How To Make AI Remember, Specialize, and Show Up Ready To Work

Radio people understand this instinctively: tools don’t matter until they make the work better. Faster. Sharper. Consistent, Reliable. I suspect that's not how you'd describe your experience with AI.

If 2025 was the year you started talking to AI, asking questions, kicking ideas around, seeing what it could do, think of that as step one. Step two, the one that matters in 2026, is learning how to make AI remember what matters to your work, create exactly what you need on the first try, and show up ready to work instead of starting from scratch every time.

 

And look, if you spent last year watching others experiment with AI from a safe distance, that’s fine. The noise was loud. The hype was relentless. AI didn't pass you by. You didn't miss the secret handshake. Starting to use AI right now isn't about catching up; it's about doing it. Jump in.

 

Radio needs this to be the year AI stops being a curiosity and starts being truly useful. Not flashy, trendy, or futuristic. Useful. That means using it intentionally. Giving it structure, instructions, and feedback so it starts pulling its weight. That's where Projects, a feature in ChatGPT and Claude, come in.

Quick note: Projects can be created in ChatGPT and Claude using a free account. Both are worth investigating. Unless noted, references to ChatGPT below will work in Claude.

 

Think of a Project as your own private workspace inside ChatGPT, dedicated to a specific recurring task or job. You build a specialized version of ChatGPT that delivers a particular output based on the instructions and resources you provide (rather than what it was trained on or what it guesses).

 

Let's use writing imaging copy as an example. A killer image sweeper that nails the essence of why your station is special is the backbone of defining your brand and making it memorable. That's why they're so hard to write.

 

Click 'New Project' in ChatGPT and name it 'Imaging Writing'. Before you click the 'Create project' button, click the gear icon to open the Project's memory settings and select 'Project-only.' I'll explain why this is important in a bit. 

Now, create a knowledge document by copying and pasting the best imaging copy you've ever written into a document. Name it Imaging Examples, and upload it to the Project. Next, upload a document with writing instructions: 2 lines maximum; always put the station name at the end; write two versions of each; emulate the tone, style, and emotion of the writing in the Imaging Examples document.

 

Then test it. Enter a prompt, something like this, in the Project's chat: 'Write 10 image sweepers to play during our ten-in-a-row ride home.' Copy the entire response and paste it back into the chat, with notes on what's good and not so good about each one.

 

It's important to explain in detail why the good ones are good and the bad ones don't work, because a Project is a living workspace. It responds and generates content based on your instructions, the knowledge files you provide, and the entire chat history in that Project. That's why we changed the memory setting to Project-only.

 

The more precise you are with feedback, the better the results will become. Eventually, the need for feedback and direction disappears because you've trained the Project to deliver what you need, exactly the way you want it, on the first try.

 

Individual Projects are great for writing imaging, promos, blogs, and social posts. With the right instructions and knowledge docs, you can create a single Project that generates all of them with a single prompt.

 

Want a quick start? I've created a new batch of various Project instructions you can get for free, here. You'll also get specialized prompts that create custom station brand identity and style guides, target listener personas, and an audience behavior compass that you can use as knowledge docs in any Project.

 

Projects can change the way you think about what AI can do for your station, your show, your creativity, and your time. The only limit on how much they can help is not making one. So, go and do. Expect it to be a bit clunky at first, like learning a new audio board. Stick with it, and you'll shift into high gear quickly and never look back.   

Next
Next

Start treating AI as a collaborator, not a vending machine.