When you ask ChatGPT "what's a good project management tool?", it doesn't randomly pick Asana or Linear. There's a pipeline behind every brand mention, and understanding it is the first step toward what the industry now calls GEO (Generative Engine Optimization).
I'm Jakub, builder at Inithouse. We run 14 products across different verticals, and one of them, Be Recommended, was born from trying to reverse-engineer exactly this: how do LLMs decide which brands to cite?
Here's what we learned, technically.
The RAG Pipeline: Where Brand Mentions Actually Come From
Most production LLM systems (Perplexity, ChatGPT with browsing, Gemini with grounding) don't rely purely on parametric knowledge. They use Retrieval-Augmented Generation, a two-stage architecture:
