Use ← → to navigate

Kagels Trading × Daniel

Article Rewrite SEO Skill

An AI-powered tool that rewrites your published articles for the 2026 AI-search era — with your data, your voice, and your existing rankings preserved.

A guide for Karsten

The Workflow

9 phases, run in order

1Read article
2Pull GSC
3Pull SEO data
4Match video
5Brief + your input
6Rewrite
7Auto-QA
8Editor pass
9Ship

Phase 5 is a pause point — the skill stops and waits for your data before writing anything.

Each phase produces a file. If anything goes wrong at phase 7, the skill loops back automatically and tries to fix itself before bothering you.

Phases 1–3

Research: deeper than any human would do by hand

Phase 1

Read your article

Pulls the full article from your WordPress site automatically. Locks in your URL slug so it never changes (changing slugs destroys SEO equity).

Phase 2

Google Search Console

Looks at exactly what people are searching to find this page right now — including queries you almost rank for but don't quite reach page 1.

Phase 3

DataForSEO

Pulls 8 different datasets: keyword volume, the AI Overview Google currently shows, what ChatGPT/Claude say about your topic, who your real competitors are.

All German-market data. All current. No assumptions, no guesses.

Phases 4–5 — The Pause Point

The skill stops here and asks for your data

This is the most important step. Without your proprietary inputs, the article will look like every other AI-generated finance article — and Google will penalize it.

You'll be asked for things like:
  • A real trade screenshot from TradingView
  • The actual win-rate of one of your indicators
  • A first-hand anecdote from your trading history
  • An annotated chart you marked up yourself

If you can't provide any of these, the skill will mark the gaps with [INSERT KARSTEN: ...] placeholders so you can fill them in later — but the article won't be its strongest until you do.

Phase 6 — The Rewrite

Plan first, then edit. Not "regenerate from scratch."

This is the part most AI tools get wrong. Generic AI tools throw away your whole article and write a new one — losing your voice and your good content. This skill works like a senior editor would.

A
Plan agent reads everything Looks at your article, the data, the brief — then decides which sections to KEEP, which to LIGHTLY EDIT, and where to ADD new material.
B
Apply step executes the plan Your article is treated as a canvas. Sections marked KEEP are copied untouched. Edits are small, surgical — one sentence at a time, never a whole section regenerated.
C
Polish read at the end One final pass to make sure the whole thing flows together as one article — not a Frankenstein of stitched-together edits.

Phases 7–8

Quality control: a robot pass and a human-style editor pass

Phase 7 — Auto-QA

The robot pass

Runs ~25 mechanical checks: title length, keyword presence, broken links, banned buzzwords, German anti-patterns ("die Edge" not "der Edge"), price-leak detection. If something fails, it tries to self-heal up to 3 times before stopping.

Phase 8 — Editor Pass

The human-style pass

An AI agent reads the article 7 times like a senior editor would: cold read, goal check, fact spot-check, cohesion. Returns a verdict — READY, NEEDS POLISH, or SEND BACK — with specific fixes if needed.

Both passes run automatically. You only see results after the article is as clean as the skill can make it.

Output Structure

What you get back: 3 folders, German labels

✅

1-zur-freigabe

Everything you need to review and approve. The article, an English translation, a side-by-side diff against the original, the editorial review.

FOR REVIEW
📤

2-zum-veroeffentlichen

Once approved: the final article ready to paste into WordPress, the title and meta description, image alt texts, the schema block.

FOR PUBLISH
📊

3-daten-und-logs

All the research data, GSC/DataForSEO results, the QA report, cost ledger. You don't normally need to open this — it's there for transparency and for Daniel to debug if something goes wrong.

DATA & LOGS

Setup — One-Time

What you'll need to set up (only if you run it yourself)

1
Claude Code The AI tool that runs the skill. Free desktop app. You already have this installed.
2
DataForSEO API key Paid service, ~$0.30–0.80 per article in actual SEO data costs. You sign up at dataforseo.com and add the key to a settings file. One-time setup, ~5 minutes. Daniel can help.
3
Google Search Console connector Free. Lets the skill see your real search data for the article. Already set up if Daniel has worked on your GSC before — otherwise a one-time browser login.

If Daniel runs it for you, you skip all of this and just send him the URL.

How to Run It

Two paths. Pick whichever fits your week.

Option A — Hands off

Daniel runs it for you

You send Daniel:

  • The article URL
  • Your trade screenshot / win-rate / anecdote (when asked)

Daniel sends back the 3 markdown files: the rewritten article, an English translation, and a change log. Full data + logs available on request.

Option B — Hands on

You run it yourself in Claude Code

In Claude Code, just type the word rewrite and paste the URL. The skill auto-triggers, no special command needed.

rewrite https://www.kagels-trading.de/krypto-prognosen/

It runs phase by phase. When it pauses for your inputs, you paste them. You get the same 3 folders saved locally.

The Most Important Slide

Your inputs are the difference between ranking and getting penalized.

"Generic AI rewrites of finance content get demoted by Google. Articles with first-hand expertise rank."

— Lily Ray, Neil Patel, Edward Sturm (consensus)

What "first-hand expertise" looks like in practice:
📸
ScreenshotsA real TradingView chart you marked up yourself. Shows you've actually done this trade.
📈
Numbers"My Fibo Cross indicator hit 64% win-rate on BTC daily over 12 months." Specific. Verifiable.
💬
Anecdotes"In March 2026 I spotted a Wyckoff accumulation on DAX..." Stories Google can't generate.
🎯
AnnotationsAn original chart with your notation showing exactly where you'd enter and exit.

Cost Per Article

Transparent. Tracked automatically.

~600k
AI tokens used per article
(reading, planning, writing, reviewing)
~$0.30–0.80
DataForSEO costs per article
(actual USD, not estimate)
Every run produces a cost-summary.txt in the data folder showing exactly what was spent. No hidden costs. No surprises.

After You Get Draft 1

Want changes? Just say what you want, in plain language.

You don't need to know how the skill works internally to ask for edits. The article is a normal markdown file you (or Daniel) can change directly.

Examples of edits you can request

Plain-language requests

  • "Shorten the intro by half"
  • "Add a paragraph about XRP halving cycles"
  • "Use this trade screenshot in section 4"
  • "Replace the Bitcoin example with Ethereum"
  • "Make the tone more cautious"
When to re-run the skill vs. edit by hand

Rule of thumb

Edit by hand: small changes, one or two paragraphs, fixing a fact, adding a screenshot.

Re-run the skill: if Google updates a key dataset, if you write a totally new section worth integrating, or after 12 weeks once you have ranking data.

Important — Set Expectations

This is an ongoing project. Not a finished product.

What that means in practice:
  • It will sometimes make mistakes
  • The first draft won't always be perfect
  • It needs your judgment, not a rubber stamp
  • Every article we run improves the skill
What we're doing about it:
  • Tracking what goes wrong, fixing the script
  • Patterns from your feedback shape the next version
  • The skill gets better the more we use it
  • By article 5, it'll be far sharper than article 1

Think of it like a new junior editor on the team — competent, fast, eager to improve, but still needs review.

Let's Run One

Ready to try it?

Pick an article that's losing clicks, prepare one or two pieces of your own data (screenshot, win-rate, anecdote), and we run it together.

1
Pick an articleSuggested first one: /elliott-wellen-analyse/ — your strongest E-E-A-T story.
2
Gather your dataOne screenshot, one number, one short story. That's enough for the first run.
3
Send Daniel the URLOr run it yourself in Claude Code. Either way, you get the review folder back within an hour.

Questions? Send Daniel a message.

1 / 14