How does “PMM Search” work?

PMM Search is a tiny search engine focused on product marketing.

It only indexes blogs, newsletters, and social posts from growth and marketing operators in B2B. Because of that, questions like “What should a homepage contain?” or “What is an SQL?” will get proper answers related to marketing and sales.

To do that, I’ve used Google’s Programmable Search Engine, which lets you define a set of websites where the search should be performed. For Linkedin posts I’m indexing pages that have this common pattern*

After you ask it a question (or enter a search query) it retrieves the top most important pages by first asking GPT-4 what would be the ideal search keyword for your question, processes the results using Firefox “Reader mode” to extract the text, and then asks ChatGPT to summarize them.

The ChatGPT prompt looks something like this:

Based on the following sources what is the correct answer to "How..."

1. ...
2. ...
3. ...

The code for querying ChatGPT and processing the results is based on Mckay Wrigley’s open-source clone of Perplexity AI.

The sources used for PMM Search are currently these ones. Write to me to add new ones.