At the Last Question Before the Sale, Local Mortgage Brokers Do Not Exist

Not because the answer leaves them out. Because the question does.

I measure what AI assistants say when somebody is looking for a mortgage. This week I ran it on an independent brokerage in the Phoenix market. Own brand, own website, years in business, real reviews. Thirty-four questions a borrower would actually type, put to ChatGPT, Claude, Gemini, Perplexity and Google's AI answers. One hundred and seventy live answers.

They were named six times.

Rocket Mortgage was named thirty-three. Guaranteed Rate, loanDepot, Fairway and Movement took most of the rest. Of every brand mention in their own category, in their own city, this brokerage held nine percent.

That is a familiar kind of bad news and it is not what stopped me.

Ask about them by name and everything works

The audit also runs six control questions that include the company name. Is this brokerage legitimate. What are their reviews like. How do they compare to the big lenders.

Six out of six. Every assistant knew them, answered clearly, and cited their own website as the source.

So this is not a business the machines have never heard of. It is a business the machines know perfectly well and never bring up.

The part I did not expect

The thirty-four questions are grouped by how close the buyer is to deciding. The last group is six questions: someone who knows what they need and is choosing a provider. The final question before the sale.

This brokerage appeared in none of them.

Then I read the six questions themselves.

Is Quicken better than Rocket Mortgage. Guaranteed Rate versus Better. Should I use a local lender or Wells Fargo. Fairway reviews. Guild versus Caliber. loanDepot customer service complaints.

Every single one is a comparison between national brands. Not one of them contains the possibility of a local broker at all. One asks whether to use "a local lender" — as a category, unnamed, against a bank that has a name.

The brokerage is not losing that comparison. It is not in it. By the time a borrower reaches the moment of choosing, the vocabulary he is choosing within was set somewhere upstream, and it contains six national names and one anonymous category.

Why this happens

An assistant answers with what has been written about, and national lenders have been written about endlessly. Comparison articles, review roundups, complaint threads, news coverage, forum arguments. Twenty years of the internet arguing about Rocket and Quicken and Wells Fargo.

Almost nobody writes "should I use this Phoenix broker or Wells Fargo". So that sentence does not exist for the assistant to learn from, and the question it asks itself is built from the sentences that do.

The local brokerage's own website is not the problem. It is well built and it gets cited the moment someone types the company name. The problem is that everything else ever written about the choice was written about somebody else.

What follows for anyone originating locally

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

First, an assistant describing you well when asked by name proves nothing. Those are two separate behaviours: knowing you and recommending you. Almost every business I measure is far better at the first.

Second, the stage that matters most is the one you can least reach from your own website. Your site answers questions about you. The decision questions are about the category, and they are settled on other people's pages.

Third, in the answers where this brokerage did appear, it was listed in passing and never assessed. Zero positive, zero negative, six neutral mentions. Being named is not the same as being recommended either.

There is a fourth thing, quieter than the others. The one local outcome in that whole set of decision questions was the phrase "a local lender" — no name, no reviews, no track record, just a category standing in for every independent broker in the city at once. Somebody will eventually own that sentence. At present nobody does.

Run it on your own business

You can have the same measurement done for your firm. Thirty-four questions a real borrower would type, put to five assistants, live, with no simulations and no extrapolation. It shows how often you are named, which lenders are named instead, where in the buying journey you disappear, how the assistants describe you when they do name you, and which questions nobody in your market owns yet.

Run the audit on your own domain. It costs nothing and there is nothing to sign.

This article is brought to you by autonomousgrowth.io.

Method: thirty-four buyer questions plus six branded control questions, each put to ChatGPT, Claude, Gemini, Perplexity and a live Google AI Overviews check, run on 25 August 2026. A mention counts when the brand name appears in the answer body or its cited sources. Answers where an assistant skipped its live web search are counted and marked. The brokerage is not named here; the figures are its own. Results from one business in one market do not generalise, which is why the measurement is run per business.