If you have ever asked ChatGPT for “a good mortgage broker in my area” and watched it confidently describe a competitor, you already understand the problem an llms.txt is built to solve. AI assistants do not read your site the way a borrower does. They scan, summarize, and reconstruct your business from whatever signals they can verify. When those signals are thin or scattered, the model guesses, and a guess about a regulated lender is rarely flattering. An llms txt for a mortgage website gives the model a single, authoritative source to read first.
Think of llms.txt as a cover letter written for machines instead of borrowers. It sits at yoursite.com/llms.txt, it is plain text, and it states the facts that matter most: your legal name and DBA, your NMLS number, the states you are licensed in, the loan products you actually close, your service area, and links to the pages you want surfaced. It does not replace the rest of your AI-search work, but it removes the ambiguity that causes a model to describe you wrong or skip you entirely.
What an llms.txt file actually is
The llms.txt convention is a simple idea borrowed from a familiar one. For decades, websites have used robots.txt to tell search crawlers what they may access. An llms.txt file does something different: instead of permissions, it offers a curated, human-readable summary aimed specifically at large language models. The format is intentionally lightweight, usually Markdown, so the same file is easy for both a person and a model to parse.
For a mortgage business, that simplicity is a feature. You are not writing code. You are writing a short, factual brief about your practice that an AI assistant can lift verbatim when a borrower asks who to call. The goal is not to stuff keywords; it is to be unmistakably clear about the few facts that decide whether you get named.
Why mortgage websites need one
Mortgage is a high-trust, heavily regulated, intensely local business, and those three traits are exactly where AI assistants stumble without help. A model has to figure out whether you are licensed where the borrower lives, whether you do the loan they need, and whether you are a real, reputable name rather than a lead-gen page. Most loan officer websites bury those answers inside marketing copy, PDFs, or a footer that a model never reads cleanly.
- Licensing clarity. Your NMLS number and licensed states are non-negotiable facts. Stating them plainly keeps a model from recommending you in a state where you cannot lend.
- Product accuracy. If you specialize in VA and first-time-buyer loans, say so. Otherwise the model may describe you as a generic lender and lose your edge.
- Local intent. Borrowers ask AI assistants city-specific questions. A clear service area helps you win them, which we cover in our guide on how loan officers win “best mortgage lender in [city]” in AI.
This is also where llms.txt fits into the bigger picture of answer engine optimization: it is one structured signal among several that, together, make you legible to the systems borrowers now ask first.
What to put in a mortgage llms.txt
Keep it factual, current, and short enough that a model can read the whole thing. A workable structure for a loan officer or broker looks like this:
- Identity: legal name, DBA, NMLS number, and the parent brokerage if you operate under one.
- Licensing and service area: the states you are licensed in and the metros or counties you serve.
- Products: the loans you actually close most often (FHA, VA, conventional, jumbo, refinance, first-time buyer, non-QM).
- Proof: a line about reviews, years in business, or a notable result, with a link to where it lives.
- Key pages: labeled links to your highest-value pages, such as your purchase loan page, refinance page, contact page, and any city pages.
A simple example
| Section | Example line |
|---|---|
| Identity | Jordan Rivera, Loan Officer · NMLS #123456 · Cascade Home Lending |
| Service area | Licensed in WA, OR, ID. Serving greater Seattle and the Eastside. |
| Products | First-time buyer, FHA, VA, conventional, and rate-and-term refinance. |
| Proof | 4.9-star average across 180+ verified reviews; 11 years in lending. |
| Key page | Purchase loans: https://yoursite.com/purchase |
Notice what is missing: hype, rate promises, and anything you cannot back up. AI assistants weight verifiable facts and penalize claims that conflict with your other public sources, so the cleaner and more consistent your file, the more usable it is.
llms.txt is necessary, not sufficient
Here is the honest part. An llms.txt will not, by itself, make ChatGPT recommend you. Across the audits we run for loan officers, the file is the easy 10 percent; the visibility comes from the signals it points to. A model still wants corroboration before it puts your name in front of a borrower.
That means the foundation has to be in place: a complete Google Business Profile, a steady stream of real reviews, identical name-address-phone details everywhere you appear, and pages that answer borrower questions directly instead of burying the answer three scrolls down. We saw how much that combination matters with Keith Akada, a Seattle mortgage broker who went from invisible in AI search to the number-one AI-recommended broker in his market, generating roughly 30 leads and four closed deals in six weeks once those trust signals were built. The llms.txt helped the models read the work; the work is what earned the recommendation.
How it fits with your other AI-search work
Treat llms.txt as the index card on top of a well-organized file. It points models at the right pages, and those pages need to be worth landing on. If you are deciding where to invest first, our breakdown of AEO vs SEO for mortgage brokers explains how answer-first content and traditional ranking work together, and our broader AI SEO guidance for mortgage businesses shows how the pieces connect.
- Structured data on your pages confirms the same facts your llms.txt states.
- Reviews and directories give models the third-party proof they look for.
- Answer-first pages give them clean text to quote when a borrower asks a question.
When those layers agree with each other, a model has every reason to trust and name you. When they conflict, it hedges or moves on.
The bottom line
An llms.txt is a small, low-effort file that punches above its weight when the rest of your house is in order. It will not rescue a site with no reviews, mismatched licensing details, and thin pages, but it will make a credible mortgage business far easier for AI assistants to read, trust, and recommend. Write it plainly, keep it current, and treat it as the front door to the trust signals that actually win the borrower.