See All the Custom GPTs I'm Working On #BehindTheBot

By Dan Sanchez · Episode published

Build small custom GPTs around real marketing tasks, test them on repeated work, and improve the instructions through play, failure, and revision before sharing the assistant with a team.

Get the next practical AI marketing episode wherever you listen.

Apple PodcastsSpotifyYouTube

Build small assistants around real work

When I recorded this behind-the-scenes episode, I was experimenting with a collection of custom GPTs rather than waiting for one perfect assistant. I had simple tools for writing, newsletters, SEO planning, naming, and running my show. Some were public, some were private, and some were still works in progress. The point was to learn by building around tasks I actually repeated.

A custom GPT is most useful when it has a narrow job and enough instructions to perform that job consistently. A newsletter builder should know the audience, format, and standards. An SEO article planner should ask for the inputs that affect the recommendation. A showrunner should help with the recurring production steps rather than attempt to be an all-purpose expert.

Treat the first version as a prototype

I was paying for ChatGPT because the time savings already justified the subscription. But the value did not come from turning on a feature. It came from testing an assistant, noticing where it made weak assumptions, and improving the instructions. The first version of a GPT is rarely the last version that deserves to be used.

Start with the workflow. Write down what happens before, during, and after the task. Add examples of the output you want and explain what the assistant should do when information is missing. Then run several real requests through it. If the output is inconsistent, the instructions or the use case probably need to be clearer.

Learn through play and failure

These small experiments also reveal where AI is genuinely helpful. A tool that saves minutes on a task I do once a year is less important than one that removes friction every week. Keep the assistants that improve the work and retire the ones that merely look interesting.

There is no need to protect an early prototype from criticism. Play, fail, revise, and keep the useful parts. Building these tools gives a marketer a practical understanding of prompting, memory, context, and workflow design. That knowledge compounds as the assistants get better.

The best next step is to choose one repeated marketing task and build the smallest assistant that can help. Use it yourself before sharing it with a team. Once it works, document the process so another person can understand when to use it and what to check.