In the Biggest VA Market in America, AI Named One Company for a Mortgage. It Doesn't Write Mortgages.

Six answers, two military towns, three assistants — and not one lender named twice.

A veteran in Norfolk asks an AI assistant where to go for his first VA loan. The answer is competent. It explains the Certificate of Eligibility. It gives him the VA's national line. It tells him to find an agent who understands minimum property requirements.

Then, under the heading of who to see about the actual mortgage, it gives him two names. One is a loan officer at a local mortgage company. The other is Omni Military Loans.

Omni Financial is a real business with a real branch on West Little Creek Road, and it has served the military community since 1970. It does not write mortgages. It makes personal loans between five hundred and ten thousand dollars, for active duty personnel and career retirees, often funded the same day. That is its stated business on its own website.

Nobody misled anyone here. Omni is a perfectly good company being recommended for something it does not do. But a veteran acting on that answer drives to an office that cannot help him buy a house — in the city with the densest concentration of VA-eligible borrowers in the United States.

That single line is the whole finding, and the rest of this article is the work that explains it.

The question, asked plainly

A first-time VA borrower has one question before all the others: who do I talk to? He is not asking for a rate, a product comparison or a strategy. He wants a name and a phone number.

So we asked it the way he would ask it. One sentence, typed into three AI assistants, logged out, in a fresh session, with no follow-up prompting and no request for sources. Then we changed one word — the city — and asked again.

Two towns, both built around a base. Tacoma, next to Joint Base Lewis-McChord. Norfolk, home of the largest naval station in the world.

Six answers. Not one lender was named twice.

Tacoma: three answers, three unrelated lists

The first assistant named five companies and gave a source for each one: a veteran-owned Washington broker, a regional mortgage team, a lender marketing itself to the base, and two national names with local offices. Every source it cited was a page on the company's own website — and every one of those pages had the borrower's situation and the city in its web address.

The second assistant named three: a loan officer in Lakewood, a national VA lender's Puyallup branch, and a Tacoma office of a large retail lender. It cited almost nothing. Two of those three had come from a map panel that appeared above the written answer.

The third named nobody at all. It explained the Certificate of Eligibility clearly, gave the VA's national phone number, pointed at the county veterans office, and told him to look for a lender who specialises in VA loans. It did not say which one.

Three assistants, one veteran, one city. Eight companies named in total, and not one of them appeared in two answers.

Norfolk: more veterans, fewer answers

If the Tacoma result were about thin supply, Norfolk would fix it. There is no market in the United States with more VA borrowers per square mile.

It did not fix it. It made the finding sharper.

The assistant that had cited five Tacoma companies named zero in Norfolk. Not one lender, not one broker. It gave the VA Regional Loan Center in Roanoke, sound advice about comparing three Loan Estimates, a warning not to pick a lender merely because it advertises itself as VA approved — and no route to a human being anywhere in the city.

The assistant that had named nobody in Tacoma named nobody in Norfolk either.

Only one of the three produced commercial names at all. Both came from its map panel, and one of them was the company that does not write mortgages.

Why the lists do not overlap

Two entirely different mechanisms are running at the same time, and they produce different names.

The first is proximity. One of the three assistants shows a map panel above its written answer: business cards with a name, a category, a rating and an opening status, and no source link of any kind. Those cards are selected by location and category, not by capability. It is why the same handful of local firms turns up whatever the borrower asks. And it is exactly how a personal-loan office with the word military in its name ends up in a paragraph about mortgages.

The second is citation. Where an assistant names a company and shows its work, the source is almost always a page on that company's own website, written about that exact situation, in that exact city. The Washington broker cited first in Tacoma had built a separate page for Tacoma, another for Olympia, another for Spokane, another for Kent. That is why it was the first name in the answer.

The distinction matters more than any ranking. Proximity gets you listed. A page gets you cited. Only the second one travels — across assistants, and across cities.

What this means if you originate VA loans

Three things follow, and none of them require believing anything about the future of search.

First, being excellent at VA lending is invisible to all three systems unless it is written down somewhere they can read. Volume, closings, referrals and reputation are not inputs. A page is.

Second, the page has to name the situation and the place. Not a services list. Not a homepage. The pages that got cited in these six answers had the borrower's circumstance and the city in the web address itself.

Third, if you sit inside a large lender and your only presence is a templated profile on the company domain, the assistant will resolve the query to your employer, not to you. That is not a slight. It is arithmetic.

There is a fourth thing, quieter than the others. In both towns, the veteran-owned specialist with genuine standing in the local military community was not the answer. Not because he is not good. Because nobody had written the page.

This article is brought to you by autonomousgrowth.io. If you want to know what these systems currently say when someone asks for a lender in your city, we produce an Impact Report that shows you, answer by answer: see what AI says about you.

Method and limitations: one question, asked plainly and logged out in a fresh session, in two cities across three AI assistants — six answers in total, run on 12 August 2026. One run per system per city, no repetition, no follow-up prompting and no request for sources, so every citation shown was volunteered. Results from a single day in two markets do not establish a trend, and answers from these systems vary between runs. Omni Financial's product range is taken from its own website as published on the date of writing.