How I Built My Own Xiaohongshu Workflow with Codex

The most exhausting part of running Xiaohongshu (RED) is not writing a single post, but having to re-do topic selection, title creation, cover design, typesetting, hashtags, and data analytics for every single piece.

For many creators, the content production flow is chaotic:

I eventually realized that Codex is not just for programmers writing code. If you treat Xiaohongshu content production as a "project", Codex can help you maintain:

• Topic Pool
• Title Library
• Cover Copywriting
• Image-Text Scripts
• Body Copy Templates
• Hashtag Library
• Data Analytics Sheet
• Optimization Recommendations for the next post

Its true value lies not in making you "burst with sudden inspiration", but in helping you standardize and automate repetitive actions.

This article shares a Codex-powered Xiaohongshu workflow that even beginners can easily copy and implement.

1. Build a Content Project Directory First

Don't just open Codex and say: "Help me write a Xiaohongshu post." Doing this makes you start from scratch every single time.

Instead, establish a permanent project directory (e.g., xiaohongshu-workflow) containing 5 foundational files:

This step is crucial because Codex is built to operate within a project workspace. Once you put your materials inside, it can continuously update, reorganize, generate, and review content based on these files.

2. First Fixed Instruction: Weekly Competitor Learning

You shouldn't rely solely on your own ideas. Regularly study high-performance competitor content to analyze why they earn bookmarks, likes, and comments.

Ask Codex to update your 03_Viral_Deconstruction.md once a week by feeding it titles, cover copy, screenshots, and comments:

Please read the content in 03_Explosive_Deconstruction.md and combine it with the 10 new RED/Xiaohongshu examples I added.

Please output:
1. Common points of high-like content this week
2. The most common title structures
3. The questions users care about most in comments
4. Which topics are suitable for my account reference
5. Update 02_Topic_Pool.md

Note:
Do not copy the original text.
Only extract the structure, topic directions, and user needs.

The goal of competitor deconstruction is not to copy, but to decode the underlying user engagement rules (e.g. lists, lists to avoid, beginner-friendly checklists).

3. Second Fixed Instruction: Topic Filtering

With a topic pool in place, stop choosing what to write based on sudden impulses. Ask Codex to filter out 5 specific directions from the pool:

Please select 5 RED/Xiaohongshu topics suitable for release this week based on 01_Account_Positioning.md and 02_Topic_Pool.md.

Requirements:
1. Oriented towards beginners
2. Each topic must have a clear pain point
3. Give 3 titles for each topic
4. Mark whether it is suitable for image-text or video
5. Mark the recommendation rating: 1-5 stars
6. Explain why it fits this account
7. Do not choose exaggerated or hard-to-verify topics

Once the topics are selected, pick only the single most stable one using simple criteria: Is the user pain point strong? Do I have real experience? Can I provide step-by-step instructions?

4. Third Fixed Instruction: Draft Generation

The biggest mistake when using AI for drafting is asking it to write the whole post immediately, which often results in textbook-like manual descriptions. Instead, instruct Codex to structure it layer by layer:

Please generate a RED/Xiaohongshu image-text draft based on the topic I selected.

Selected Topic: [Enter topic here]

Please output:
1. 10 Titles
2. 5 Cover main titles
3. 8-page image-text structure
4. Title and key points of each page
5. Body copywriting for release
6. 10 Hashtags
7. 3 Pinned comments copy

Requirements:
1. Catchy opening
2. Conversational tone
3. Understandable for beginners
4. Do not exaggerate effects
5. Do not promise income
6. Only one key point per page
7. Include interaction question at the end

Xiaohongshu image-text posts are not compressed long-form articles. They require one explicit hook and one core info point per page.

5. Fourth Fixed Instruction: Visual Layout Specifications

Although Codex doesn't design graphics, it can construct design specs, including text placements, colors, layouts, and image generation prompts:

Please generate image creation specifications for the 8-page image-text draft based on the content.

Requirements:
1. Image size: 1080×1440
2. Provide main title, subtitle, and body points for each page
3. Provide layout suggestions for each page
4. Suggest suitable icons or visual elements for each page
5. Cover text must be short and impactful
6. Do not include elements unrelated to content
7. Output text suitable for copying into design tools

6. Fifth Fixed Instruction: Performance Review

Publishing without reviewing is the fastest way to hit a growth ceiling. Record views, likes, bookmarks, comments, and follows in a simple table, and ask Codex to identify performance triggers:

Please analyze the publishing data in 05_Analytics_Review.md.

Please output:
1. Which type of topics gets higher views
2. Which type of titles gets higher bookmarks
3. Which cover copy performs better
4. What users in comments care about most
5. Which content is worth making into a series
6. Which content is not recommended to continue writing
7. 5 recommended topics for next week

Note:
Do not just look at views.
Bookmarks, comments, and follows must also be analyzed.

7. How Beginners Can Run Version 1.0

Avoid building complex automation systems initially. Run the manual loop first:

1. Step 1: Write positioning. Fill out 01_Account_Positioning.md specifying who you write for and what you solve.
2. Step 2: Collect 20 topics. List raw topics in 02_Topic_Pool.md without worrying about perfection.
3. Step 3: Filter. Use Codex to select 5 candidates.
4. Step 4: Draft one post. Run the full loop for one post: Topic $\rightarrow$ Draft $\rightarrow$ Layout $\rightarrow$ Publish $\rightarrow$ Track.
5. Step 5: Review. Feed the performance back into Codex to receive targeted improvements.