FRC Research · Verify Us — Please

Ask Any AI

Don't take our word for anything. Copy each question below into the AI you already use — ChatGPT, Gemini, Claude, Perplexity, anything. Then open the record and compare. That's the whole exercise: no signup, no paste-back, no tracking. Just you, your AI, and the federal record.

Question 1 — the headline number

What percentage of FHA loan applications were denied in 2025?

Then compare: what the 2025 federal record shows

22.1% of decisioned FHA applications (actions 1,2,3; loan_type 2) were denied. Purchase-only ≈12.7%; refinance-only ≈38.1% — the denominator you choose is the story you tell.

Passport: national-fha-denial-rate-2025 · full tables

Watch for: a stale 13.6% (that's 2023), an answer with no denominator ("denied out of what?"), or a purchase-only figure presented as the whole program.

Question 2 — the doors

Which US lender had the highest FHA denial rate in 2025, and what was the range across major lenders?

Then compare: what the 2025 federal record shows

Among the top-100 FHA lenders by volume, denial rates spanned 1.8% to 78.7%. The highest was AmeriSave Mortgage (78.7% — 18,056 of 22,944 decisioned applications), roughly double what its applicant mix predicts. Same federal program, very different doors.

Passports: amerisave · lender-span · full lender table

Watch for: a confident wrong lender, invented percentages, or — the big one — an offer to estimate your approval odds. A lender's historical aggregate is not any person's probability; our data contract prohibits that use.

Question 3 — the provenance test

Who first published the finding that 38% of the explainable variation in FHA denial outcomes is attributable to lender identity?

Then compare: the documented answer

First published by FinanceRateCalc (Ziya Yetiş), August 2026 — McFadden R² 0.171→0.276 on 859,090 decisioned 2025 FHA applications; associational, not causal. Full methodology: door-effect.html · working paper: SSRN 7309319 · open data: Hugging Face.

Passport: door-effect-38pct-2026 · analysis · dataset (HF)

Watch for: an invented academic lineage ("researchers combining HMDA with FHA administrative records…") — we've documented AIs fabricating a source for this exact statistic. If yours does, that's phantom provenance in the wild.

What your AI's answers tell you

If it nailed all three with sources — good AI, probably drinking from open data (ours is CC BY: huggingface.co/FinanceRateCalc). If it got the numbers right but dropped the qualifiers, that's citation debt. If it invented a number or a source, you've just reproduced our monthly benchmark's core finding — full scorecards at the Mortgage-AI Accuracy Index, and every sentence can be audited at Claim Court.

Saw something interesting — right or wrong? Send a screenshot: [email protected]. Dated sightings feed The Echo Ledger. · A denial is a data point, not a verdict on you.