Marketing content creation

Last edited time
Mar 25, 2024 6:59 PM
Reading list
Status
1 case
Tools
tags
marketing

Use case 1 : Write content as Persona follow Tone of Voice

Configuration

  • GPT-4 Turbo preview model was used.
  • Temperature set to 1.
  • Utilized the STAR introduction prompt framework.

Introduction

Writing marketing content is time-consuming, particularly for entrepreneurs managing their own practices. They often engage in marketing activities and generate content to maintain engagement. However, creating this content takes a significant amount of time, leading some entrepreneurs to hire content managers. These hired professionals take notes in various forms (voice messages, text notes, etc.), craft the content, and then return it for review. This content is then distributed across different social media platforms, each requiring different formats and catering to unique audiences. The process of tailoring content for each platform adds additional time to post-writing due to the varying presentation requirements.

Hypothesis

The hypothesis suggests utilizing ChatGPT to generate content, aiming to eliminate the need for extensive human intervention. It addresses the concern that direct use of ChatGPT often results in content that feels artificial, highlighting the challenge of creating material that doesn't seem crafted by a robot. The hypothesis proposes creating a tone profile based on the user's previous materials and analyzing the audience to create a model of a typical reader. This approach aims to produce posts that closely replicate those written by a professional human by mimicking the person's tone and understanding the audience's preferences.

image

Solution Implementation

Step 1: Sampling Before anything, we gather between 6 to 10 existing posts from the Author. It’s crucial to choose a diverse set of topics to avoid topic-specific bias in our analysis. This data will be used to understand and mimic the user’s unique tone and style effectively.

Ideally, after collection, split these materials into two equally sized datasets. One dataset will be used to extract the Tone of Voice. Another dataset will be used as examples(few-shot prompting) to improve accuracy.

Step 2 Building Tone of Voice Framework: The process begins with the need to evaluate the author's tone of voice accurately. Recognizing the challenge of assessing tone without professional expertise in literary criticism or copywriting, the strategy involves leveraging Large Language Models (LLMs). The initial step requires submitting a request to the model to "Build a framework to evaluate voice tone." This action generates a comprehensive set of criteria that can guide the evaluation of an author's tone. If necessary, further requests can be made to refine or expand this framework, ensuring it aligns closely with the project's specific needs.

Prompt to build Tone of Voice Framework

💡
After you execute the prompt above. You can ask the model several times “Do you have anything to complement?” And pick up more items to your framework.

Step 3. Applying the Tone Framework: The next critical step entails applying the previously developed tone of voice framework to the first dataset of selected posts. By analyzing these examples through the lens of the established criteria, it's possible to fill the framework with concrete data unique to the author's style. This process effectively translates the theoretical aspects of the tone framework into a practical, data-informed profile of the author's voice. This profile is a key artifact used in the subsequent steps of creating tailored marketing posts.

Prompt to describe Author’s Tone of Voice

💡
For a large framework, consider analyzing each section individually or in groups of 2-3 sections instead of having LLM complete all fields in one request.

Step 4: Building Persona Framework: Personas are crucial for creating content that aligns with the author and the audience. This step includes creating a framework for two persona profiles: the author's and the audience's. LLM can help create this framework, which should include details like age, education level, income, interests, professional activities, etc. I will use my existing framework.

Using this framework, we can build detailed personas based on available data, such as public social media information and personal interactions. This helps make the content creation process more personalized.

Prompt to create Persona

Step 5: Transcription and Expansion of Content: New post content originates from the transcription of voice or text notes from the author.

Once we establish the Audience Persona, we can emulate its behaviour to predict, ask questions, and express concerns. This stage requires using the LLM to formulate relevant questions that an audience member might have after going through the content. The author can respond to these in written form or via additional voice notes. This method enhances the initial content, creating a more captivating and thorough post.

Step 6: Combining Elements for Post Creation:Once all elements—Tone of Voice, Author's Persona, Audience Persona, and enriched content—are ready, the next step is to combine them to create draft posts. This is done by sending a request to the LLM that merges the tone, personas, and content into unified marketing posts. We improve the model's precision in replicating the author's style by including examples from a second dataset of posts (not used for tone analysis) through few-shot prompting.

Final Prompt

Evaluation of Results

The results of this automated content creation are evaluated in several ways:

  1. Author's Evaluation: The main assessment comes from the author. They review the posts to ensure that the content aligns with their style and communication preferences. They check for authenticity, tone consistency, relevance, and potential engagement.
  2. Multiple Variant Generation: The system generates several versions of a post, letting the author choose the one that best represents their message or that will resonate with their audience. After choosing the best content variant, the author can make minor edits to refine the tone or adjust details. This ensures the final post matches their style and requirements.
  3. Efficiency & Cost-effectiveness: The efficiency and cost-effectiveness of the content creation process are also considered. The author compares this method to traditional ones in terms of time, effort, and costs. This automated process aims to reduce these factors, making content creation more accessible and manageable.
  4. Stress Reduction & Accessibility: This process reduces the stress and perceived barrier associated with content creation. The system makes it easier and less time-consuming for the author to generate posts, enabling more frequent and diverse content interactions.

In summary, results evaluation is a comprehensive process that includes the author's subjective assessments and objective measures of efficiency and impact.

Next steps

  1. Full Process Automation: Implement an automated system to handle transcription of voice notes, tone and persona analysis, content variant creation, and author feedback. This minimizes manual work and accelerates content creation.
  2. Prompt Refinement with Professional Insights: Involve expert copywriters and literary critics to improve the tone assessment and persona framework. Their knowledge will increase the content's accuracy, authenticity, and appeal to the audience.
  3. Incorporation of Author Feedback Loop: Establish a system for authors to offer direct feedback on content. This will allow for stylistic preferences and identification of undesirable phrases, personalizing the content to even better reflect the author's unique style.
  4. Multi-Network Content Adaptation: Build a system that can create content variants for different social networks simultaneously. Each platform requires unique content to engage its specific audience.
  5. Voice Note and Interaction Optimization: Enhance the efficiency of voice note recording and analysis. Incorporate a system that can prompt and record responses to generated questions immediately.