This Blog Is Written and Published by AI — Rokuse's Full Pipeline Revealed

This article is itself an output of the pipeline

In Running Content Production with AI — Rokuse’s Full Operations Playbook, we disclosed the high-level strategy. This article is the “blog production line” detail — a full account of how this very article was generated and published.

Read it as a meta article if you like, but it is also a reproducible design document. “Making our own system publicly visible” is what makes it a genuine demonstration of Rokuse’s content production service.

The full pipeline — 7 steps

Rokuse’s automated blog publishing runs on the following 7-step process:

Humans are only responsible for the final step.

Step 1: Backlog lookup and article selection

docs/content-backlog.md is the single source of truth for “what to write next.” The design publishes 3 articles per batch (one execution), with publication order managed in a table like this:

| Batch | Article 1                    | Article 2            | Article 3                   |
|-------|------------------------------|----------------------|-----------------------------|
| B01 ✅ | discord-bot-capabilities-map | slack-bot-capabilities-map | ai-native-content-operations |
| B02   | discord-bot-serverless-workers | slack-bolt-socket-mode | ai-blog-automation-pipeline |

Each time Claude Code is triggered on a schedule, it takes the first 3 uncompleted ([ ]) articles from the front of the batch queue and begins generating. After completing, it marks those rows as [x] and includes the update in the PR.

Step 2: Japanese article generation

Article structure and body text are generated by Claude Code (claude.ai/code). The rules for generation are written in the repository’s CLAUDE.md and .claude/skills/blog-bilingual/SKILL.md, which the AI reads before writing.

Key rules include:

• Headings from ## downward only (H1 is rendered by the page template)
• 3–5 FAQ questions in frontmatter (auto-output as FAQPage JSON-LD)
• 3+ internal links (for topic cluster strengthening in the knowledge graph)
• Technical articles with code snippets require human validation

This article (ai-blog-automation-pipeline.md) was generated following these same rules.

Step 3: Thumbnail generation

Rokuse’s thumbnails are designed without AI image generation. Adding an entry to scripts/blog-thumbnails.config.mjs and running bun run thumbnails produces a deterministic SVG:

{
  file: 'ai-blog-automation-pipeline',
  eyebrow: 'PRODUCTION',
  watermark: 'PIPELINE',
  lines: [
    { text: 'AI writes and' },
    { text: 'auto-publishes.', highlight: true },
    { text: 'Full pipeline revealed' },
  ],
  description: 'Revealing how this blog is automatically generated and published.',
},

Text-based SVGs maintain visual consistency across mass production. A future article (“Text-Based Thumbnail Generation”) will cover this in detail.

Step 4: English translation

Once the Japanese version is complete, the English version is generated following the translation rules in .claude/skills/blog-bilingual/SKILL.md.

• src/content/blog/{category}/{slug}.md (JP) → src/content/blog/en/{category}/{slug}.md (EN)
• Add lang: en and translationOf: {category}/{slug} to frontmatter
• Convert internal links from /blog/{path} to /en/blog/{path}
• Apply glossary (e.g. 六瀬合同会社 → “Rokuse LLC”, コミュニティ → “community”)

The page template automatically outputs <link rel="alternate" hreflang> tags and a JP/EN switcher button on each article.

Step 5: Knowledge graph update

After all 3 articles are generated, bun run knowledge-graph runs. This command:

1. Analyzes the internal link structure across all articles
2. Updates docs/knowledge-graph/graph.json, graph.mmd, and report.md
3. Detects orphan articles (those with few incoming links) and adds cross-links from existing articles

This knowledge graph ensures that even as articles are mass-produced, they remain interconnected as an organic topic cluster. The mutual links between What You Can Do with Slack Bot and API and What You Can Do with Discord Bot and API were designed by this graph system.

Step 6: PR creation

Git commit, branch push, and GitHub PR creation are all automated. PR titles follow the format “Add content batch B02 (3 articles, JP+EN)”, with the body listing which articles are included and the outcome of the quality check (bun run build).

Human reviewers see the diff in the PR, comment on any factual errors, code issues, or tone problems, and the AI applies fixes before the reviewer merges.

Step 7: What humans are responsible for

Even with automation, the following always require human review:

• Fact-checking proper nouns, statistics, and prices: AI can hallucinate
• Code snippet validation: Bot/API code should actually be tested
• Sensitive content review: Ensure no inappropriate tone crept into auto-generated output
• Merge decision: Once the review is satisfied, merge triggers automatic deployment to Cloudflare Pages

Deployment runs on the main branch push trigger in Cloudflare Pages.

Why we’re making this public

“If it’s written by AI, can it be trusted?” is a reasonable reaction. But Rokuse’s position is actually the opposite.

Publishing the system we actually use is the strongest proof it works. Mass-producing content with AI while maintaining quality requires precisely designed rules, guardrails, and human intervention points. Demonstrating that this design is in place is what the transparency of this pipeline provides.

Designing “what not to automate” is harder than “what to automate” in content production. Rokuse uses its own production as a live demonstration while helping clients build their own systems.

Summary

• Rokuse’s blog is automatically generated in Japanese and English and published by Claude Code running on a schedule
• Humans are responsible for only one of the 7 steps: final review and the merge decision
• Backlog management, article generation, translation, thumbnails, knowledge graph, and PR creation are all automated
• Explicitly defining what is not automated (fact verification, code validation, sensitive content review) is the key to maintaining quality

If you’re looking to design or launch an AI-driven content production system, see Rokuse’s Content Production Service.

Related articles

Running Content Production AI-First — Rokuse's Production Operations, Fully Disclosed

A pillar article that fully discloses how Rokuse runs blog, audio, and design production "AI-natively," split across the three lines — text, audio, visual — and the steps where humans must stay in the loop.

rokuse.com/en/blog/production/ai-native-content-operations

What Can Discord Bots and APIs Do? — A Complete Map for Automating Community Operations

A pillar article mapping what Discord bots and APIs can do across four areas — messaging & UI, members & roles, moderation, and events/voice/analytics — including when to use off-the-shelf bots versus building your own (vibe coding), for operators and developers.

rokuse.com/en/blog/community/discord-bot-capabilities-map

What Can Slack Bots and APIs Do? — A Complete Map for Automating Community Operations

A pillar article mapping what Slack bots and APIs can do across four areas — messaging & UI, reactions & automation, operations & management, and integration & AI — from no-code Workflow Builder to building your own (vibe coding), for operators and developers.

rokuse.com/en/blog/community/slack-bot-capabilities-map

How to Build a Discord Community — A Complete Guide from Server Design to Launch and Retention

A complete guide for first-time operators on launching a community on Discord, walking through purpose design, channel structure, role permissions, entry design, moderation, the launch, and retention in practical order.

rokuse.com/en/blog/community/discord-community-setup-guide

Expressing Community Operation in Code — Implementation Patterns for Policy, Moderation, and Invite Flows on Slack

We explain implementation patterns for declaratively managing operational policy, moderation, and invite flows in code on top of Slack. We show concrete file structures, automation designs, and a phased adoption path for moving from person-dependent operation to a reproducible system.

rokuse.com/en/blog/community/community-as-code-slack