Hey friend!

You might have seen me post about my year-long educational quasi-sabbatical as a Tower Fellow at UT Austin. Hook ‘em!

I've been going deep in storytelling. Learning how myths form and persist. Studying history and its making — and its revisions.

Contemplating why we believe what we believe, and how credibility is earned . . . and lost. And of course, how AI figures into all of this.

I think we’re in a delicate time. If you’re anything like me, you’ve probably started to question your sources. You’ve started a new habit of fact-checking. Maybe you’ve become skeptical, or even cynical.

For me, communication is underneath it all. And communication is getting complicated. Even manipulated.

Large language models are embedded into almost everything we’re consuming now, and things are starting to get interesting.

Which brings me to Idiograph.ai - the thing I’ve been laboring over for about a year and a half.

I’ve been building and delivering positioning and messaging playbooks to clients for years. Quick definition: a messaging playbook in its most basic form is a guide that governs what you say about your business, and just as importantly, what you don't.

Anyway. My clients have happily paid me to create these playbooks for them because knowing exactly what to say, it turns out, is hard. Why? Because competitors are trying to do the exact same thing with the exact same target audience, AND, more importantly, knowing what to say and actually saying it over, and over, and over again is super hard.

Operationalizing your company narrative, from top to bottom, is one of the most challenging things in business because it reaches into every function. It’s one thing to have a playbook. It’s a different thing to run the plays week in and week out.

But then ChatGPT happened. Suddenly it felt like the problem was solved. We started cranking out beautiful prose. It felt like a revelation. I thought my entire raison d’etre - my reason for being — was DOA. It was an existential moment.

But it didn’t take long until we noticed — hey, everyone else got better right alongside us! Boo! No one got an advantage.

And then the models started “improving”! But as they did, they did weird things to our writing. In trying to sound human, the models overcorrected in some areas. At first it was “In a world…” Then “It’s not X, it’s Y.” And now, “Here's the sharper version”… “to sit with”… “quietly”…“in the room.” Gag.

Everyone’s communication that comes out of an AI model kinda sounds the same stylistically … and it’s all platitudes. Big obvious generic crap.

And it turns out there’s a real reason for this baked into the way large language models are designed. You cannot prompt your way out of it. It would be like prompting a fish to start walking around on land and breathing air. Not possible.

AI models are trained to predict the most expected next word (actually, not even a whole word, a token, which is just a small handful of characters), and this prediction is the most important thing to understand about LLMs.

How do we predict? We look at a large sample of things, we spot the trend, and then we predict the most probable thing that would happen next.

But there’s a problem. When you give your AI a juicy story to “improve” — like a way that you are differentiated from your competitors, which is your whole value proposition!! — what it’s trained to do is look your stuff over and insert the next most probable thing. It’s not likely to protect the most unusual true detail that you gave it. So you end up just having to work it over again. You’re playing tug-of-war with your model.

Every time you run something through AI — draft it, tighten it, adapt it for another use — it pulls a little further from what actually happened and a little closer to what a normal version of that story sounds like. And get this. When a real, specific detail falls out, the AI doesn't leave a gap. It fills the space with something that sounds just as plausible. Which is why nobody catches it. And now you’ve got some back-stepping to do.

So back to Idiograph. I realized, after doing battle with AI too many times, that there’s an antidote to the crap AI writes and it’s been sitting under our noses the entire time. ⇒ It’s all the stuff that gets said less formally as companies actually walk the talk.

Here’s the insight: We need to pivot from storytelling to history-telling. In other words, we need to tell the true tales that no other company can tell except yours.

I built Idiograph, an AI agent, to use a messaging playbook as its ‘constitution’ and then find stories that pay off the brand promise with rich, specific, REAL detail. It doesn’t invent stories, it finds your real stories that are sticky. The ones people will remember and repeat.

Cool, huh?

I’ve implemented this as a pilot with my client, XPRIZE and we’re starting to scale it there. And now I’m ready to start rolling it out further — maybe at your company, or one you know of that needs to stop the crazy-making of bogus and RISKY AI-generated content.

Idiograph is a way to go back into your company's own raw material — calls, transcripts, notes, whatever's sitting there — and pull out the real, provable stories before they are forgotten or get smoothed away. It’s a way to find, catalog, and protect the stuff that's actually true and specific about you, before it disappears into the same generic version everyone else already sounds like.

I’ll be candid about why I'm telling you this. Part of it is business — I built this, I believe in it, it works, I'd like more people to see it and pay for it. But that's not the whole reason.

The bigger thing is the question I’ve been turning over this whole sabbatical: how do we actually put AI to good use, without losing the thing that made any of us worth listening to in the first place? How do we protect what’s real and true? How do we avoid accidentally blowing up our credibility?

I don't think these are small questions. We’re in the middle of a credibility war. I’m sure you’ve felt it. Credibility starts with what’s true and consistent. And you can win this war. All you have to do is start mining your own true, lived history.

So that's the update.

Take a look at Idiograph.ai if you're curious — and I hope you are. And if you want to keep following where this goes, please consider following me and subscribing to my Substack where I’m getting into the nitty-gritty.

One of my goals has been to understand how large language models really work, and how AI figures into what they’re doing to our communications, how they establish or erode credibility, and how we can use AI responsibly. Part of that means knowing and understanding all the AI-related vocabulary that has entered our lexicon. I’m sharing all that stuff on my Substack.

‍