Connect through text

Most real estate agents use ChatGPT the same way. They open a new chat, paste in the details on a house, and ask for a listing description. Or they keep one long chat going for months and wonder why it stopped sounding like them.

Here's what's happening. As a chat gets longer, the AI stops going back to the top to reread the context you gave it. It works mostly from the last few exchanges. So the voice examples you loaded at the beginning get buried, like one rolled-up piece of paper on a desk covered in coffee cups and clutter. The AI grabs whatever's closest, fills the gaps with generic realtor content from across the internet, and your voice starts to erode.

The fix is already built into both Claude and ChatGPT. It's called Projects, and most agents aren't using it. In this episode I walk through how I organize mine: one project per client, not per property, so a client's sale and their next purchase live in the same folder. Every new chat inside that project reads the instructions and files again, and that's what keeps the output consistent and sounding like you.

And once your work is organized this way, a lot of the paid "AI employee" tools start to look unnecessary. They're folders with instructions. So are Projects, and they come with the plan you already pay for.

In this episode:

• Why long chats lose your context and drift toward generic output
• A simple way to picture your AI workspace: the cluttered desk and the organized desk
• Why I set up projects by client instead of by property
• What goes in a project: instructions, your Voice Foundation, and client files like tax records, listing agreements, walkthrough notes, and offers
• The difference between AI's built-in memory and intentional operating context
• Scheduling a weekly client update email from inside a project

Next up: a full walkthrough building a project around a mock listing consultation.

Get the Voice Foundation, as a free worksheet or the full Claude plugin, and learn more at agentslearnai.com