, start to finish.

Does it optimize for AI citations?

Yes, by construction. It flags GEO-critical questions (where competitors get cited and you do not) and structures content to be the cited source.

Stop guessing what to write.
Run the pipeline.

A domain and a seed keyword in. A complete, prioritized, citation-ready site plan out, on your own machine, in one run.

From $149, one time. No subscription.

Newsletter

Stay ahead of AI-driven SEO.

Occasional field notes on running SEO with AI, new tactics worth stealing, and course updates. No spam, unsubscribe anytime.

We email a confirmation link first. See our Privacy Policy.