Coralville

Aric Pearson

Human Led Optimizer

A recent customer journey optimization project clarified something I keep returning to that I think is getting lost in the current AI conversation: for strategy work, the first draft needs to come from the human. This can feel counterintuitive when the pressure to 10x efficiency is constant, but I’ve found it matters for two reasons. First, giving AI a human-generated draft provides context that significantly improves the final output. The second reason is more personal, and I’ll come back to it.

Building the Human-Led Foundation

The project began with a familiar problem: our conversion rate was slipping. Google Analytics confirmed friction in the user journey, and I reviewed flow data to identify the weak points, then manually audited the site to develop hypotheses about where trust was breaking down and what content gaps were affecting progression. The pattern that emerged was clear: customers were heading straight to site search instead of engaging with our homepage. I’d wanted to redesign our homepage for years, but it had always felt daunting and easy to deprioritize in favor of more urgent work.

I started the project with a blank Word document instead of a chat window. It was strategic work grounded in context: understanding the audience, understanding the business, and understanding what the user needed at each stage. I drafted a proposed journey and mapped the available assets, page components, and content blocks that could support users along the way. That rough document became the core brief for everything that followed.

Using ChatGPT, I worked with multiple custom GPTs across a single thread to refine my brief. The ability to call multiple customGPTs in a single thread is still a differentiating feature for OpenAI, though Claude’s skills offer something comparable. Each GPT had a distinct role. Some represented key customer personas we’d developed over several years, pressure-testing the journey from different user perspectives. Another focused on conversion optimization and content strategy, to help keep CRO and UX best practices in mind. Together, they challenged my assumptions, helped tighten the structure, and surfaced sections I wouldn’t have otherwise considered.

Collaboration to Automation: Scaling Execution with AI

Once that planning document was solid, I shifted from collaboration to automation. First, I moved the finalized plan document into a local folder and used Claude Code to build the page elements in HTML. I gave the model access to the plan and additional context, including access to our existing CSS, so it could work within our established design system. Rather than generating full pages at once, I had it build the components one by one (hero areas, trust sections, supporting content blocks) as individual text files. This made it easier to review and adjust individual pieces after assembly.

By the end of a single Claude Code session, I had a full set of usable page elements as text files, organized by page in separate folders. In about an hour, I assembled those into a working prototype in our staging environment. My own HTML knowledge handled the stitching and final cleanup. After reviewing the rendered code, I requested several small changes to individual sections.

That prototype changed the internal conversation. Instead of asking leadership to react to a concept document, I could walk them through a navigable experience: above-the-fold, mobile presentation, content sequence in context. The project moved from idea to demonstration in days.

What would previously have been a quarter-long initiative became a multi-day workflow thanks to the right amount of context and the right sequence.

In this project, AI did substantial work, but I was guiding it at every stage with my perspective, priorities, and taste. The sequence was: define the problem and draft the strategy independently; use AI as a collaborative pressure-test; then shift to autonomous tooling for production. Human thinking at the front of the process provided critical context and direction, giving the AI something worth building on.

Starting with a blank document has turned out to be as much about protecting my edge as it is about producing better work. My personal experience aligns with what researchers at MIT’s Media Lab and Microsoft Research are beginning to document about cognitive overreliance on AI tools. This became real to me last year: on a flight without Wi-Fi, assembling arguments into paragraphs for a report felt unexpectedly difficult. I kept reaching for an AI tool that wasn’t there. That discomfort was informative. It’s part of why I’m deliberate now about when AI enters my process.