Every piece of advice about AI tells you to be more specific. Here is the one place that advice is exactly backwards.
The Takeaway
When AI keeps getting something wrong, the instinct is to describe it more precisely. Sometimes the precision is the problem. The instruction gets so tight there is no room left to satisfy it, and you end up with a worse result than if you had said less.
What Happened
I build presentation decks with AI. An agent puts in the details, and out comes a finished, branded deck with their photo on it.
For a stretch, one thing kept going wrong. A card would come back with an empty gray box where a photo should be. The space was there, reserved and sized and waiting. Nothing in it.
So I did the obvious thing. I got more specific.
I described where the image should sit. Still empty. I described how large it should be. Still empty. I gave dimensions. I described the position relative to the text. I got more exact with every pass, because that is what you do when something is not landing. You clarify.
It got worse.
At some point I tried the opposite, mostly out of frustration. Instead of describing the placement precisely, I described almost nothing. Just what the image was and roughly where it belonged. A photo of the agent, near the top.
It rendered perfectly.
I tested it enough times to be sure it was not luck. It held every time. Loose description works. Tight description fails. And the tighter I got, the more reliably it failed.
The precision was causing the empty box.
Why It Happened
Here is what I eventually understood, and it changed how I write everything I give AI.
When you specify something exactly, you are not just describing what you want. You are defining a set of conditions that have to be met simultaneously. Put it here, at this size, at these proportions, in relation to that. Every added detail is another constraint.
At some point the constraints stop describing a thing that can exist. Not because any one of them is wrong, but because together they leave nothing that satisfies all of them at once.
And when that happens, you do not get an error. You do not get “those requirements conflict.” You get the system doing the best it can with an impossible instruction, which in my case meant reserving the space I demanded and then being unable to put anything in it that met every condition I had listed.
The empty box was not a failure to understand me. It was the exact consequence of understanding me too literally.
Loose wording works because it leaves room to solve the problem. I care about the outcome, a photo of the agent, near the top, looking right. I do not actually care about the geometry. But the moment I started describing the geometry, the geometry became the requirement, and the outcome I actually wanted stopped being the target.
That is the whole mechanism. Over-specifying replaces your real goal with your guess at how to achieve it.
What This Means for Your Business
When it is not working, try removing instructions, not adding them. This is the counterintuitive move and it is the whole lesson. Your next instinct after a bad result will be to explain harder. Try cutting the prompt in half instead. It costs you fifteen seconds to find out.
Describe the outcome, not the mechanics. “Write a listing description that makes the backyard the reason someone tours this house” is a goal. “Write a listing description with the backyard in the second paragraph, three sentences, mentioning the pergola and the mature trees and the fence” is a specification, and you have now made yourself responsible for a structure you did not need to design.
Notice when you are specifying how instead of what. The tell is that you are describing a method rather than a result. You do not want three sentences. You want the backyard to land. Say that.
Save the detail for the things that actually are requirements. Some things really are non-negotiable. Compliance language, your license number, the fair housing constraints, a specific price. Be exact about those and loose about everything else. The mistake is treating your preferences with the same rigidity as your requirements.
One Thing to Try This Week
Take the longest prompt you have written recently. One of the ones you were proud of, with all the detail in it.
Cut it to two sentences. What you want, and who it is for. Nothing else.
Run both.
I expect the short one to be better more often than you would guess. When it is not, you will at least know which of that detail was actually load-bearing, which is worth knowing either way.
The instinct that more explanation produces better results comes from working with people. It does not transfer cleanly.
About the Author
Clay Duncan is a Huntsville, Alabama Mortgage Loan Originator with Princeton Mortgage, NMLS #118739, and runs practical AI training for real estate professionals across North Alabama. He is currently building a suite of AI-powered tools for REALTORS®. This series covers what he learns doing it, including the parts that break.
If You Want to Go Deeper
Clay’s AI training for REALTORS® is built for practical adoption: what to use, when to use it, and how to make it stick. Sessions run for offices, teams, and partner groups.