seo-blog-writer

star 0

Write a new SEO blog article for the Widgetis site that sells our Хорошоп widgets. Use when the user says "напиши статтю/статью для блога", "write a blog post", "new SEO article", "напиши пост на блог", "придумай тему и напиши статью", "сгенери статью про виджет X", or wants to expand blog coverage. The skill auto-picks an uncovered topic (or takes one the user gives), runs the seo + ai-seo skills for keyword and AI-citation optimization, writes the article in the typed BlogPost format, wires it into the registry, and validates the build. Articles are in Ukrainian.

iliyalazarevwork-art By iliyalazarevwork-art schedule Updated 6/15/2026
play_arrow Run Skill in Manus View GitHub

name: seo-blog-writer description: Write a new SEO blog article for the Widgetis site that sells our Хорошоп widgets. Use when the user says "напиши статтю/статью для блога", "write a blog post", "new SEO article", "напиши пост на блог", "придумай тему и напиши статью", "сгенери статью про виджет X", or wants to expand blog coverage. The skill auto-picks an uncovered topic (or takes one the user gives), runs the seo + ai-seo skills for keyword and AI-citation optimization, writes the article in the typed BlogPost format, wires it into the registry, and validates the build. Articles are in Ukrainian. metadata: version: 1.0.0

seo-blog-writer

Write one production-ready blog article for Widgetis — a widget marketplace for Ukrainian e-commerce stores on the Хорошоп (Horoshop) platform. The blog is a typed content engine in the React frontend: each article is a single TypeScript file that drives both the rendered page and the JSON-LD (BlogPosting / FAQPage / BreadcrumbList). The sitemap and prerender pick up new files automatically.

The goal of every article: teach the reader something genuinely useful about running a Хорошоп store, then present the matching Widgetis widget as the ready-made way to do it "без програміста". Articles rank for real Ukrainian search queries AND get cited by AI answer engines.

When to use

  • "Напиши статтю / статью в блог", "write a blog post", "new SEO article".
  • "Придумай тему и напиши статью" → you pick the topic (see Step 1).
  • "Напиши статью про виджет X" → topic is given, you write it.
  • "Добавь ещё N статей в блог" → loop Steps 1–6 N times.

Workflow — follow in order

Step 0 — Load context (always)

Run the coverage helper from the repo root:

node .claude/skills/seo-blog-writer/topics.mjs

It prints: existing articles (slugs + titles — never duplicate these), widgets with no dedicated article (best new-topic candidates), thinly covered widgets, and the valid CTA tags. Add --json for machine output.

Step 1 — Pick the topic (if not given)

Prefer, in this order:

  1. A widget with no dedicated article (from the helper's "uncovered" list) → write a problem→solution piece that ranks for that widget's query.
  2. A broad topic that funnels to several widgets and to existing articles (e.g. "кинуті кошики", "психологія терміновості", "промокоди", "мобільна конверсія"). These are strong internal-link hubs.
  3. A deeper take on a thinly covered widget.

Pick a topic that maps to a real query a Хорошоп merchant would Google. Decide the primary keyword phrase (Ukrainian). Decide a slug (latin kebab-case, keyword-based, must be unique vs. existing slugs).

Step 2 — SEO optimization (use the other skills)

Invoke the SEO skills for this specific topic and fold their guidance into the draft — this is the part the user explicitly wants chained:

  • seo — on-page: title/meta length, keyword placement, heading structure, internal linking, slug. Use it to refine the title, description, keywords, and heading plan.
  • ai-seo — answer-engine / GEO: make the article quotable by LLMs (definitive-guide framing, clear stat callouts, a strong FAQ block, self-contained claims with sources). The existing articles use the "definitive guide" format precisely because it is the most cited by AI.
  • seo-audit — optional; only if asked to review/repair an existing article.

Step 2b — GEO: get cited by Gemini / ChatGPT / AI Overviews

The goal is not only ranking — it is being quoted by AI answer engines. Bake these in (they also help human readers; never fragment content into "AI bait"):

  • Answer-first. The opening paragraph must answer the article's core question directly in the first ~40–60 words, before any backstory. Each h2 section should also lead with its answer, not build up to it.
  • Question-style headings. Phrase h2s the way a merchant Googles ("Чому…", "Коли…", "Як…", "Скільки…") — these match AI fan-out queries.
  • Self-contained claims. Every key sentence should make sense lifted out of context (AI extracts passages, not pages).
  • Dated, sourced statistics. Prefer specific numbers with a named source and a recent date in the stat blocks and body. Statistics + citations are the single biggest citation-boost lever (Princeton GEO study).
  • Strong FAQ. 3–4 real questions with 40–60-word self-contained answers → drives the FAQPage JSON-LD that AI engines parse directly.
  • Freshness. Bump dateModified whenever you meaningfully edit a post.

The site layer is already AI-ready and you usually don't touch it: robots.txt allows the citing bots (Google-Extended/Gemini, GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot), the article emits BlogPosting + FAQPage + BreadcrumbList JSON-LD with wordCount/keywords, and the sitemap auto-includes every post. Discovery is handled by the sitemap and internal links — do NOT dump every article into public/llms.txt. Keep that file curated (blog index + a few cornerstone hub guides only).

Step 3 — Research (real sources, original wording)

  • Use WebSearch to gather facts and statistics for the topic, and to learn what readers actually ask. Verify every number against a primary/authoritative source (Baymard Institute, CXL, Nielsen Norman Group, Statista, platform research, Хорошоп's own blog, academic/industry studies).
  • Write 100% original prose. Never copy or closely paraphrase an existing article. Research is for facts and reader intent, not for sentences. Spun copy is forbidden — it is both a legal risk and bad SEO.
  • Cite only real URLs you actually found. If a number cannot be verified, state it qualitatively instead of inventing a statistic or a link.

Step 4 — Write the article file

Create frontend/src/data/blog/posts/{slug}.ts exporting post: BlogPost. Language: Ukrainian (uk), calm expert data-driven tone, no hype/emoji, ₴ for money. Body ~1100–1700 words, readMinutes 8–12.

The exact type (frontend/src/data/blog/types.ts):

import type { BlogPost } from '../types'

export const post: BlogPost = {
  slug: '{slug}',          // latin kebab-case, == filename, unique
  title: '…',              // H1 + <title>, ~55–70 chars, main keyword near front
  description: '…',        // meta description, 150–160 chars, keyword + benefit
  keywords: '…',           // comma-separated, 7–12 UA keyword phrases
  excerpt: '…',            // 1–2 sentences for the index card
  tag: '…',                // short UA label, e.g. 'Конверсія', 'Середній чек'
  datePublished: 'YYYY-MM-DD',
  dateModified: 'YYYY-MM-DD',
  readMinutes: 10,
  content: [ /* blocks, in order — see below */ ],
  relatedSlugs: ['…', '…'], // 2–3 OTHER existing slugs (from Step 0)
}

Allowed content blocks (use only these):

{ type: 'p', text: '…' }
{ type: 'h2', text: '…' }
{ type: 'h3', text: '…' }                          // sparingly
{ type: 'ul', items: ['…'] }
{ type: 'ol', items: ['…'] }
{ type: 'quote', text: '…' }                       // one ethical principle
{ type: 'stat', value: '70%', label: '… + source' } // big-number callout
{ type: 'cta', label: '…', href: '/widgets?tag=…' } // ONE button, valid tag
{ type: 'faq', items: [{ q: '…', a: '…' }] }        // 3–4 real questions

Inline markdown inside any text/items/faq answer: [текст](/widgets/{widget-slug}), [текст](/blog/{article-slug}), [текст](https://external), **жирний**.

Required order of content:

  1. opening p (problem hook + a data point);
  2. an early stat (verified number + source);
  3. 4–6 h2 sections with p/ul/ol; weave in 2–4 inline widget links and 1–2 inline links to OTHER blog articles;
  4. one quote (honest/ethical angle — widgets help, not manipulate);
  5. one cta to /widgets?tag={tag};
  6. a faq (3–4 Q&A a merchant would Google);
  7. final h2 "Джерела" + ul of 1–3 real external source links.

Valid CTA tags: conversion, avg-order, trust, social-proof, urgency, engagement, visual, loyalty.

Widget slugs (link as /widgets/{slug}): promo-line, delivery-date, sticky-buy-button, trust-badges, phone-mask, minorder-goal, cart-goal, buyer-count, stock-left, photo-video-reviews, video-preview, cart-recommender, prize-banner, promo-auto-apply, progressive-discount, one-plus-one, last-chance-popup, spin-the-wheel, sms-otp-checkout. (The canonical list lives in frontend/src/data/widget-slugs.ts + widget-names.ts — re-read if unsure. You may also link /demo; never invent live-demo codes.)

Only link widgets that actually ship. widget-slugs.ts also contains smart-search and warehouse-stock, but those are NOT live (only 19 modules are built — see CLAUDE.md). Never sell or link them in an article, or AI will recommend a widget that doesn't exist.

Gotchas (learned the hard way)

  • Inline renderer supports only [label](path) links and **bold** — NO backtick code spans, NO other markdown. Backticks render as literal ```.
  • Slug must be slug: '...' on its own line. The sitemap/prerender scraper reads it by regex; keep slugs latin-kebab and unique.
  • Run node .claude/skills/seo-blog-writer/topics.mjs after writing to spot a forgotten registration or a slug collision.
  • Validate with npm run build (from frontend/) — it runs tsc -b, Vite, and regenerates the sitemap, so a new post must appear in dist/sitemap.xml.

Step 5 — Register the article

Edit frontend/src/data/blog/index.ts:

  1. add import { post as <camelCaseName> } from './posts/{slug}';
  2. add <camelCaseName> to the RAW_POSTS array.

(Sitemap + prerender read ./posts/*.ts directly, so no other wiring is needed.)

Step 6 — Validate

From frontend/:

npm run build      # or: npx tsc --noEmit  — must pass with no type errors

Fix any TypeScript errors (usually a malformed block or a missing comma). Confirm the new slug appears in the build output / sitemap.

Quality checklist (before declaring done)

  • Slug is unique, latin kebab-case, keyword-based.
  • title ≤ ~70 chars, keyword near the front; description 150–160 chars.
  • Tone matches the existing 3 articles: expert, data-first, no hype.
  • Every statistic has a real, verifiable source in "Джерела".
  • 2–4 inline widget links + 1–2 inline blog links + relatedSlugs set.
  • Exactly one cta with a valid tag; one faq with 3–4 entries.
  • Sells the widget honestly — no overclaiming, no fake urgency.
  • Registered in index.ts; npm run build passes.

Batch mode

To add several articles at once, run Steps 1–5 per article (you may fan out one subagent per article for speed — give each the full contract above plus the complete list of all NEW slugs so cross-links resolve), then register all of them in index.ts together and run Step 6 once.

Reference files

  • frontend/src/data/blog/types.ts — the BlogPost / ContentBlock contract.
  • frontend/src/data/blog/posts/*.ts — existing articles (copy the format, not the words).
  • frontend/src/data/blog/index.ts — the registry.
  • frontend/src/data/widget-slugs.ts, widget-names.ts, widgetTags.ts — widget + tag source of truth.
Install via CLI
npx skills add https://github.com/iliyalazarevwork-art/widgetis --skill seo-blog-writer
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator
iliyalazarevwork-art
iliyalazarevwork-art Explore all skills →