How GPT-4o decides which brands to mention (and how to be one of them)

I ran 500 brand queries through GPT-4o last month. Same questions, different brands, across 12 industries. The patterns were consistent enough to build a playbook around.

Here's what GPT-4o actually does when someone asks "what's the best [your category]?"

When a user asks GPT-4o about brands in a category, the model doesn't search the web (unless browsing is enabled). It generates text token by token, drawing on patterns learned during training.

Step 1: Query parsing. GPT-4o breaks the query into semantic components. "Best project management tool for remote teams" becomes: category (project management), qualifier (best), context (remote teams), intent (recommendation).

Step 2: Pattern matching. The model activates neurons associated with the category and qualifier. It's not looking up a database. It's generating the most probable sequence of tokens given the context.

Step 3: Brand selection. This is where it gets interesting. GPT-4o selects brands based on four weighted factors:

| Factor | Weight | What it means | |--------|--------|---------