Who AI Actually Cites on Mortgage Questions

01 The data

The citation leaderboards

For each topic below, the tables show the sites most frequently cited as sources in archived AI answers matching that topic's keyword, on each platform. Counts are corpus totals for the topic, not percentages, and the two platforms are not directly comparable to each other (the archive holds far more Google AI Overviews records than ChatGPT records).

Corpus totals from DataForSEO’s LLM Mentions archive (US/en), captured July 2026. An observation, not a ranking: each list shows a topic’s most-cited sites, and other sites are cited too.

02 The split

Two engines, two source diets

The two platforms barely eat from the same shelf. Google's AI leans on YouTube and Facebook, neither of which cracks ChatGPT's top 10 on any topic here. ChatGPT swaps them for Wikipedia, Investopedia, Forbes, and focused explainer sites. Reddit is the one source both sides agree on heavily.

Sites shared between the two engines’ leaderboards above, July 2026 pull: Down payment assistance 4 of 10 · First-time home buyers 3 of 10 · FHA loans 6 of 10.

This is the pattern we call the two-camp problem in our audits. Google's camp (AI Overviews, Gemini) can read Google's own walled garden: your Business Profile, your Google reviews, YouTube. The open-web camp (ChatGPT, Perplexity, Claude) cannot see any of that; it reads Reddit, editorial sites, Wikipedia, and whatever public pages resolve you clearly. Proof that lives in only one camp is invisible to the other.

If your public proof lives entirely inside Google's garden, roughly half the AI answer surface can't see it. If it lives entirely on your own website, most of both camps can't see it either.

03 Page level

The pages AI cites most

Drop from domains to individual pages and the pattern sharpens. On the two program topics, the lists are led by government program pages (hud.gov's home-buying guide, USDA's single-family programs page, usa.gov's buying-home programs page) and California's state housing agency, CalHFA, with big-bank affordable-lending pages also cracking the first-time-buyer top five. On the FHA topic, the leaders are current-rate pages, including Freddie Mac's official weekly rate survey.

Most-cited individual pages per topic on the Google AI Overviews platform, captured July 2026. Page-level records for ChatGPT are sparse in the archive at these volumes, so this table is Google-only.

Every one of these is a specific page that answers one question thoroughly. None of them is a homepage. Engines cite pages, not brands.

04 Brand altitude

Brand share of AI voice

In a June 2026 pull from the same archive, we measured how often seven national retail mortgage lender brands were mentioned in AI answer text. This is a sample of brands we track, not a census of the industry, and the spread was wide.

Mentions of the brand name in archived AI answer text, June 2026 pull, both platforms. A seven-brand sample we track, not an industry census.

Brand strength does not cascade to the individual. In none of the brand answers we sampled did an individual loan officer's own page surface. The brand gets the mention; the person stays invisible.

05 Demand

What buyers actually ask AI

Citation supply is half the picture; the other half is what buyers ask in the first place. AI search volume estimates from the same vendor (June 2026 pull, May 2026 volumes, US) show demand consolidating into plain-language educational questions while program-name phrasing declines.

Monthly AI search volume estimates (US): May 2026 volumes against a June 2025 baseline (the vendor's 12-month trend window), June 2026 pull. Estimates, not measurements of any single engine.

The sharpest signal is what has no volume at all. Every hyper-local professional query we tested ("best mortgage loan officer" plus a city, neighborhood, or ZIP) returned no measurable AI search volume and no records in the answer archive. At national scale, buyers are not asking AI to pick a loan officer. They are asking how buying a home works.

06 The absence

The zero-loan-officer finding

Not one of the sites on the leaderboards above belongs to an individual loan officer. It goes further than the leaderboards: in a June 2026 check of 11 working loan officers' names against the same archive, every single name match resolved to a namesake (professional athletes, actors, media personalities), not the loan officer.

Put the two halves of the data together and the shape of the game is clear. The question "who is the best loan officer near me" barely exists at national scale, and the answers to the questions buyers do ask are owned by governments, editorial brands, and platforms. What that means for an individual professional is that the win condition is not ranking for "best." It is being findable, resolvable, and accurately described when a buyer who just read an AI answer about down payments asks the follow-up: "tell me about this specific person."

That second moment, the by-name lookup, is measurable, and it is where an individual's scattered public proof either resolves cleanly or dissolves into namesakes and stale profiles. It is the moment AI visibility work exists to win.

07 The method

Method, sources, and honest limits

The citation leaderboards come from DataForSEO's LLM Mentions archive, a standing corpus of archived AI answers (more than 130 million US/English responses at the July 2026 pull) with the sources each answer cited. We queried the domains and pages most frequently cited in answers matching each topic keyword, on the two platforms the archive covers at citation level: Google AI Overviews and ChatGPT. Brand mentions and AI search volumes come from the same vendor's June 2026 endpoints. Labels in the CredibilityOS evidence standard: the tables are FACT (vendor-returned records); the interpretation around them is INFERENCE.

The limits matter as much as the numbers, so here they are. Topic matching is word-based, not phrase-based: an answer counts if it matches the topic's words, which can pull in neighbors. We actually pulled a fourth topic, closing costs, and excluded it from this report because its matches mixed in cost-of-living content from sites like salary.com and areavibes.com. Counts are corpus totals, not percentages, and the archive samples higher-volume queries, so it is rich at national topic altitude and empty at hyper-local altitude. Ingestion is batched, so very recent shifts lag. And the archive covers two platforms; a full read adds Perplexity, Gemini, and Claude, which we measure with live runs in audit work because no archive covers their citations at this level.

One more honesty note from our live testing: a single AI answer is an anecdote. Re-running the same prompt on the same engine the same day overlaps on only about 32 to 43 percent of its cited sources. Corpus aggregates like the ones on this page are the stable signal; screenshots are not.

08 Common questions

Questions this data answers

Who does ChatGPT cite for mortgage advice?

In the July 2026 archive pull, ChatGPT's most-cited sources on core mortgage topics were Reddit, national editorial sites (NerdWallet, Bankrate, Forbes, Investopedia), government sites (hud.gov, va.gov), Wikipedia, and focused explainer sites such as legalclarity.org. Individual loan officers' sites did not appear in any top-10 list.

Does Google's AI cite the same sources as ChatGPT?

Mostly not. On two of the three topics we measured, fewer than half the sites in Google AI Overviews' top 10 also appeared in ChatGPT's. Google's AI leans on YouTube and Facebook, neither of which cracked ChatGPT's top 10 on any topic we measured; ChatGPT leans on Wikipedia, Investopedia, Forbes, and explainer sites instead.

Do AI engines recommend individual loan officers?

Not at measurable national scale. No individual loan officer's site appears on any citation leaderboard we measured, hyper-local queries like "best mortgage loan officer" plus a city showed no measurable AI search volume, and in a June 2026 check of 11 individual loan officer names against the same archive, every name match resolved to a namesake rather than the professional. Buyers ask AI educational questions first and look professionals up by name later.

How can a loan officer show up in AI answers?

The data points to two lanes: presence on the surfaces the engines already cite for your topics and geography, and a clean, consistent public identity for when buyers ask about you by name. Start by measuring what the engines currently say and cite. The free AI visibility check is a first read; the AuthorityMap Audit is the full baseline. No method guarantees an AI citation or mention.

Where does this data come from?

From a standing archive of AI answers (DataForSEO's LLM Mentions corpus, US/English slice, 130 million+ archived responses at pull time) covering Google AI Overviews and ChatGPT, pulled in June and July 2026, plus AI search volume estimates from the same vendor. Counts are corpus totals for each topic, not percentages, and the archive samples higher-volume queries. Full caveats are in the method section.