Jimmy Kim breaks down three things that make AI-generated marketing copy actually work: feed it rich brand context like customer profiles and past campaign data, write specific prompts that detail audience, offer, tone, and format, and have a skilled editor who knows when to revise versus when to manually fix things.
3 quick primary things. The first is like the quality of the actual brand context. So is it loaded in with kind of the project for the brand voice? Does it have the ICP? Does it have examples of past campaigns and past data? People don't feed it all of this necessary context. So without it, you're going to be at the 50%. But with some iteration, with some finesse, with some training, with the data dump, I do believe that 80% does become more realistic for kind of the ones that are using it well. And then the second one is the quality of the prompt. So you really need the prompt to be very specific. Specific. And when I talk about specific, it needs to be entailing the audience, the offer, the angle, the tone, the length, the structure. At the end of the day, vague prompts produce vague output. And then the last one, this third one would be like the really good operator editor at the end of understanding how to do the editing, knowing what to edit, what to leave, where to push the AI for revisions versus when to just do it yourself.
