We Need to Talk About How You’re Using ChatGPT | To the Power of X, Ep. 3

Welcome back to To the Power of X, Xponent21’s podcast on AI and marketing. In Episode 3, Chuck McCarthy, Kayleigh Crandell, and Jenna Pace ask where AI stops speeding up your thinking and starts quietly replacing it. The crew names the habits they’re seeing everywhere, from marketers to clients to that one Instagram caption Jenna refused to read. They explain what a language model is actually doing when it hands you a confident answer, including the shadow prompting that rewrites your question before it ever gets answered.

In This Episode

• Chat models predict text. They don’t strategize, and per the crew, treating fluent output as a strategy is the fastest way to execute a bad idea with total confidence.
• Hallucinations aren’t spooky. Chuck explains why a system built to sound like an expert will happily invent the statistic an expert would cite, and tells on the fake stat that almost got him.
• Sameness is now a double penalty. Kayleigh and Jenna break down why readers scroll past homogeneous AI content, and Chuck describes watching AI-heavy sites spike in visibility and then fall off a cliff.

Where AI Stops Helping and Starts Thinking for You

An AI output is only as good as the brain attached to it. Jenna puts it plainly. “The thing about AI and ones like Claude and ChatGPT is that it gives everyone just enough knowledge to be a little bit dangerous” (3:47). Her point is that these models predict text. They organize, they accelerate, they make a genuinely great sounding board. What they don’t do is reason about your business, which becomes a problem the moment someone hands one a half-formed idea and receives a standing ovation in return.

Chuck calls the models the ultimate yes-man, and Kayleigh lands the comparison of the episode. “Facebook’s never going to tell you, ‘Don’t post that ugly photo of yourself.’ … Chat says, ‘Yes, execute on that terrible idea'” (7:22). The platforms want you logged in and feeling affirmed, and an AI chat window is still a platform. The model sounds just as certain when it’s inventing the evidence. Chuck shares about a report that paired one real survey statistic with one completely invented one, both delivered in the same confident voice.

The shadow prompting explanation is the episode’s most useful mechanical detail: the fan-out of expert-level questions a model runs behind your original prompt before it generates a word. That same machinery is why a model asked for “a marketing strategy” may be learning strategy from whatever listicle slop is floating around the internet.

Jenna and Kayleigh take it from the reader’s side. People are getting sharp at spotting non-human writing, and homogeneous content gives audiences nothing to connect with. Chuck adds a warning from his read of the visibility data. Google, he says, is getting more aggressive about filtering what it identifies as pure AI-generated content. The episode closes exactly where it should, with the crew comparing AI tokens to arcade tokens and making a case for Skee-Ball.

Key Moments

• 3:47 “Just enough knowledge to be a little bit dangerous.” Jenna names the whole episode in one line.
• 5:51 The ultimate yes-man. Why “great idea!” is the most dangerous thing an AI can tell you.
• 7:35 Hallucinations, de-spookified. What’s actually happening when AI invents a statistic.
• 13:10 Shadow prompting. What happens to your question in the split second before the answer.
• 17:12 “Human brains are really, really good at identifying other human brains and really, really good at identifying non-human brains.” Jenna on why readers can always tell.
• 26:21 Grown-up tokens. AI companies are burning through compute tokens, and Chuck can’t hear the word without picturing a cup full of arcade tokens and zero Skee-Ball. 

Full chapter markers are in the video description on YouTube.

Go Deeper

• Don’t Outsource Thinking: AI in the Workplace
• The Truth About AI Content and SEO: Why Strategy and Expertise Matter More Than Ever
• Previous episode: The Data Nobody Has | To the Power of X, Ep. 2

Watch and Subscribe

Watch the full episode on YouTube and subscribe so the next one finds you. You can also listen on Spotify and Apple Podcasts.

If this episode described someone at your company a little too accurately, that’s fixable. A marketing team that knows when to use AI and when to use a brain is the entire pitch. Book a discovery call and we’ll talk about which is which for your business.

To the Power of X is Xponent21’s podcast on AI and marketing, recorded in Richmond, Virginia. New episodes and their companion articles publish at xponent21.com/insights.