Patrick Kemp

Hi, I'm Patrick. I build agentic marketing systems. This is where I post things I've built, written, and learned.

My goal is for this to be a resource for others building AI marketing infrastructure. I've found it hard to find resources like this. Connect with me on LinkedIn if you have thoughts or feedback on anything you see here.

I do fractional and advisory work. Mainly with early-stage B2B tech startups. I'm vague on purpose sometimes to protect confidentiality. If you're trying to grow your company, I'd love to chat about working together.

Projects

Things I've built.

See all projects →
GEO Measurement System

Agentic workflow to measure and improve how a company is represented by LLMs.

PinnedBuiltJun 2026

An agentic system that measures how LLMs portray a company in buyer prompts, audits on- and off-site dimensions that influence LLMs, and suggests high-leverage fixes. It measures LLM responses using performance scores and audits dimensions using deterministic scoring rubrics. Every score is reproducible run-to-run. There are two human-approval gates. The final artifact is a summary of results and recommendations in HTML.

GitHub ↗ Example report ↗ Methodology ↗
Brand Voice Generators

A skill to build a brand voice reference file for a company’s agents

Jul 2026

Agents need a brand voice reference file to write content reliably. The shape of the file is converging to a canonical structure (see Anthropic’s plugin). This skill allows a marketer to build a brand voice reference file in a ~45m working session with an AI that does research, proposes examples, and locks the canonical file other agents use.

GitHub ↗
Claude Code Slack Wrapper

Slack agents connected to local claude code sessions.

Jul 2026

This is how I quickly prototype Slack agents with client teams. A claude code session configured as a marketing agent (system prompt, MCPs, skills, commands) connected to Slack for a team to use that requires my approval for write actions. I’ll write a post about how I use this and how its different from Claude Tag soon. This is a generic wrapper you can clone + build off.

GitHub ↗
Content

My writing about agentic marketing.

See all content →

How to measure what AI actually says about your company (with a codebase)

Why I use a deterministic performance score instead of mention rate to measure AI discovery.

Jul 2026 · 6m read

How to Build a Context System for Marketing Agents

Marketing agents fail on context, not prompts. Here’s how I build and deploy context systems for my clients’ marketing agents.

Jun 2026 · 4m read
Free GEO Audit

Audit how AI sees your company right now for free

Run audit →