Use Deep Research to Build a Knowledge Base for Your AI Sales Agent

Most people building AI agents give them a two-sentence system prompt and call it done. Ben from Founders Table shows a better approach: use Gemini's Deep Research to generate a full knowledge base that acts as the agent's brain. The result is an agent grounded in real sales methodology rather than generic LLM output.

The Problem With Simple Prompts

When you tell an LLM "you are a salesperson," it draws on its general training data to generate responses. That training data is broad and unfocused. The agent has no specific framework for how to handle inbound leads, when to push, when to back off, or how to structure a value proposition. You get generic responses that sound like a chatbot, not a trained sales professional.

The Deep Research Method

Ben walks through the setup of a sales AI agent named Jordan inside the Founders Table AI organization chart. The chart itself is worth noting because it shows a structured system of agents across Sales, Marketing, Community, and Events, each with defined roles, sub-agents, and shared information layers.

Instead of writing Jordan's prompt by hand, Ben opens Gemini's Deep Research and asks it to create a knowledge base for an inbound sales AI agent. The specific prompt: "Can you create a knowledge base that will act as a brain for our sales AI agent? We want it to be a non-pushy professional salesperson."

Deep Research then does what would take a human hours. It crawls hundreds of sources across the internet, pulling best practices for inbound sales, written sales communication, and professional tone. The research plan shown on screen includes specific tasks like finding articles on sales communication techniques, identifying best practices for professional tone in written sales, and exploring characteristics of high-performing sales professionals.

What the Knowledge Base Covers

The output is a detailed document titled "AI Sales Agent Knowledge Base" that covers:

Sales psychology and how to read buyer intent from written messages
Core principles of optimal written sales, including clarity, brevity, and directness
Value proposition crafting, with a focus on benefits over features
Empathy and understanding in sales responses, including techniques for acknowledging prospect feelings without sounding scripted
Personalization strategies that make the AI feel attentive to individual needs
Authenticity and friendliness, with guidance on avoiding language that comes across as cold or transactional

The document goes deep. It references Natural Language Processing techniques for detecting emotional states, discusses how to project warmth in text-based communication, and provides frameworks for building trust over long-term interactions.

Why This Works Better Than Prompt Engineering

The key insight is that the knowledge base becomes a retrieval source, not just a system prompt. When Jordan receives an inbound sales email, it pulls from this research-backed document to craft its response. That means the agent's behavior is shaped by synthesized best practices from real sales experts, not by whatever the base model happens to generate.

This is a practical example of a pattern that applies beyond sales. Any agent that needs domain expertise benefits from a pre-built knowledge base generated through systematic research rather than a few lines of instruction.

Key Takeaways

Two-sentence prompts produce two-sentence quality. If you want your agent to perform like a professional, give it professional-grade source material.
Gemini Deep Research can synthesize hundreds of sources into a structured knowledge base in minutes, covering ground that would take a human researcher hours.
Use the knowledge base as a retrieval layer, not just a system prompt. The agent should reference it when generating responses to real inputs.
Define the agent's personality in the research prompt. Ben specified "non-pushy professional salesperson" upfront, and that constraint shaped the entire output.
This pattern works for any domain. Customer support, technical writing, recruiting, or any role where the agent needs specialized expertise.

Resources

Gemini Deep Research - Google's AI research tool used to generate the knowledge base
Founders Table - Ben's AI agent content on TikTok
Google AI Studio - Build and test AI agents with custom knowledge bases