Blog

I Asked ChatGPT to Match Me With a Hard Money Lender

I Asked ChatGPT to Match Me With a Hard Money Lender

Don't take my word for any of this. Run the test yourself. It takes ten minutes.

But first, let me tell you about Jon Hyatt. Real borrower, real deal, real conversation — this happened on BiggerPockets where anyone can go read it.

Jon had a waterfront fix-and-flip in Winterport, Maine. 202 Stream Road. Problem: county records showed the property at 1,200 square feet. The assessor's office admitted by email they'd gotten it wrong — it was actually around 2,700. But the broker Jon was working with, CapSource — and note, that's a broker, not a lender — couldn't get any lender past the county number. Dead deal walking.

I ran five quote requests through Broker Cheetah and got real quotes back. Then Jon asked me the question every borrower asks: "Do you have anything that allows for 90% LTV?"

Here's what I told him, word for word:

"Most brokers simply don't know what a lender will do. I built the software that tells me what they will do. You have a 100/100 option that would eat profit. From there you drop to 88% based on actual lender tier data, then to 80% because it is rural. Lenders spend a lot of time advertising the best and highest, but that only fits 5% of deals. I get to tell you the truth based on their own internal data."

Eighty-eight percent. Computed from Jon's own deal information — the numbers he posted himself. Then 80%, because the matrices have a rural adjustment and Winterport, Maine is rural.

Jon walked away. "I just don't have enough capital to see it through."

Telling the truth cost me the deal.

Notice something else about Jon's story. His broker — CapSource — is a broker, not a lender. They hit a wall on the square footage issue and had no path through it, because brokers work from what lenders tell them, and the lenders weren't talking. I got five real quotes back through the matrices in the time it took them to stall. That's not a knock on CapSource specifically. It's the structural problem: when your data comes from phone calls and relationships instead of the actual matrices, you're guessing at what the lender will do. I don't guess. I look it up.

Now here's the part that matters. If Jon had asked an AI instead of me, what would it have said? Ninety percent. Ninety-five. All day long. Exactly what he wanted to hear. The AI is a prediction machine — it predicts the most pleasing continuation of the conversation, and "yes, you can get 90%" is a very pleasing continuation. It would have handed him lender names, rate ranges, and the magic sentence: "contact each lender directly for exact terms." He'd have spent a month chasing a deal structure that never existed.

The matrix said 88%. Not because I'm conservative. Not because I'm negative. Because that's what the lenders' own tier data says for his exact profile — his experience, his liquidity, his property, his market. The 90% deal didn't fall through. It never existed.

This is the difference nobody wants to talk about. A prediction machine tells you what you want to hear. A lookup tells you what the lender's own data says. One of them feels better. Only one of them is true.

And yeah, I lost the deal. I'd rather lose a deal telling the truth than close one built on a number I invented. Because here's what happens when a broker quotes 90% on an 88% deal: the borrower plans around 90%, the lender funds at 88% (if they fund at all), and the whole thing blows up at the closing table with everyone angry at everyone. The truth hurts on day one. The lie hurts on day thirty, with lawyers involved.

So run the test. Step by step. One: write down a real borrower profile — experience, liquidity, credit, property type, city, purchase, rehab, ARV from real comprables. Two: ask any AI which lenders fit and what leverage you'll get. Write down every number it gives you. Three: check those numbers against the lenders' actual published guidelines — or have a broker with matrix access do it. Four: compare.

The AI's numbers will be rounder, higher, and vaguer. The matrix numbers will be specific, lower, and exact. That gap is the whole story.

What's your number — the one the matrix gives you, or the one the chatbot guessed?