A Three‑Step Method to Rebuild Your WeChat Article Workflow so AI Mirrors Your Style

1. Building a Knowledge Base

The first step is not tweaking prompts but teaching the AI "who you are" by curating past writings. From over 100 original articles, the author discards irrelevant pieces (e.g., pure lifestyle or unrelated technical details) and extracts style features such as frequently used words, sentence length distribution, paragraph pacing, and typical opening/closing phrasing. The extracted assets are saved as high‑frequency vocabularies, tone characteristics, chapter‑progression patterns, and opening/closing habits.

Next, the author extracts the stylistic fingerprint: consistent first‑person narration, a judgment‑then‑reason structure, and a final action recommendation. These become reusable style templates.

Finally, the author solidifies the analysis into four asset categories—high‑frequency terms, tone traits, chapter‑progression methods, and opening/closing habits—so future drafts can reuse them without re‑training each time.

2. The Three‑Step Writing Process

With the knowledge base ready, the author structures writing into three repeatable stages to avoid AI hallucinations and ensure consistent output.

Step 1 – Blueprint

Convert voice recordings or meeting notes into a requirement document, then produce a topic blueprint that fixes "what to write, in what order, and which assets to include".

## 1) Goal
Organize input into a topic blueprint; no body text.

## 2) I/O
Follow the unified input/output conventions in `prompts/starter.md`.

## 3) Execution Focus
1. Strictly follow chapter‑title rules in `starter.md`.
2. Do not read whitelist article bodies at this stage.
3. Write “to be added” for missing info; no fabrication.
4. Output any high‑value structural suggestions as `Pending Items`.

## 4) Checklist
1. Topic viability + rationale.
2. Audience + concerns.
3. 3‑5 title candidates + 1 recommendation.
4. One‑sentence main line.
5. Chapter plan: goal, core points (2‑5), required assets.
6. Risk reminder: common drift + mitigation.

Step 2 – Manuscript

Using the requirement document, blueprint, and knowledge‑base assets, generate the first full draft while preserving personal expression.

## 1) Goal
Produce a publishable first draft from `需求文档.md + 01_选题.md`; no outline changes.

## 2) I/O
Follow `prompts/starter.md`. Mandatory input: `knowledge_base/style_system/author_style_card.md`.

## 3) Workflow
1. Write first draft covering each chapter’s goal, core points, and required assets; limit label repetitions (`观点:`/`依据:`/`动作建议:`) to ≤1 per adjacent paragraphs.
2. After draft, compare against `author_style_card.md` and revise any tone, structure, or factual deviations.
3. Perform at least one round of vocabulary and prose optimization using `knowledge_base/style_system/all_reference_whitelist.txt`.

Step 3 – Polish

Finalize the draft by removing AI‑specific phrasing and making the text sound natural, without altering the main line or factual boundaries.

## 1) Goal
Refine `02_撰写.md` without changing the main line or facts.

## 2) I/O
Follow `prompts/starter.md`.

## 3) Workflow
1. Remove AI “flavor” – at least two iterations of locate‑revise‑review.
2. Full‑text oralization – convert explanatory/technical tone to conversational language; create an “AI‑flavor analysis checklist” covering template phrases, slogans, promises, and explanatory tone.
3. Minimum three iterations of analyze‑rewrite‑review; only language changes, no structural or factual edits.

The effectiveness of this loop lies in repeatedly comparing generated output with a high‑quality reference, correcting deviations, and regenerating until the result satisfies the author’s style.

3. Practical Demonstration

Step 1: Voice → Requirement Document

The author converts a Feishu meeting recording into a requirement document, defining chapter directions and key points.

The transcript is fed to the AI, which produces the requirement document.

Step 2: Requirement Document → Blueprint

Using the document, the author creates 01_选题.md, fixing chapter structure, goals, core points, and required assets.

Human review adjusts the AI‑generated outline to keep the direction on track.

Step 3: Blueprint + Knowledge Base → Manuscript

The author generates 02_撰写.md, the first draft that embeds the personal style.

The draft follows the three internal steps described earlier, ensuring each chapter meets the style card and reference whitelist.

Step 4: Manuscript → Polish

The final stage removes residual AI tone and converts the text to natural, conversational language, producing the finished article.

4. Summary

AI’s greatest value is not to replace writing but to formalize the process so busy authors can output consistently.

The decisive factors for quality are a personal knowledge base (the foundation) and a clear, step‑by‑step workflow (the track).

Expect iterative refinement: generate, compare, correct, and regenerate until the output feels truly yours.