GenAI Prompt Pilot, Teaching AI to Think Creatively

This program started as an open question more than a project. The brand wanted to explore whether GenAI tools could produce genuinely on-brand creative content, but nobody had defined what "good" looked like for a machine generated asset. On the ground, the writers on my team were unsure whether they were being replaced by the tool or equipped by it, and that uncertainty was quietly slowing everything down.

I treated prompt writing the same way I would treat briefing a junior creative team, with structure, context, and clear tone guardrails. I built more than 80 prompts mapped to specific product contexts and created a scoring system so we could actually measure whether an output was on brand, instead of relying on gut feel alone. I mentored five junior writers through the shift so they came to see the tool as an extension of their craft rather than a threat to it.

The real test came when engineering pushed back on how much iteration the creative side kept asking for. I sat between both teams, translating what "close enough" meant technically versus what it meant creatively, until we landed on a shared definition of quality that neither side had to compromise to reach.

Model output accuracy improved by 30 percent, measured against our QA scoring rubric, by pairing structured prompt engineering with a formal creative review loop. The pilot earned a greenlight for a Phase II expansion, and the framework became the reference model for future creative and AI collaborations across the team.

Training an AI was not about creativity alone. It was about clarity, structured thinking, and strategic iteration, just like leading human teams.