"AI-Powered Music Content Strategy for Xiaohongshu: From Title Analysis to GPT Prompt Optimization\”

“AI-Powered Music Content Strategy for Xiaohongshu: From Title Analysis to GPT Prompt Optimization”

AI-generated images are used as video footage, and pleasant music is used as the background music to form video content.

1. Automatic Collection, Crawler/Script

Here, I used a crawler software called “Houyi Collector.” I originally intended to write a tutorial on using Houyi Collector to crawl Xiaohongshu notes, but the Xiaohongshu page seems to have been upgraded recently, causing some issues when using Houyi Collector for crawling. Since the theme of this article is not about data collection, I will share the tutorial once I research a new method.

In summary, after collection, export to an Excel table. Generally, video titles with high likes are well-written, so I use Office to open the table, click on filtering, and sort by “likes” in descending order. This way, the titles, copy, and other content with the highest likes are displayed.

2. Data Cleaning

The copy of the benchmark blogger is templatized, with fixed front and back parts, and only the middle sentence is the real copy content.

💡 If you are also doing Xiaohongshu recently, consider using this modular method to write copy.

Next is data cleaning, which essentially means removing unwanted data and keeping the desired data. Similarly, data cleaning has two methods: manual/automated.

• Manual cleaning: Copying and pasting, deleting by yourself. If the data is small, this method is acceptable. But in the AI era, such repetitive work should definitely be done by automated tools/machines.

3. Title Analysis Dimensions

Initial Analysis Dimensions

1. Title Length
2. Language Style and Tone
3. Structure and Format
4. Problem Solving
5. Target Audience
6. Relevance to Content

Potential Dimensions Supplemented by GPT

1. Culture and Background
2. Timeliness
3. Interactivity
4. Comparison and Contrast
5. Mystery and Curiosity
6. Reference to Other Platform Strategies
7. Brand and Consistency

Filtered Applicable Dimensions

Excluding irrelevant dimensions such as “reference to other platform strategies” and “brand and consistency”, the retained analysis angles more suitable for Xiaohongshu video notes are:

• Title length, language style, structure format, problem-solving orientation, target audience, content relevance, cultural background, timeliness, interactivity, comparison and contrast, mystery creation, and thumbnail influence.

Specific Operation Process for Title Analysis

Filtered Applicable Analysis Dimensions

Retained dimensions: Title length, language style, structure format, problem-solving orientation, target audience, content relevance, cultural background, timeliness, interactivity, comparison and contrast, mystery creation.

Framework for Issuing Analysis Instructions to GPT

💡 When analyzing the pure text characteristics of Xiaohongshu bloggers’ video titles, you can 切入 from the following dimensions:{List of filtered analysis dimensions}

Specific questioning patterns:

1. Provide about 20 high-quality titles (focus on quality over quantity to avoid GPT performance degradation due to token overload).
2. Require listing 3-5 techniques for each dimension, including:
3. If a dimension is not reflected, directly mark “No reflection”.

Application Logic of GPT Analysis Results

1. Data Simplification Principle: Control the number of titles to about 20 to ensure GPT analysis accuracy.
2. Output Requirements: Each dimension corresponds to specific techniques, avoiding vague descriptions.
3. Follow-up Actions: Extract reusable title creation templates through the techniques output by GPT.

Notes

• Token Limitation: The total length of analysis dimensions + title content should be controlled within the efficient processing range of GPT to avoid performance degradation.
• Dimension Adaptability: If a title has no obvious characteristics in a specific dimension (such as “cultural background”), directly mark “No reflection” without forced interpretation.

4. Creating Prompts

The above steps all aim at one goal: What methodology should we use to teach GPT to create, that is, “know how”.

Next, write these methodologies and examples into the prompt. The fastest way is to rewrite the prompts written by experts, such as Teacher JK’s “Happy Moments Circle Jokes Production V0.3”.