How we get paid
Nobody can pay to be higher on this page.
Ask for a rate on a big comparison site and, inside the hour, twenty phones ring. That is not a bug in those sites — it is the product. Your file was sold, once to each buyer, and every buyer calls because the first voice to reach you usually wins. LoanXray does not sell files. That is the easy half. The hard half is the order of the table, and the order is where the law actually lives.
disclosure alone does not make illegal fees legal under RESPAA disclaimer does not cure a violation. You cannot fix a table that steers by writing a paragraph under it. So this page is not the guarantee — the guarantee is how the table is built.
So this page is not the guarantee. The guarantee is how the table is built: one written sorting rule at the foot of it, the same type for every lender, and whatever does not qualify sitting at the end in gray instead of quietly disappearing.
Where the money stands today
Four numbers, and none of them is a promise.
Lenders paying us
0
Not one signed agreement. Everything in the comparison comes from public records — HMDA, FHFA, HUD, the Census — because there is nothing else it could come from.
What you pay us
$0
Nobody is charged for using this today. If that ever changes, it gets written here before it gets written anywhere else.
Commission at closing
None
We take nothing that depends on you closing, or on which lender you close with. Connecticut writes that exact condition into its lead-generator safe harbor. We hold it in every state.
Your data sold
Never
Not once, not in aggregate, not “with a trusted partner”. There is no lead here for anyone to buy, which is a design decision and not a policy.
And the market makes that honest instead of noble: the ten largest originators together hold 23.5% of all mortgages, and Rocket Mortgage — the biggest in the country — holds 6.3%. No set of private agreements could ever cover this market, so the comparison has to stand on the public record. It is the only surface wide enough.
The two business models
There are two ways a comparison site can make money. Only one of them lets you finish reading in peace.
Both are legal. They are not the same product, and the difference shows up in your phone rather than on the screen.
Model one
Sell the file
- 1You fill in the form and ask for rates.
- 2The site sells your file — commonly to four, six or ten buyers at the same time.
- 3Every buyer calls. Speed beats price, because whoever reaches you first gets the conversation.
- 4The site was paid the moment it sold you, whether or not you ever close, and whether or not the loan was any good.
The incentive is volume of files, not quality of loans. Nothing on the screen has to be false for this to work against you.
Model two
Charge for taking part
- 1You fill in the form. Nobody is told, because there is nobody to tell.
- 2The table is ordered by one written rule, identical for every lender in it.
- 3Nothing moves until you say so. When you ask, you send the same packet to the lenders you picked, on the same day.
- 4Lenders pay a flat fee to appear in the comparison — never contingent on your closing, never for a position in it.
The incentive is a comparison worth coming back to. That is the model the CFPB wrote down in February 2023, and it is the one this product is built against, line by line.
for their participation on a platform that both neutrally uses and neutrally presents information is receiving payment for compensable services, and thus would be compliant with RESPA section 8The regulator wrote our business model down before we did. A platform may charge lenders for taking part, as long as it uses and presents their information neutrally — because then the money is buying a service, not a referral.
The sorting rule
Sorted by what you actually pay, over the years you actually stay.
Every table on this site carries its sorting rule at the foot, in words, not hidden in a tooltip. And there are only a few orders we are allowed to use. We did not pick them from a design meeting: they are the short, closed list the regulator treats as neutral.
Why total cost and not the rate: in a single county, on 523 loans an underwriter could not tell apart — same program, same term, same occupancy, LTV between 79 and 81, DTI between 36 and 43 — the borrower at the 90th percentile paid 3.71 times what the borrower at the 10th paid in closing costs. $3,628 against $13,471. The rate is the number everyone advertises. This is the number that moves.
Orders the regulator treats as neutral
APR, lowest first
in useObjective customer-satisfaction information
not usedWhatever criterion you choose yourself
in useTotal cost including every charge, adjusted to the tenure you tell us — blessed by HUD in 1996 as a neutral display
in use
The one we leave alone is customer satisfaction. It is allowed, but we have no satisfaction data we could verify or that you could check behind us, and an unverifiable ranking input is exactly the place where a thumb ends up on the scale.
APR calculated to include all charges and to account for the expected tenure of the buyerHUD blessed this as a neutral way to display loan options in 1996, and the CFPB cited it again in 2023. Sorting by everything you pay, over the years you actually plan to keep the house, is not our clever idea: it is the presentation the regulator already approved.
Ties, and why the tie-break is boring on purpose
When two lenders land on the same figure, the tie breaks by LEI — the Legal Entity Identifier, the twenty-character code the regulator files them under — in ascending order. It is arbitrary, and that is the point: it correlates with nothing a lender could pay to move. A tie-break by volume, by name, or by anything called “relevance” would.
The Advisory Opinion says in a footnote that a complex algorithm is not a defense. So the rule here is blunt: if we cannot write in one sentence why a row is where it is, the order does not ship. Every row below carries that sentence.
The rule, running
The same rule, applied to a real county.
This is not an offer and not a recommendation. It is what these lenders actually charged, counted in the register they file with the regulator, in one county and one profile band. It is here because a sorting rule you cannot see running is just a paragraph.
| # | Lender | Loans in the band | Median closing cost | vs. the county median |
|---|---|---|---|---|
| 1 | ChaseLEI 7H6GLXDRUGQFU57RNE97 | 32 | $5,921.46 | −$890.42 |
Median of $5,921.46 across 32 loans in this same profile band in Harris County, 2025. Ascending by that median. | ||||
| 2 | United Wholesale MortgageLEI 549300HW662MN1WU8550 | 49 | $7,402.17 | +$590.29 |
Median of $7,402.17 across 49 loans in this same profile band in Harris County, 2025. Ascending by that median. | ||||
Below the threshold — shown, not hidden | ||||
| — | Rocket MortgageLEI 549300FGXN1K3HLB1R50 | 23 | — | — |
Only 23 loans in this band, under the minimum of 25. Below that count a median means nothing, so none is published for it and it is not ranked. It stays on the page, at the end, with the number that leaves it out. And it is the largest originator in the country — 429,332 mortgages in 2025, 6.3% of the market. The rule makes no exception for size, in either direction. | ||||
Order: median closing cost actually charged, lowest first, among lenders with at least 25 loans in this profile band in Harris County, TX, 2025 (HMDA). Ties break by LEI, ascending. Lenders below the threshold appear at the end, in gray, with the count that leaves them out. No lender pays to appear, and none can pay to move.
History, not an offer. HMDA carries no origination date, so part of any rate difference is simply the calendar. That is why this table compares costs and never rates.
Scope: this figure is blocks A, B and C of the Loan Estimate — what the lender charges and what it sends you to buy. It does not include transfer taxes, prepaids, the initial escrow deposit or the owner’s title policy. Those are the county’s side of the closing table, and they are calculated separately.
The blacklist
What this product will never do.
This is not a code of conduct written for a wall. Each line below is something that, the day we did it, would turn the fee we charge into an illegal referral fee.
The penalty is treble damages on the settlement-service charge, per borrower, plus attorney’s fees, plus a three-year window for state attorneys general. There is no version of this business that survives doing any of it once.
even where the fees paid by those providers are the same as one anotherCharging everyone the same does not save the arrangement if the screen still steers. A flat fee is necessary and it is not sufficient — which is why the work is in the table, not in the price list.
Auction off a position. Nobody bids for the top row.
The Advisory Opinion treats position auctions and per-participant pricing as evidence of an illegal referral fee.
Charge one lender more than another to be in the comparison.
Same reason — and note that charging everyone the same would still not save a screen that steers.
Transfer you live to whichever lender picks up first. No warm handoff, ever.
The CFPB has already brought an action over exactly this: Planet Home Lending, 2017.
Email or text you promoting one particular lender.
A perfect table does not survive an imperfect email. The message is itself an action that can affirmatively influence you.
Say “best match”, “recommended for you”, or “the right lender for your profile”.
Those phrases are a recommendation wearing the clothes of a filter. They do not exist here and they are not coming.
Give one participant bigger type, a bigger logo, a highlighted row, or two rows instead of one.
Typography is presentation, and non-neutral presentation is the first of the three elements that make the fee illegal.
Paginate. The good offers on page two, page one for whoever paid.
No table on this site paginates. Everything that qualifies is on the same page, in the same list.
Link only to the ones who pay.
Either every lender in the table is linked or none of them is. There is no middle version of this.
Show you only the top result when you come back.
The table comes back whole. A screen that remembers only the winner has quietly become a recommendation.
Use your financial profile to place your file with one particular lender.
Selective placement by the consumer’s own financial characteristics is what Connecticut’s lead-generator safe harbor forbids by name.
Rank by anything we cannot explain in one sentence.
Footnote 47 of the Advisory Opinion: a complex algorithm is not a defense. Opacity is the violation, not the excuse for it.
And one thing we will not say: that we are impartial. Saying it without being it is a violation on its own — the FTC fined LendEDU $350,000 for describing itself as “objective” and “unbiased”. What this page claims is smaller and checkable: here is the rule, here it is running, and here is how to reproduce it.
What goes in the packet
Five you hand over, and the one you never do.
Six pieces of information turn a question into an “application” under 12 CFR 1026.2(a)(3)(ii). The packet you take to lenders carries five of them and stops. The sixth is your Social Security number, and this site never asks you for it — there is no field for it anywhere.
| The six | In the packet? |
|---|---|
| 1Name | Yes |
| 2Income | Yes |
| 3Property address | Yes |
| 4Estimated property value | Yes |
| 5Loan amount requested | Yes |
| 6Social Security number, to pull the credit report | Never |
Stopping there is not squeamishness, it is the whole architecture. Without the sixth item there is no application, so no obligation to issue a Loan Estimate, no hard credit pull, and no trigger lead — the file the credit bureaus sell within hours of an inquiry, which is where the twenty phone calls come from. It is a legal decision dressed as a form field.
What we do instead
45
days, one inquiry
3–5
Loan Estimates
3
business days, by law
We hand you the packet and the script, and you request 3 to 5 real Loan Estimates yourself. The Loan Estimate is free, it is legally due within 3 business days, nothing can be charged for it except the credit report — and multiple mortgage inquiries inside a 45-day window land on your credit report as one. Then we normalize them and lay them side by side. Everyone else builds on advertised rates. The Loan Estimate is the only comparable document that exists.
“Within a 45-day window, multiple credit checks from mortgage lenders are recorded on your credit report as a single inquiry”
The law caught up on 4 March 2026: the Homebuyers Privacy Protection Act cut the bureaus’ sale of trigger leads. It left an exception for consumer consent. Asking you for that consent would be reproducing the problem with your signature on it, so we do not ask.
No lender calls you, because we never introduce you to one — an introduction that can affirmatively influence which lender you pick is a referral under Regulation Z, and a paid referral is the violation. It is you who sends, to the ones you chose, on the day you choose.
What this becomes
The honest version, said now instead of when it happens.
A comparison layer is a badly paid business, and pretending otherwise is why its three independent players walked out of it.
LendingTree’s home segment billed $151.8M in FY2025 against 6,822,375 mortgages originated in the country — about $22.25 per mortgage. Origination captures $10,321 per funded loan. That is a ratio of roughly 464 to one, and it is why Zillow bought its own lender and NerdWallet bought a brokerage. Anyone who tells you a pure comparison site is a great business is selling you the site.
Sources: LendingTree, Inc., FY2025 results (Home segment); Better Home & Finance, revenue per funded loan; national originations from HMDA 2025 (FFIEC/CFPB)
Today
A comparison medium that does not originate
- Who pays
- Nobody, yet. The plan is a flat participation fee from lenders.
- How much, and the limit on it
- Identical for every participant, never contingent on your closing or on receiving an application.
- What that money cannot buy
- A position, a typeface, an email, a live transfer, or a place in the table at all.
- Licensing and the rules that apply
- No federal MLO licence required: the SAFE Act test is conjunctive — you must take the application AND offer or negotiate terms. RESPA §8, Reg Z 1026.24, Reg N and ECOA all apply anyway.
Tomorrow
A licensed broker, in three to five states
- Who pays
- One side. Never both.
- How much, and the limit on it
- Reg Z 1026.36(d)(2) forbids dual compensation: either you pay us or the lender pays us, never both in the same transaction. It kills the “subscription plus commission” model, and we are saying so now.
- What that money cannot buy
- A rate. 1026.36(d)(1) forbids compensating an originator based on a term of the transaction, so nobody here can earn more because your rate went up an eighth.
- Licensing and the rules that apply
- Company licence per state, individual MLOs, 20 hours of pre-licensing education, 8 hours a year after that, and a test at 75%. It is the only structure that pays for the work this product does.
Never
A lender
Becoming the lender adds TRID, ATR/QM and HMDA obligations and closes no gap that the first two phases leave open. It is the obvious move and it is the wrong one.
The line, and which side of it we stand on
Market benchmarks, cost calculated from published rules, and what each lender actually charged last year: all fine, none of it is anybody’s terms. A rate attached to a lender’s name and to your financial profile is the other side. The distance between those two sentences is the entire business.
And a note on what does not change between phases: no phase of this pays anyone here more for sending you to one lender rather than another. That is not a policy we could revise. It is the condition under which we are allowed to charge at all.
referring includes any oral or written action directed to a consumer that can affirmatively influence…This is the line, and it is thin. Showing the market benchmark, the cost the published rules produce, and what each lender actually charged last year is not offering anybody’s terms. “Your rate with this lender would be 6.5%” is. We stay on this side of it, in writing, from day one.
Check us
Do not take our word for it. The point is that you do not have to.
Trust in this category is worth nothing, because everyone claims it. What is worth something is a claim you can go and check without asking us for anything.
- 01
Every figure carries its source and its date
Constants live in one file with the primary source, the effective date, and a provenance flag. Anything not verified against a primary source is marked as such on the screen where it appears — disputed, unverified, or market convention.
- 02
The sorting rule sits at the foot of the table
Not in a help centre, not behind an info icon. Written where the table ends, in the language you are reading the page in.
- 03
Every row says why it is where it is
One sentence per row, generated from the same numbers that produced the order. If the sentence cannot be written, the row does not get ranked.
- 04
The order is reproducible by a stranger
The HMDA file is public and needs no key. Download the same county, apply the same band, and you get the same order — including the tie-break, which is why the tie-break is a public identifier and not a score of ours.
- 05
Nothing that qualifies is hidden
What fails a rule goes to the end of the list, dimmed, with the reason and the rule cited. A screen that hides its exclusions is deciding for you and calling it design.
One more, and it cuts against us: although we do not lend, a platform that selects which lenders see a file is a creditor for anti-discrimination purposes under Reg B 1002.2(l). So the sorting rule has to be tested for disparate impact — and that is a second, independent reason no ordering criterion here depends on who you are.
The English notice at the top of this page appears on every screen that personalizes a figure, in type no smaller than 12 points. It is regulated text: we did not write it and we are not allowed to improve it. The Spanish line under it is ours, and it is there because reading a warning you do not understand is not a warning.
What we do not know
The opinion this business stands on has never been tested.
The Advisory Opinion has never been the subject of an enforcement action, so there is no case law marking where the real edge is — only the text. Which is why we build literally against the text, in writing, from the first commit.
notwithstanding that after such act or omission has occurred, the interpretive rule is amended, rescinded, or determined… to be invalidGood faith is protected even if the opinion we rely on is later changed, withdrawn or struck down — but only if the conformity was documented while it was in force. That is why the rules live in the code and in this page, dated, instead of in somebody’s head.
Two things are moving under all of this and it would be dishonest to leave them out. Federal enforcement is depressed while the CFPB’s funding is in dispute, which changes the odds of being challenged and changes nothing about the law. And the CFPB’s July 2026 request for information asks whether the Loan Estimate could be delivered earlier in the process — if that changes, the packet-and-script design on this site changes with it.
Now go look at the table this page is about.
It opens with a real case already loaded, so there is something to read before there is anything to fill in.