Human + AI Collaboration in prompt chaining: 100x quality for LLM output

A simple guide on prompt-chaining with human inputs, even for beginner.

I want to share with you all a tip I regularly use to maximize the output quality when using LLMs.

Prompt chaining may be familiar with some of you. However, my tip take a step further by incorporating “human inputs” into the chain. This method helps maximize AI’s efficiency + human’s expertise and creativity.

I temporarily called it “human-in-loop-capability” in a prompt chain or an AI workflow.

So what does this look like?

Basically, we break down a large task into smaller, manageable subtasks. In each task, we - as human, provide AI with resources to make sure these subtasks be completed with high-quality. The final output we receive from AI will be of highest quality possible.

Let take a practical example of how this technique can be applied to create high-quality SEO blog posts with AI.

We design a process for AI to work just like how a human writes a good blog post:

(1) brainstorming unique ideas → (2) researching (keywords, competitor’s blog post, the topic,…) → (3) drafting the first outline → (4) first draft: writing each subtopic comprehensively → before (5) finalizing the final SEO blog post → (6) on-page SEO optimization.

→ Next, we provide AI in each step with every possible “human inputs” for AI to learn from, and process the results that we expect. Inputs could be a resource, your ideas, your requirements in any form of text, files, images, url.

1- Brainstorming Unique Ideas:

Human inputs: • Company details • Target audience information • Desired topic

In this initial stage, the AI is provided with essential information to generate relevant and unique ideas tailored to the company's target audience. The human reviewer then selects the most promising concept that aligns with the overall content strategy.

2 - Researching:

Human inputs: • Selection of the best idea that matches expectations from AI-generated list • Refinement of AI-conducted research • Focus on most relevant and valuable insights

AI conducts comprehensive research on keywords, competitor content, and topic-specific information. The human collaborator then refines this research, ensuring that the most relevant and valuable insights are prioritized for the content.

3- Drafting the First Outline:

Human inputs: • Additional insights to be included • Suggestions for structure refinement • Guidance on how to best serve the target audience

Based on the research, the AI creates an initial outline. The human reviewer provides additional insights and suggestions to refine the structure, ensuring that the outline is optimized to serve the target audience effectively.

4- Writing Comprehensive Insights:

Human inputs: • Feedback on the AI-expanded outline • Suggestions for changes or additions to content • Specifications on desired tone and style

The AI expands on each section of the outline, creating detailed content. The human collaborator then reviews this draft, providing feedback, suggesting changes or additions, and specifying the desired tone and style to ensure the content aligns with the brand voice.

5- Finalizing the SEO Blog Post:

Human inputs: • Feedback on the near-final draft • Last-minute adjustments to ensure quality standards are met • Approval of overall content structure and flow

The AI refines the draft based on human feedback. The human reviewer then makes any final adjustments, ensuring that the content meets all quality standards and approves the overall structure and flow of the blog post.

6- On-Page SEO Optimization:

Human inputs: • Internal linking strategy guidance • Meta description refinement • Title tag optimization • Additional SEO elements specific to the company's strategy

In this final stage, the AI optimizes the post for search engines. The human collaborator provides crucial input on internal linking strategies, meta descriptions, and title tags, as well as any additional SEO elements specific to the company's strategy, to maximize the post's search engine visibility.

By combining AI’s efficiency with human creativity and expertise like this, we can create superior content that is both engaging and optimized for search engines.

Further notes: Other tip I want to share is: You can even automate prompt-chaining by bringing prompt chain into a multi-agent workflow with human inputs between every step. Manually writing all these prompts again and again will takes a considerable amount of time, whereas in a workflow, you just need to provide inputs, like an AI app, to use these prompts without “even typing any of them again”. Not many multi-agent tools already include user inputs in between steps, you can refer to MindPal to find out more.

For any task, you can create a step-by-step “human-in-the-loop” prompt chain like this. Drop down any task, you want me to give examples of these prompt chains, I will send them to you. And also, tell me what do you think about this? Any enhancement or something?

4 comments