Parabolica #29 - How I automated market research and script writing with AI agents
Only $0.37 to save hours of work
Yo! Manson here.
Quick update: My buddy Mark gave me his buddy pass to join Season 4 of Buildspace. Excited to be on this journey! Hope to bring a cool idea into reality. I’ll keep y’all updated.
Okay let’s get to it.
You’ve probably heard of AutoGPT or gpt-engineer by now.
GPT-Researcher is an autonomous agent that is designed to research your topic by scraping web sources. This is helpful for conducting market research. It uses a ‘planner agent’ and an ‘execution agent’ to help.
The planner agent generates the research questions based on your prompt and the execution agent goes and scrapes websites for relevant data. Then it creates a report for you based on the summaries of all the sites the agents visited.
Here’s a diagram from their Github:
The common feedback after testing AutoGPT was that it came into issues too frequently. Perhaps the limited scope of autonomous agents like gpt-engineer and GPT-Researcher makes it more successful.
Why use this instead of a LLM on its own?
At the moment, ChatGPT Plus doesn’t have a browsing mode anymore.
Hallucinations are an issue with LLMs.
Time is of the essence.
You want sources to the links like how Bing Chat operates.
You can use app.tavily.com to use the default agents they’ve created. You will need an OpenAI API Key with GPT4 access.
Initial run
My question was: “What's the target market for a men's mental health app? What are good marketing strategies to promote an app for men's mental health?”
It chose the ‘Business Analyst Agent’ for me.
I ran it locally and a run costed $0.37.
If you’re curious what the default ‘Research Report’ output looks like take a look here.
My customizations for Marketing
The default agents aren’t well-suited for Marketing so I made some changes.
So I wondered how I could customize the code to make it fit my use-case.
I created a new agent named: “Customer Listener Agent”. Its role is to look for reviews, testimonials, conversations, and language online on the given niche.
"Customer Listener Agent": "You are an experienced marketer. Your purpose is to compile research on what people are saying about a product or niche and learn to leverage their language for direct response copy."
I then created a new report type “Customer Language Report” that compiles what people are saying on Reddit and forums about your question.
If you’re curious what the ‘Customer Language Report” output looks like take a look here. (‘Customer Language Report’ is not the best name but it’ll do for now.)
I also created a ‘Generate Scripts’ prompt that uses the research to generate 3 video scripts following a proven script framework.
Here’s what it generated. It’s a good starting point.
All of the URLs it visited and summaries are in an ‘outputs’ folder which is valuable in itself for your own research.
To summarize my edits:
Changed research agent prompt to only generate Google searches on Reddit and forums
Added ‘Customer Listener Agent’
Added ‘Customer Language Report’
Added ‘Generate Scripts Report’
Added drop-down options in the .html file for the reports
Next steps:
Push video scripts to ElevenLabs API for voiceovers
Code another text input to capture a description of your product (I’m ‘hardcoding’ it in the ‘generate_scripts_prompt’ right now.)
Add more script frameworks to be generated
Generate scripts from existing research summary
What would you add to make your life easier as a marketer? I’m not a software engineer but I’ll give it a shot.
Any feedback on the prompts? I’m keen to learn.
it didn't work