Ask ChatGPT how much house you can afford, and it will give you an answer in seconds. The math may even be correct. The problem is that ChatGPT has no way of knowing whether the income you typed qualifies under mortgage guidelines, which of your debts an underwriter will actually count, what your real interest rate and mortgage insurance cost will be, or whether the house you're looking at carries $250 or $800 a month in taxes and HOA dues. That's why an AI affordability estimate can look precise and still land tens of thousands of dollars away from the price you can actually finance.
This is a fast-growing search behavior. In a LendingTree survey of 2,001 consumers published August 31, 2026, 72 percent said they'd use AI for at least one homebuying or selling task, including 35 percent specifically to find homes within their budget and 24 percent to locate down payment assistance. A separate Veterans United survey found 59 percent of buyers had used at least one AI platform somewhere in their homebuying research, and 45 percent of prospective buyers said they'd used AI specifically to help with buying a home.
I'm Elliott Bowman, NMLS #1982189, a mortgage broker at Your Mortgage Copilot, powered by Xpert Home Lending, working with 120+ lenders across conventional, FHA, VA, USDA, jumbo, and non-QM loan programs. Here's what an AI affordability number is actually built from, what it can't see, and how to get one you can actually shop with.
How many homebuyers are already using AI to estimate affordability?
A significant and growing share. LendingTree's August 2026 survey found 72 percent of consumers would use AI for at least one homebuying or selling task, with 35 percent using it to find homes within budget and 24 percent to locate down payment assistance. Veterans United separately found 59 percent of buyers had used at least one AI platform during their research, and 45 percent had used AI specifically to help with the buying process itself. These tools are fast, free, and increasingly the first stop before anyone talks to a lender.
Is the 28/36 rule how mortgage approval actually works?
No. It's a reasonable consumer budgeting heuristic, but it isn't how modern automated underwriting decides what you qualify for. Many consumer affordability calculators and AI tools default to a version of the 28/36 rule: keep housing costs near 28 percent of gross monthly income and total debt near 36 percent. That's a fine sanity check for budgeting. Actual mortgage approval runs through an automated underwriting system, and allowable debt-to-income ratio varies substantially by loan program, credit profile, reserves, and other risk factors. Fannie Mae's automated underwriting can allow total debt-to-income up to 50 percent for well-qualified borrowers, while its manually underwritten guidelines start around 36 percent and can extend to 45 percent with compensating factors. The rule of thumb and the real approval math are related, but they are not the same thing.
What does AI actually not know when it estimates your affordability?
Four things: your real qualifying income, which debts underwriting will count, which loan program you'll be placed in, and the specific property's costs. Tell ChatGPT you earn $180,000 a year and it will divide by 12 and hand you a clean monthly figure. Underwriting doesn't work that way. Bonus and overtime income typically need a history and get averaged rather than annualized from a recent pay period. Commission income is calculated differently than salary. Self-employed income is built from tax-return analysis, not gross revenue. Rental income has its own calculation method entirely. Fannie Mae's guidelines require qualifying income to be stable, documented, and reasonably expected to continue, which is a judgment call, not a division problem. On top of that, underwriting also weighs cash reserves, lender-specific overlays layered on top of standard guidelines, and your actual credit-based pricing for rate and mortgage insurance, none of which a generic calculator has access to.
What does this look like with real numbers?
Imagine you tell ChatGPT you make $12,000 a month, carry a $500 car payment and $300 in student loan payments, and plan to put 10 percent down. It can produce a perfectly reasonable affordability estimate from those three numbers. But what if $3,000 of that monthly income is overtime you've only been receiving for six months? What if the home you're considering carries a $450 HOA payment? What if homeowners insurance runs $400 a month instead of the $150 the calculator assumed?
The math wasn't wrong. The inputs were never underwriting inputs to begin with.
AI isn't necessarily bad at mortgage math. It lacks the information needed to know which numbers belong in the math.
Why can two lenders approve the same borrower for different amounts?
Because they may offer different loan programs, apply different overlays, price the risk differently, or document non-W-2 income differently, not because the underlying agency guidelines change from lender to lender. For a standard conventional loan, the core Fannie Mae or Freddie Mac requirements don't shift based on who you call. What does shift is which programs a lender offers, what overlays they layer on top of the minimum guidelines, how they price your rate and mortgage insurance, and how flexible they are in documenting income that doesn't fit neatly into a single W-2.
Elliott can review your actual income and debts and tell you what you qualify for — before you rely on an AI estimate or a generic calculator.
Is a preapproval always more reliable than a calculator estimate?
Only if you know what was actually reviewed to produce it, because the industry doesn't use the terms "prequalification" and "preapproval" consistently. Some lenders use "preapproval" for a lightly reviewed estimate, while others reserve it for a file where credit and documentation have actually been verified. Don't get hung up on what the letter is called. Ask what was reviewed to produce it. When I issue a preapproval, I've pulled credit and reviewed actual pay stubs, tax returns, or bank statements, depending on how you're paid, rather than accepting numbers you typed into a form. That gives you a far more reliable number to shop against, though the specific property's taxes, insurance, HOA dues, and other costs still matter once you're under contract.
AI tools are a reasonable way to get comfortable with terminology and rough out a price range before you start touring homes. Run our mortgage calculator first if you want a starting point. Before you make an offer, get a preapproval built from your actual income and debt, reviewed against real underwriting guidelines, not an estimate built from a generic formula. See why getting preapproved is the first real step, or call or text (206) 949-5563, or start at YourMortgageCopilot.com.
Elliott Bowman | NMLS #1982189 | Your Mortgage Copilot, powered by Xpert Home Lending, Inc. (NMLS #2179191)
(206) 949-5563 · Erie, Colorado