ChatGPT
Plan: Controlled test plan to be recorded
Protocol ready · controlled run pending
Tool detailsChatGPT vs Claude vs Gemini: Korean Writing Test
Compare a source-based Korean Naver blog draft under identical editorial constraints.
Common prompt
Using only the supplied source notes, write a 900–1,000 character Korean Naver blog post for beginners. Use a neutral helpful tone, five short sections, one checklist, and no invented statistics. Avoid translated phrasing and promotional superlatives.
Controlled conditions
- New chat with memory and custom instructions disabled.
- One generation only, followed by a timed human edit.
- The same source packet and temperature-equivalent default are used.
Published criteria
Instruction accuracy
25%Follows format, length, tone, and prohibited-expression rules.
Korean naturalness
30%Counts awkward particles, translations, spacing, and honorific errors.
Editing effort
25%Measures minutes needed to reach publishable quality.
Cost efficiency
20%Compares usable output against plan and run cost.
Tool records
A controlled test has not been run yet. Raw outputs and scores are therefore hidden and do not affect tool ratings.
Claude
Plan: Controlled test plan to be recorded
Protocol ready · controlled run pending
Tool detailsGemini
Plan: Controlled test plan to be recorded
Protocol ready · controlled run pending
Tool details