AI SEO

In short, SEO for LLMs! Analytics for marketers to assess the visibility of their products in AI-generated content.

Comment

Inspiration

As a business owner myself attempting to improve my brand's SEO, I started to research how prominent my company was in GPT-4/3.5 output (as well as other models). For every digital marketer in the world, a tool like this will be must-have as consumers migrate from search engines to chat bots for their content.

What it does

Given a company name, description, customer segment, and competitors, the tool provides analytics on how prominent the given company is in GPT-4/3.5 output compared to its competitors.

How we built it

Two primary steps:

  1. Given a company description and customer segment, use GPT-4 to generate a robust, diverse list of realistic prompts a consumer may have for an AI chatbot that may cause the company's name (or one of its competitors) to appear.
  2. Simulate each of these generated prompts and analyze how prominent the given company name appears compared to its competitors. Output summary statistics to highlight the comparison, as well as the list of the prompts and prompt outputs used for the analysis. Using the Together AI API, insights can be seamlessly compared across output from different models.

Challenges we ran into

The biggest challenge is generating a realistic and diverse set of prompts given a customer segment and product description. Without careful prompt engineering, all of the generated prompts would be some version of "Provide me a list of companies that do ...", which may not accurately reflect the real prompts that consumers would input.

Accomplishments that we're proud of

The tool is accurate and flexible across many industries and customer segments - its current version is already capable of providing value to thousands, if not millions, of marketing specialists around the world.

What we learned

In applications like these where you have a chain of LLM outputs feeding into LLM inputs, it is important create prompts that ensure the output in each link of the chain is formatted in an easily parsable manner (e.g. as a JSON).

What's next for AI SEO

The long-term vision is to launch a startup venture that can provide specific, accurate recommendations for marketers to take to improve their brand's visibility on AI chatbots (similar to SEO tools today). As the data collection and training of commonly used LLMs (e.g. GPT-4) is not transparent, this will require long-term research and iteration as the AI industry develops.

Built With

Share this project:

Updates