Directing AI

This post was originally featured in the March 25th, 2026 newsletter found here: INBOX INSIGHTS: Directing AI, AI Digital Clone Part 1

Moving from Executing to Directing AI

If you know me, you know I have opinions about AI replacing people.

Strong ones. Loud ones. The kind that earns me the “ranty pants” label on LinkedIn.

So when I tell you I just rebuilt katierobbert.com using Claude’s Cowork mode — and I mean really rebuilt it, the whole site, over the course of two days — I need you to hear what I’m actually saying. Because it’s not “AI built my website.”

It’s “AI operationalized twenty years of my expertise into a website. And it didn’t write a single word I wouldn’t have written myself.”

Here’s the thing.

Most people are using AI backward. They open a tool, type a prompt, get generic output, and then spend more time fixing it than if they had just written the thing themselves. Sound familiar?

That happens because they’re starting at Platform—the fourth P in the 5P Framework by Trust Insights™—and skipping Purpose, People, and Process entirely.

I didn’t do that.

We started with Purpose. (Obviously)

Before we touched a single tool or wrote a word of copy, I had to answer the question I ask every client: what is this site actually for? Because katierobbert.com had been sitting there for a while without a clear reason to exist. Was it a portfolio? A blog? A consulting page? A vanity URL? If I couldn’t articulate the purpose, no amount of AI was going to fix that.

So we worked through it. The site exists to establish my authority as a speaker, strategist, and CEO — separate from Trust Insights — and to give people a place to book me, hire my company, or read my thinking. That clarity shaped every page we built. Without it, we would have just been making pretty pages that didn’t go anywhere.

Then we documented my brain.

Before I wrote any website copy, I worked with Claude to build the system. Not the website. The system. I fed it my existing materials — bios, podcast transcripts, blog posts, LinkedIn content, framework documentation — and together we built a comprehensive writing style guide that captures how I actually think and communicate. My signature phrases (“here’s the thing”), my anti-patterns (if I ever say “in today’s rapidly evolving landscape,” check me into a hospital), my argument structure, how I use vulnerability, how I close a post. The AI analyzed my patterns. I validated them. Over 500 lines of documented voice patterns — built collaboratively, not from scratch.

Then we built a brand system. Not “pick some nice colors.” A real brand guide — my color palette (forest greens, warm ambers, dark mode), typography rules (Rubik across the board, weight variations for hierarchy), visual tone rules, logo usage guidelines, the whole thing codified into a format AI can reference every time it works on my stuff.

Then a WordPress theme skill. Block patterns, CSS class conventions, page structure rules — so that any content generated would actually work inside my site’s existing architecture without a developer having to clean it up.

And then I packaged all three of those into a reusable plugin. So every time I open a new session with Claude on this project, it’s not starting from scratch. It’s starting from me. My brain. My brand. My technical setup. Every time.

Here’s what we actually built across two days — no code, no developers:

And when I say “we,” I want to be specific about what that means. Claude Cowork is an autonomous AI agent. I wasn’t typing code. I wasn’t manually editing CSS. I wasn’t handwriting meta descriptions. I was directing — giving strategic input, making decisions, approving or correcting output, and pointing Claude at my existing materials. The AI did the execution. I did the thinking.

That’s a meaningful distinction. Because the work that happened below? I didn’t do most of it with my hands. I did it with my brain.

Session one was about building the foundation and the first pages:

• Created a comprehensive writing style guide documenting my voice across seven different modes (ranty pants, teaching, vulnerable, CEO, collaborative, podcast, thought leadership) — including good vs. bad examples so the AI knows exactly what “sounds like Katie” means
• Built a complete brand system skill with color palettes, typography hierarchy, spacing rules, and visual identity guidelines
• Built a WordPress theme skill documenting block markup patterns, CSS custom properties, and page structure conventions
• Packaged it all into a reusable Cowork plugin that persists across sessions
• Completely rebuilt the About page from scratch — pulling from my existing bios, speaking history, credentials, and frameworks (such as 5P, TRIPS, 7D, 9C) to write a full page in my documented voice
• Built the Speaking page with a speaker bio, topic areas, and past engagements
• Built the Hire Trust Insights page, positioning our consulting and workshop services
• Built the Contact page
• Fixed a duplicate title CSS bug that was affecting every page on the site
• Set up Rank Math SEO on the Homepage and About page — custom meta titles, descriptions, and focus keywords
• Pulled in existing Trust Insights blog content and resources to populate the blog — so the site launched with real content, not placeholder posts, drawing from work I had already published

Session two was about completing SEO across the remaining pages and polishing:

• Worked through the strategic purpose of the site itself — why does katierobbert.com need to exist separately from trustinsights.ai? Who is the audience? What action should someone take when they land here? That clarity drove the SEO and content decisions for every page
• Set up Rank Math SEO on the Speaking page (score went from 14 to 16 — that one needs more content to score higher, but the metadata is solid)
• Set up Rank Math SEO on the Hire Trust Insights page (score went from 15 to 56)
• Set up Rank Math SEO on the Contact page (score went from 8 to 47)
• Set up Rank Math SEO on the Blog page (score went from 6 to 71)
• Homepage went from 21 to 82; About page went from unscored to 80
• Cleared the WP Engine cache and verified the live site

Total pages built or rebuilt: 6. Total developers hired: zero. Total hours spent pretending to be someone I’m not: also zero.

My role in all this was strategic direction. I decided what the site needed to say. I pointed Claude at the right source materials. I corrected things that didn’t sound right. I made the judgment calls — which frameworks to highlight, what the CTAs should be, and how to position the brand. Claude handled the WordPress editing, the CSS fixes, the SEO configuration, and the content structuring. It navigated the admin panel, wrote the markup, and configured the plugins.

That’s what autonomous AI looks like when you use it well. You’re not a typist. You’re a director.

And the reusable plugin means the next time I need to update the site, write a new page, or create content, I don’t have to re-explain who I am or how my site works. It’s already there. That’s the Process piece of the 5P that most people skip.

So why did this work when most AI-generated content sounds like garbage?

Because the AI wasn’t generating my expertise. It was executing it.

Every word on that site traces back to something I created — a bio I wrote, a talk I gave, a framework I developed, a phrase I’ve been saying for years. The AI’s job was to organize it, structure it for the web, match my voice patterns, and apply it consistently across six pages. That’s execution. That’s what AI is good at.

What AI can’t do is have my twenty years in pharma, operations, management, and marketing technology. It can’t decide that people come before platforms. It can’t develop the 5P Framework. It can’t know that I hate the word “utilize” or that I’d rather say “dumb stuff that works” than “strategic optimization.”

That’s mine. And because I documented it, the AI could use it. Because I didn’t document it for an AI — I documented it because I’m a nerd who believes documentation is love. (That’s for nerds. I’m a nerd.)

Here’s where you come in.

If you’re sitting there thinking, “I should document my own voice and frameworks before I let AI loose on my content” — yes. You should. Here’s where to start:

1. Write down your top 5 signature phrases. The things you say so often your team finishes your sentences
2. Write down your anti-patterns. The words and approaches that make you cringe when you see them in your industry
3. Document a framework that you use often. Transfer your thoughts onto a page without worrying about perfection
4. Write a short “what I am NOT” list. Sometimes defining what you’re not is more useful than defining what you are
5. Use all that as your starting brief the next time you work with any AI tool

The difference will stun you. Not because the AI got smarter. Because you gave it something worth executing.

I’m not saying AI replaces your expertise. I’m saying it operationalizes it. There’s a massive difference. And if you don’t see that difference yet — if your AI-generated content still sounds like everyone else’s AI-generated content — that’s not a tool problem. That’s a you-haven’t-done-the-work problem.

The good news? The work isn’t hard. It takes intention. And maybe a little bit of that documentation nerdiness I keep going on about.

How are you letting AI work for you? Reply to this email or join the conversation in our Free Slack community, Analytics for Marketers!

– Katie Robbert, CEO