Bootstrapped · Seoul, Korea · Est. November 2025

AI content operations for Korean publishers, with a human in the loop.

SSTACK turns search demand into accurate, readable Korean articles. Claude drafts and checks; a person approves before anything is published. We run the pipeline every day on our own channels and are packaging it for other independent publishers and small businesses.

See the workflow Our channels

The problem

Korean solo publishers spend hours per article on keyword research, fact checking and image work. Most "AI writer" tools skip verification and produce generic, error-prone text.

Our approach

Small, auditable steps: demand research, title research against the live top results, a fixed article spec, a fact and consistency QA gate, and human approval.

Who it is for

Independent Korean bloggers, local SMBs and info publishers who want consistent, honest content without hiring a content team.

How the pipeline works

  1. Demand research. Scheduled jobs collect candidate topics from Naver Creator Advisor and Google Trends every few hours, dedupe them and score them by freshness and search demand.
  2. Title research. For each topic we review the current top 15 Naver results and write titles that match real search intent. Risky topics (health claims, crime victims, personal rumors, unverified financial advice) are skipped.
  3. Drafting with Claude. Claude writes the article from a fixed spec (about 1,300 Korean characters, four sections) using only facts gathered for that topic.
  4. Quality gate. A second pass checks facts, figures and dates, and consistency between title, body and images. Failing drafts are regenerated or dropped, never published.
  5. Human approval and publishing. The founder approves each batch. Every post is logged (date, URL, keyword, length, image count) for auditing and performance tracking.
SSTACK publishing log: every article is recorded with date, post ID, URL, keyword, content type and character count
Live operations log (Google Sheets). Every published article is recorded with its date, post ID, URL, keyword and length.

Where Claude fits

Today

  • Claude Sonnet 5.5 is our primary drafting model. Other models are used only as fallback when Claude is rate limited.
  • Claude Projects hold our writing spec and style references so output stays consistent across hundreds of articles.

Next (building now)

  • Move drafting and QA to the Claude API with structured outputs, so each article returns as validated JSON (title, sections, sources, image briefs).
  • Use Claude as an automated fact and consistency reviewer before human approval, with prompt caching for our shared spec and the Batch API for nightly runs.
  • Offer the pipeline as a hosted workflow for other Korean publishers, with a human approval step built in.

Traction

265,000+lifetime visits on our flagship Naver blog, 팁저장소 (Oct 2026)
109articles shipped through the pipeline since its launch on Oct 6, 2026
200+topic candidates screened per day before anything is written
RevenueSelf-funded by advertising revenue (Naver AdPost) from our own channels

Our channels

팁저장소 Naver blog showing 265,502 total visits and 182 neighbors
Flagship channel 팁저장소 on Naver Blog.

Content principles