I've seen a lot of people talk about grounding and RAG in AI and then follow it up with a load of technobabble that they clearly don't understand themselves. In simple terms, grounding converts a prompt into one or more traditional search queries then scrapes the results in Bing/Google. So for example, the following prompt:
"I'm running a 10k at the weekend and still want to look fabulous. Can you recommend a foundation which won't run?"
Becomes "best long lasting foundation waterproof running 2025" or "foundation won't run for running makeup".
Understanding how grounding queries are created can improve your understanding of how LLMs generate their answers as well as informing what you need to do to get yourself seen. But how do you find out what the grounding queries are? In ChatGPT, follow these steps:
Step 1: Open Chrome Dev Tools (F12)
Step 2: With Chrome Dev Tools open, send your prompt (make sure it requires a search and doesn't just use pre-trained data)
Step 3: Under Network, click the Fetch/XHR button and search for a script called "conversation". Click on it to open and select the Response tab.
Step 4: The easiest way to do this is to copy the JSON you can see into a text editor. You can then search the text for instances of "q" (with the speech marks). This will show you the grounding queries.
I'm working on a solution to make these steps a lot easier/quicker so watch this space.
This post was originally shared by on Linkedin.