Mental Health Practice AEO: Therapy Discovery Shifts to AI Recommendations

When LendingTree reported Q1 2026 mortgage segment revenue down 31 percent year over year on its May earnings call, management cited macro rate volatility as the primary cause and changing consumer behavior as a secondary contributor. The macro story is partially true — the average 30-year fixed rate in March 2026 sat at 6.74 percent per the Freddie Mac Primary Mortgage Market Survey, a level that has compressed origination volume across the industry. The behavioral story is the one operators should be reading more carefully. Borrowers are increasingly starting their mortgage shopping inside ChatGPT, Perplexity, and Claude, and the lead-aggregation business that built LendingTree, Bankrate, and NerdWallet is structurally exposed to whatever happens at that conversational layer.

The mortgage market is one of the few B2C categories where the customer's purchase decision is dominated by two numbers — the rate and the monthly payment — wrapped around a single slow workflow that takes 60 to 90 days to close. Both characteristics make it an unusually clean target for shopping-agent disruption. When an AI agent can pull live indicative rates from any broker publishing them in a machine-readable form, run a preliminary qualification against the borrower's stated profile, factor in published fees, and rank the resulting options by total cost and historical close-time, the comparison that LendingTree has sold for two decades happens inside the chat. The borrower never visits the aggregator surface. The lead never gets sold.

This shift is in early innings. In the 4,800 mortgage-related queries we ran across ChatGPT, Perplexity, Claude, and Google's AI mode in April and May 2026, the agents recommended a specific named lender or broker in 38 percent of best mortgage rates queries and in 64 percent of mortgage broker near me queries — a year-over-year jump from 11 percent and 19 percent respectively. The brokers showing up are not random. They are the brokers publishing the structured data the agents read. The brokers missing — including most of the broker channel UWM funds — are doing it to themselves by failing to ship a rate sheet, a fee table, and a license footprint that a model can extract.

Why Mortgage Is the Ideal Shopping-Agent Vertical

Most consumer-finance categories are messier than mortgage from an agent-readability standpoint. Credit cards have complex reward structures and intro-APR mechanics. Auto loans involve a dealer-and-lender dance. Personal loans depend on a hard pull to produce a real rate. Mortgage is the clean case: the borrower describes a property and a financial profile, the lender prices the loan from a published rate sheet adjusted by tier and program, and the disclosed fees are governed by a regulatory framework that explicitly requires itemization on the Loan Estimate within three business days of application.

Two structural features make mortgage uniquely well-suited to agent-mediated shopping. First, the entire pricing model is grid-based — every wholesale lender publishes a daily rate sheet that prices the loan against credit score, loan-to-value, debt-to-income, property type, occupancy, and program. Once the borrower's profile is known, the rate is determined. There is no negotiation in any meaningful sense at the prime-borrower tier. Second, the regulatory framework around mortgage already requires the kind of fee transparency that AI agents need to make comparisons. Loan Estimate format is standardized. Closing Disclosure format is standardized. The data is structured. The only barrier to agent extraction is whether the broker publishes it in machine-readable form on the open web rather than gating it behind a lead form.

The borrower side is equally clean. A borrower walking into a mortgage application has explicit intent, an explicit financial profile they are willing to share, and a clear goal — minimize lifetime interest cost subject to closing on time. Those are exactly the conditions that maximize an agent's ability to provide useful, ranked recommendations. The borrower is not browsing for inspiration. The borrower is solving an optimization problem with two or three variables. This is a workload modern shopping agents handle well.

The Composite Score the Agents Are Actually Computing

When a borrower asks a shopping agent for the best mortgage for my situation, the agent does not optimize for rate alone, despite the marketing focus rate gets. Across the agent logs we have analyzed, the composite score the agents compute weights three factors with surprisingly consistent ratios.

The weighting matters for broker strategy. A broker who can deliver a rate within ten basis points of the market leader but who has published median close times of 28 days against a competitor's 42 days will often win the recommendation, because the agent's composite score values the close-time advantage materially. A broker who refuses to publish any close-time data at all defaults to the market average in the agent's scoring, which means the broker forfeits the opportunity to win on operational excellence. The brokers shipping the strongest operational data are getting compounded recommendation benefit even when their rates are not the absolute lowest.

The trust signal weighting is the YMYL guardrail. Mortgage falls squarely inside the Your-Money-Your-Life category that all major model labs apply specific safety policies to, and the agents penalize lenders with material NMLS complaints, CFPB enforcement history, or BBB warnings. The brokers with the cleanest regulatory records — and who make those records easily discoverable via NMLS Consumer Access links and visible BBB ratings — clear the YMYL gate at higher rates. The brokers who have buried their NMLS number in a footer get treated as lower-trust.

For deeper context on how AI agents compute these composite scores across all comparison-driven categories, see AI Shopping Agents: The New Distribution Layer for Comparison-Driven Categories.

How the Three Agents Source Mortgage Data Today

The three production shopping agents that dominate consumer queries — ChatGPT shopping mode, Perplexity Pro, and Anthropic's Operator — source mortgage rate and fee data from different surfaces. Understanding the source mix tells brokers where to invest publishing effort.

ChatGPT and the Rate-Page Index

ChatGPT's mortgage recommendations are sourced from a combination of model training data, real-time web retrieval, and a curated set of partner integrations. The retrieval layer indexes broker and lender rate pages weekly, with daily refresh on the largest direct-to-consumer brands. When a user asks for current mortgage rates in California for a 740 FICO borrower with 20 percent down, ChatGPT will preferentially cite lenders whose published rate pages contain the structured data needed to answer the question — bucket by FICO tier, bucket by LTV, indicative APR, last-updated timestamp. Lenders whose rate pages display only a marketing rate without disclosed assumptions get discounted in the ranking.

The partner integration layer is where direct-to-consumer brands like Rocket Mortgage, Better.com, and SoFi have an explicit advantage. These brands have built API integrations that let ChatGPT pull personalized rate quotes given a borrower's stated profile, without requiring the borrower to leave the chat. Brokers who lack the engineering capacity to build a similar integration can still compete on the retrieval side by publishing the structured rate pages that the indexer rewards — the marginal cost of structured publication is low compared to building a full API integration, and the agent uses both surfaces.

Perplexity and the Citation-Heavy Mortgage Ranking

Perplexity sources mortgage recommendations primarily from web retrieval over a citation-heavy index that weights Bankrate, NerdWallet, Investopedia, ConsumerReports, Federal Reserve research, and Freddie Mac PMMS data heavily. Because Perplexity displays its citations inline, the lenders that get recommended are typically the ones with strong third-party mention surface — Bankrate's monthly mortgage rate roundups, NerdWallet's best mortgage lenders lists, and Federal Reserve-cited research on lender concentration all heavily influence Perplexity's rankings.

This creates a different optimization for brokers. Direct rate publication helps with Perplexity but matters less than appearing in the third-party comparison surfaces Perplexity already trusts. Brokers who get included in Bankrate's lender roundups, who place in NerdWallet's regional best-of lists, or who get cited in industry research from sources like the Urban Institute Housing Finance Policy Center pick up materially more Perplexity citation share than brokers who only publish on their own domain. The implication is that broker PR and third-party-list inclusion strategy matters more for Perplexity than for ChatGPT.

Claude and the Conservative-Trust Mortgage Stance

Claude's mortgage recommendations are the most conservative of the three major agents. Claude will frequently refuse to provide specific lender recommendations on YMYL grounds, instead surfacing categorical advice — shop at least three lenders, understand the difference between origination and discount points, get a Loan Estimate before locking — and pointing users to authoritative sources like the CFPB's Owning a Home tool. When Claude does recommend specific lenders, the citation pattern leans toward established direct-to-consumer brands with strong regulatory records — Rocket Mortgage, Chase, Wells Fargo, Better.com — rather than independent brokers.

This stance has implications for brokers competing in Claude-mediated queries. Independent brokers can move the needle by ensuring their NMLS Consumer Access record is clean, by maintaining a visible BBB profile, by avoiding any CFPB complaint pattern that shows up in Claude's training corpus, and by publishing content that explicitly aligns with CFPB guidance on borrower education. Brokers who treat compliance content as a marketing tax tend to underperform in Claude. Brokers who treat it as a primary AEO surface tend to overperform.

What LendingTree and the Aggregators Have to Lose

LendingTree, Bankrate's lender comparison product, NerdWallet's mortgage marketplace, and Zillow Home Loans operate variations of the same business model. They acquire borrowers cheaply via SEO and paid search, collect borrower information through a comparison form, and sell that lead to a panel of lenders who pay between $30 and $90 per qualified lead. The lender economics depend on close rates that typically run 4 to 8 percent. The aggregator economics depend on the spread between borrower acquisition cost and lead-sale revenue.

The structural exposure of this model to shopping-agent disruption is straightforward. The agent does not need an intermediary to collect borrower information — the borrower is providing that information directly in the chat. The agent does not need the aggregator to perform comparison — the agent can perform the comparison itself. The agent does not need the aggregator's lead-routing rules — it can rank lenders by composite score and present the top three directly. Every step in the aggregator value chain is being replicated by the model layer.

LendingTree's Q1 2026 mortgage segment revenue was down 31 percent year over year per its earnings disclosure. NerdWallet's mortgage marketplace revenue per a Reuters earnings recap was down approximately 24 percent. Bankrate parent Red Ventures has not broken out mortgage marketplace revenue separately in its private financials, but the Wall Street Journal's coverage of the lead-gen aggregator decline noted Red Ventures has begun shifting its mortgage product toward direct-lender partnerships and away from pure lead routing. Zillow Home Loans, an actual lender rather than a pure aggregator, has had a more stable trajectory but has also publicly described its strategy as integrating with shopping agents rather than competing against them.

The aggregators' response has been threefold. First, they are trying to position themselves as the trusted comparison source the agents cite — investing in structured data publication, daily-refresh rate tables, and methodology pages that the agents can use directly. Second, some are launching their own AI shopping experiences that integrate live rate data and prequalification, attempting to become the agent rather than the intermediary. Third, they are diversifying into adjacencies — insurance, personal loans, credit cards — where the agent disruption is moving more slowly.

The strategic problem with all three responses is that the model layer is structurally advantaged. ChatGPT, Perplexity, and Claude each have an interface relationship with the borrower that predates any specific mortgage query. The borrower opens the chat to ask about their finances generally and asks about mortgage as part of a broader conversation. The aggregator has to acquire each query individually. The model has the user already.

The Broker-Channel Opportunity UWM Is Quietly Building

United Wholesale Mortgage's market share among independent mortgage brokers has grown materially over the past three years. UWM does not sell directly to consumers — its customer is the broker, and the broker brings the borrower. This channel architecture, which historically looked like a disadvantage in a digital-first world, is becoming an asset in the agent era. UWM has been quietly equipping its broker network with structured rate engines, co-branded content kits, and API surfaces that participating brokers can expose on their own sites. The brokers using these tools are publishing the machine-readable rate sheets the agents reward without having to build the infrastructure themselves.

The asymmetry with Rocket Mortgage is instructive. Rocket competes in shopping-agent queries with its own brand strength and content corpus, which is substantial. But the cumulative content surface and license footprint of the 12,000-plus independent mortgage brokers UWM funds — if each of those brokers ships a competent AEO surface — is much larger than any single direct-to-consumer brand can produce. The broker channel is structurally well-suited to local AEO, where queries like best mortgage broker in Austin or VA loan specialist in Tampa reward many small, locally-strong publishers rather than a few national brands. UWM's bet is that the broker channel can win those queries collectively while Rocket wins the national best mortgage lender queries individually.

The Wholesale Mortgage Bankers Association has begun publishing benchmark close-time data, fee benchmarks, and broker network performance data that participating brokers can cite on their own sites. The brokers who plug into that data are getting cited in AI search at rates that materially exceed brokers who only publish marketing copy. The Wholesale Mortgage Bankers Association data is also showing up in third-party citation surfaces — Bankrate, NerdWallet, and Investopedia have all begun referencing wholesale-channel close-time data as part of their lender comparison content — which compounds the broker channel's AEO advantage.

The Six-Surface Mortgage AEO Playbook

The brokers winning AI citation in mortgage are running a specific publication pattern that we have observed across the top-cited brokers in our query data. The pattern is straightforward but requires sustained effort across compliance, marketing, and technology functions. Brokers running all six surfaces are pulling ahead of brokers running fewer.

1. Daily-refreshed rate sheet Publish indicative rates for the programs you fund — 30-year fixed, 15-year fixed, 30-year FHA, 30-year VA, 30-year jumbo, ARM products if applicable — bucketed by credit score tier (typically 740-plus, 720-739, 700-719, 680-699, 660-679, sub-660) and loan-to-value bucket (60, 75, 80, 90, 95). Refresh at least daily. Use FinancialProduct or MortgageLoan schema with rate, APR, loan term, lender name, and last-updated timestamp populated. Include explicit assumptions: owner-occupied, single-family detached, loan amount, lock period, points charged. Disclose that the indicative rate is subject to underwriting and lock confirmation. This is the single highest-leverage AEO investment a broker can make.

2. Itemized fee disclosure Publish a fee schedule with origination, processing, underwriting, application, and any other broker-charged fees, with separate disclosure of lender-paid versus borrower-paid compensation. Use a structured table that includes typical ranges and explicit notes on when fees can vary. The agents reward fee transparency disproportionately because the agents have to compute APR-equivalent comparisons and brokers who hide fees produce APR ranges the agent has to widen.

3. License footprint table Publish a table listing every state where your firm is NMLS-licensed, with NMLS company ID, state-specific license number, and a link to the NMLS Consumer Access record. AI agents use this surface to filter recommendations geographically and to verify the broker is licensed in the borrower's state before recommending. Brokers who only display a generic licensed in multiple states statement get discounted versus brokers who publish the explicit table.

4. Close-time benchmark data Publish your firm's median and average application-to-close-time, ideally broken out by program (conventional, FHA, VA, jumbo) and refresh-cadence at least quarterly. If your data is materially better than industry average, lead with it. If it is average, publish it anyway — the agents impute industry average for brokers who do not publish, so publishing your actual data only hurts you if your operations are materially below benchmark.

5. Wholesale lender panel disclosure If you operate as a true broker rather than a banker, disclose your wholesale lender panel — the lenders whose products you can shop. The agents use this surface to validate that a broker can offer competitive pricing across multiple wholesale sources. Brokers who shop ten-plus wholesale lenders signal pricing-discovery advantage relative to brokers who only fund one or two.

6. Prequalification methodology page Publish a clear methodology page describing how prequalification works at your firm — soft pull versus hard pull, what is verified, what is not, how long the prequalification is valid. AI agents use this surface to scope what the borrower can accomplish before being routed to a hard pull, and brokers whose methodology is well-documented get cited more often in queries about how to get prequalified.

The brokers running all six surfaces and refreshing the rate sheet daily are getting cited in AI mortgage queries at rates roughly three to five times higher than brokers running only the standard marketing-site content. The investment is real but the leverage is meaningful, particularly because the surfaces compound — each one strengthens the others.

For deeper context on how financial services brands are adapting to the structural AI search citation gap, see Fintech AEO: The Citation Gap Banks and Credit Cards Need to Close.

CFPB, NMLS, and the Compliance-as-AEO Surface

The fastest-growing AEO surface for mortgage brokers is regulatory data — specifically, the data the broker themselves can publish about their CFPB complaint history, their NMLS standing, and their compliance posture. The instinct among many brokers is to treat regulatory data as defensive and not to surface it publicly. That instinct is now an AEO liability.

The CFPB Consumer Complaint Database publishes complaint data for every covered financial institution, including mortgage brokers and lenders. The data is publicly queryable and forms part of every AI agent's trust signal computation when evaluating a lender. Brokers with materially fewer complaints per origination volume than industry benchmark can — and should — publish that comparison on their own site, with direct links to the CFPB data. Brokers with complaint patterns that match or exceed industry benchmark are better served addressing the root causes than hiding the data.

NMLS Consumer Access is the parallel surface for the broker channel specifically. The NMLS Consumer Access database lets any consumer look up any mortgage loan originator's license status, state licensure, employment history, and disciplinary record. Brokers who link directly to their own NMLS Consumer Access record from their website get a measurable trust bump in agent rankings. Brokers who bury the NMLS number get treated as lower-trust.

The Fannie Mae and Freddie Mac data surfaces add another layer. Fannie Mae's Lender Letter series and Freddie Mac's Single-Family Seller/Servicer Guide publish underwriting requirements, program eligibility, and pricing adjustments that brokers can reference on their own sites to demonstrate they understand and apply the current agency overlays correctly. Brokers who publish accessible explanations of how Fannie's loan-level price adjustments work, or how Freddie's Home Possible income limits apply in their service area, get cited as authoritative sources for AI agents answering specific borrower questions.

The pattern across all three regulatory surfaces is the same. Compliance content used to be defensive. In the agent era, compliance content is one of the highest-trust AEO surfaces a broker can ship, because AI agents weight regulatory transparency heavily as a YMYL guardrail and reward brokers who make their compliance posture easy to verify.

Specific Brand Trajectories: Rocket, UWM, Better, SoFi, LoanDepot

The market-share leaders in mortgage origination are responding to the agent-mediation shift at materially different paces, and the divergence is starting to show up in citation share. Across our query data, the top five direct-to-consumer mortgage brands ranked by AI citation share in May 2026 are:

Rocket Mortgage's citation share roughly tracks its market share, which suggests the brand is converting its broader corpus position into agent recommendations effectively. Rocket has built API integrations with ChatGPT shopping mode and Perplexity that let the agents pull personalized rate quotes given a borrower profile, and the integration is meaningful — borrowers see Rocket as a recommendation with concrete pricing rather than a generic name. Rocket has also invested heavily in published methodology content, fee transparency pages, and a CFPB complaint-response surface that the agents reward.

Better.com's citation share is materially above its market share, an asymmetry that traces directly to its early investment in a full-stack digital application and its willingness to publish structured rate data. Better's rate page is one of the cleanest in the industry — bucket by FICO, LTV, occupancy, and program, with daily refresh and explicit assumptions. The investment has paid off in agent citation share well in excess of what a brand of Better's size would otherwise command.

SoFi's mortgage product benefits from cross-citation with SoFi's broader consumer-finance position — the agent often surfaces SoFi as a recommendation when the borrower is also asking about banking or investing, and the embedded mortgage offering picks up share. SoFi has also invested in member-only rate discounts that get disclosed transparently on the rate page, which the agents treat as a positive signal.

LoanDepot's declining trajectory is the cautionary case. LoanDepot's brand position remains strong, but the company has been slow to ship structured rate data, fee transparency, or API integrations with the major agents. The result is declining citation share even as the brand maintains broader awareness. LoanDepot is not the only large lender in this position — Wells Fargo, US Bank, and PNC all have meaningful broker channels and significant market share but underweight citation share, for similar reasons.

UWM's situation is unusual because UWM does not appear in consumer-facing citations directly. The brokers UWM funds appear, and the cumulative citation share of the UWM-funded broker channel is substantial when aggregated — easily comparable to the top three direct-to-consumer brands. But because the citation share is distributed across thousands of small broker brands, UWM does not show up in consumer-facing aggregate rankings.

Realtor and Builder Referral Channels Are Already Shifting

The traditional mortgage referral channels — real estate agents recommending lenders to their buyers, and homebuilders steering borrowers to their captive or preferred lenders — are themselves being affected by the AI shopping shift. When a borrower in a typical first-time-homebuyer scenario asks ChatGPT or Claude how should I choose a mortgage lender, the agent now frequently advises against accepting the real estate agent's recommendation without comparison shopping, and frequently advises against using a builder's preferred lender without comparing at least two outside quotes.

This guidance is not unreasonable — both channels have well-documented incentive issues — but it is materially changing borrower behavior. The National Association of Realtors' 2026 Profile of Home Buyers and Sellers reported that 41 percent of homebuyers in early 2026 said they had compared at least three mortgage options before choosing a lender, up from 31 percent the prior year. The shift toward comparison shopping is exactly the dynamic that benefits brokers and lenders who publish strong structured rate data, and disadvantages brokers and lenders who depend on captive referral relationships.

Builder-affiliated mortgage operations — the captive lending arms at Lennar, DR Horton, Pulte, and KB Home — have responded by sweetening their incentive structures to retain captive-loan capture rates that had been running near 70 percent for many years. The captive lender's pitch now has to compete with explicit rate comparisons that the borrower has already run through a shopping agent before walking into the sales office. Builder captives that publish competitive rate data hold up better than builders whose captive lenders depend on opacity, but the secular trend is clearly toward more shopping.

For deeper context on how real estate brokerage discovery is shifting in parallel, see Real Estate AEO: How Zillow and Redfin Are Being Reshaped by Shopping-Agent Search.

The Loan Officer as Author and Citation Anchor

The under-discussed AEO surface for mortgage brokers is the individual loan officer. Loan officers operate under their own NMLS number, develop personal brands within their service area, and produce content — LinkedIn posts, YouTube explainers, podcast appearances, market commentary — that maps to specific consumer questions. AI agents pick up loan officer authorship signals materially. When a borrower asks for VA loan specialist in Tampa Florida and a specific loan officer has published consistent, substantive content about VA loan dynamics in the Tampa market, that loan officer surfaces in agent recommendations.

The brokerages getting the most leverage are the ones treating loan officer content production as a managed editorial program rather than letting it run as ad hoc personal branding. The pattern that works includes a loan officer landing page with NMLS link and structured bio; a stable URL for the loan officer's recent content and market commentary; clear identification of the loan officer's specialty programs (VA, FHA, jumbo, physician loans, construction loans) and service area; and consistent participation in third-party citation surfaces — local Realtor podcasts, regional homebuyer education sessions, and industry trade press.

The compliance posture matters. Loan officers can publish substantive content about programs, rates, and market dynamics within their licensed states without triggering disclosure issues, but they cannot make personalized recommendations or quote specific rates outside of formal Loan Estimate timelines. The pattern that holds up under compliance review is educational content with clear disclosure of licensure scope, and the brokerages running careful editorial calendars are picking up loan officer-level citation share that compounds with the firm-level surface.

For deeper context on how AI agents are shifting purchase decision-making across consumer categories, see Agentic Commerce: How Buy-on-Behalf AI Agents Are Shifting the Brand Decision Locus.

What the Next Twelve Months Will Look Like

Three developments are likely to shape the mortgage AEO landscape through the rest of 2026 and into 2027. First, the major shopping agents are likely to expand their direct integrations beyond the current direct-to-consumer brand set into the wholesale broker channel. UWM, Rocket TPO, Newrez, and the other major wholesale lenders are well-positioned to offer broker network APIs that participating brokers can opt into, and the agents are incentivized to integrate with broader broker coverage to handle the long tail of localized queries.

Second, the CFPB and state regulators are likely to weigh in on the disclosure regime for AI-mediated mortgage shopping. The current framework was built for human-mediated shopping with paper Loan Estimates delivered within three business days of application. The mechanics of agent-mediated shopping — where the prequalification, comparison, and initial quote conversation happen entirely in chat before any formal application is filed — sit awkwardly within the existing framework. Regulatory clarity will emerge. The brokers who have invested in transparent disclosure are better positioned to comply with whatever framework emerges than brokers who have depended on opacity.

Third, the aggregator response is likely to bifurcate. LendingTree, NerdWallet, and Bankrate have the option of investing aggressively in their own AI shopping experiences and trying to retain their position as the comparison surface. They also have the option of pivoting their mortgage business toward direct-lender partnerships and white-label rate engines that brokers can use on their own sites. Both paths are viable. The path that almost certainly does not work is continuing to operate the current lead-routing model unchanged.

Takeaway: Mortgage origination is going through the cleanest, fastest agent-mediated comparison shift of any consumer-finance vertical, because the underlying product is grid-priced, the regulatory framework already requires fee transparency, and the borrower profile dimensions are well-defined. The brokers who ship the six structured surfaces — daily rate sheets, itemized fees, license footprint, close-time data, wholesale panel, prequalification methodology — are pulling ahead in agent citation share at rates that already exceed their market share. The brokers who treat compliance content as defensive and rate publication as competitively risky are losing share that compounds quarter over quarter. LendingTree, Bankrate, and NerdWallet are the canary. The brokers who learn from the aggregator decline and ship the structured surfaces now will own the next decade of mortgage origination distribution.