Channel RVA

RVA's independent media hub — news, video, podcasts, and community content for Richmond, Virginia.

The Data Nobody Has: Inside CARL Labs and the Future of AI Marketing | To the Power of X, Ep. 2

By Chuck McCarthy · June 30, 2026 · 26:48 read

Category: News & Politics

Show: To The Power of X

Most AI visibility tools are giving you a dashboard. What they're not giving you is data you can actually make decisions from. That's the problem Chuck, Kayleigh, and Will sit down to unpack in this e

Most AI visibility tools are giving you a dashboard. What they're not giving you is data you can actually make decisions from.

That's the problem Chuck, Kayleigh, and Will sit down to unpack in this episode of To the Power of X, and it's a problem that's been sitting in the background of almost every client conversation Xponent21 has had over the last year.

When a client asks "what is AI saying about my business?" there isn't a clean answer. Not because the question is wrong, but because the tools built to answer it aren't built on real research infrastructure. They're built on shallow signal, marked up to enterprise prices, and handed to marketers who are expected to tell a story with it.

So Xponent21 built something else. CARL — the Cognitive AI Ranking Lab — started as internal software to help the team move faster and with more precision. It's since grown into something bigger: a research lab that prompts large language models across hundreds of topics and industries, captures those responses as data points, and analyzes how AI-assembled answers change over time as the models evolve.

In this episode, Will Melton breaks down what CARL actually is, why we built it, and what the early research is starting to show about how LLMs construct answers — including things like shadow prompts, compute thresholds, and the personalization layer that makes "what is AI saying about me" a genuinely hard question to answer at scale.

Kayleigh brings the client-side perspective: what it's actually like to navigate those conversations without good data, and why the variation in LLM outputs is only going to get more pronounced as the models get more user-aware.

If you work in marketing, run an agency, or you're trying to figure out what your AI visibility spend is actually buying you, this one's worth your time.

---

CARL Labs: https://carlaio.com/

Xponent21: https://xponent21.com/

---

CHAPTERS: