3-AI Content Review Pipeline

01 Claude Draft generation

Role

Generates the article draft using project memory and past articles for context consistency

Strength

Project memory Long-form consistency Structure

Constraint

Does not review its own output. Generation and evaluation are separated by design

— run in parallel (two browser tabs) —

02 Gemini Gem

AI-pattern detection
+ fact verification

Detects

Stock phrases, over-abstraction, redundancy, structural repetition — patterns the generator can't see in its own output

Fact check

Verifies proper nouns, specs, and figures against current sources. Pulls information outside Claude and ChatGPT's memory

Note

Removing AI-ness ≠ casualization. Direction is: cut shorter, assert directly, ground in physical objects

03 ChatGPT

Style consistency check

Checks

Rhythm, register consistency, assertion/hedge balance, meta-perspective shifts, sentence-final patterns

Reference

ALL_SAMPLES.md — 5 past articles, 16,234 characters across 5 genres, loaded into Custom GPT

Output

Rates A–E axes as consistent / needs adjustment / inconsistent. Quotes the specific passage

04 Claude Integrate feedback + rewrite

Integration rule

When the two AIs conflict, defer to whichever is closer to the style samples. Don't touch passages that weren't flagged

Output

Changed passages marked with 【】. Overall argument and structure preserved

05 Me Final judgment → publish

Judgment basis

29 years of design instinct. Whether the AI's verdict of "this sounds like you" is accurate can only be confirmed by the person it's supposed to sound like

Publish to

Studio CMS LinkedIn X