The short answer
Loan officers get found and cited by AI mostly through the review, directory and ranking sites AI cites for their city, so start there. Those listings are rented, though: your own website is the only page AI can cite that you own, so link every profile to it and build it to turn visitors into leads. On that site, the best bet is a page built for one city and one loan type that answers the borrower's question in its first sentence, the shape of the most-cited ranking site in our study. The rest is ordinary search work: make pages AI crawlers can read, put answers on pages that already rank, and spend little time on questions where regulators dominate the answers.
The short version:
- Check the review, directory and ranking sites AI cites for your city, keep an accurate profile on each one that offers profiles, and link each profile to your own website. In our ChatGPT study of September 24, 2026, 88% of answers to questions naming a city cited at least one directory, review or ranking site.
- Target the questions that name your city and your loan type. In the same study, an actual lender's website was cited in 36% of answers when the question named a city, 28% for "near me" questions and 10% (one answer out of ten) for learning questions.
- Build one page per city and loan type that answers the borrower's question in its first sentence. saferate.com, the most-cited of the data and ranking sites in the study, is built from pages like that.
- Build your website to turn visitors into leads. By our estimates from building loan officer websites (October 2026), most turn fewer than 2 in 100 visitors into leads, and a well-optimized site turns 5 to 15 in 100, depending on the traffic.
- Make sure AI crawlers can read your pages, and put answers on pages that already rank. One of the uncited posts we checked in September 2026 had 0 impressions in 90 days, so there was nothing for an AI Overview to cite.
- Spend little time on questions where regulators dominate the answers. In a 50-question ChatGPT pilot we ran the same day, consumerfinance.gov was cited in 31 of the 50 answers.
- Count citations, and expect few clicks from them. A blog post of ours that AI Overviews cited for all four of its head terms earned 9 clicks from 9,310 impressions over 90 days. We saw the citation in one snapshot, so we cannot say on how many of those 90 days it was cited.
The evidence comes from our two ChatGPT studies (498 prompts, and a 50-question pilot), our own work on leadpops.com (never client data) and linked public studies. The ChatGPT figures come from one run per prompt, and ChatGPT varies between runs, so treat them as one snapshot.
Who does ChatGPT cite when a borrower asks for a lender?
Mostly review, directory and ranking sites: in our ChatGPT study, 88% of ChatGPT's answers to questions naming a city cited at least one of them, and about a third (36%) cited an actual lender's own website. The 498 prompts covered 194 cities across all 50 states, a 50-state "near me" test and a control group of learning questions. We analyzed the websites each answer cited, not the answer text. Each prompt ran once, and ChatGPT gives different answers on different runs, so treat every percentage in this section as one snapshot. The learning-question group was small, so its 10% is a single answer out of ten.
- City named
I am buying a home in [city, ST]. Who should I use for a mortgage?
44%85 of 194 answers - City named
Who are the best mortgage brokers in [city, ST]?
28%55 of 194 answers - Near me, only a state set
How do I find a mortgage lender near me?
orWho is the best mortgage broker near me?
28%28 of 100 answers - Learning questions, ten asked, one shown
How much house can I afford on an $85,000 salary?
10%1 of 10 answers
Study run September 24, 2026. Both city questions together: 140 of 388 answers (36%), across 194 cities in all 50 states. The one learning answer that cited a lender's website was to the question shown. A lender's own website means a bank, credit union, mortgage company, broker or loan officer, big banks included; saferate.com is counted as a ranking site. One run per prompt, and ChatGPT varies between runs.
An actual lender's website here means a bank, credit union, mortgage company, broker or individual loan officer, big banks included. saferate.com is run by a licensed lender (PartnerOwn LLC dba Safe Rate Mortgage), but we class it as a ranking site because its pages rank other lenders; counted as a lender, the 36% would be 55%. The wording of the city question mattered: "I am buying a home in [city, ST]. Who should I use for a mortgage?" led to a lender's website being cited in 44% of answers, and "Who are the best mortgage brokers in [city, ST]?" in 28%.
- Directory and review sites56540%
- Data and ranking sites26419%such as saferate.com, monitorbankrates.com and originationdata.com
- Actual lenders' websites23417%big banks counted separately, below
- Publishers1148%
- Government1128%
- Other1047%state housing agencies, chambers, realtors, news
- Big banks91%
Study run September 24, 2026: 1,402 citation slots across 388 city answers. A slot is one site cited in one answer; each site counts once per answer. One run per prompt.
A slot is one site cited in one answer; each site counts once per answer.
Directory and review sites appeared in 56% of city answers, led by Expertise.com (cited in 160 of them), Birdeye (100, counting either of its two hostnames), Experience.com (62) and BBB (51), and data and ranking sites appeared in 48%.
Without a city, ChatGPT leaned on government and directory sites: on the "near me" prompts, government sites were cited in 50% of answers, directories in 48% and actual lenders' websites in 28%. We ran the "near me" prompts with only a state set as the location. ChatGPT's search was visible on 20 of them, and on 15 it searched for a specific town in that state, such as Gainesville GA, Wichita KS, Oak Grove KY, Minot ND, Matthews NC and Scarborough ME. So test with your city named in the question: when a prompt gave only a state, ChatGPT's searches stayed inside that state, and we did not test real borrowers' locations.
The most-cited of the data and ranking sites was saferate.com, cited in 113 of the 388 city answers, in 44 of the 50 states. It publishes rankings of other lenders for each city and loan type, built from public HMDA data, with URLs such as saferate.com/chicago/best-mortgage-lenders/fha-loans.
How can a loan officer rank in AI search without controlling the company website?
Start with the review and directory sites AI cites for your city, and link every profile you have to a page you own. In our ChatGPT study (one run per prompt), directory and review sites took 40% of the citation slots on city questions and actual lenders' websites took 17%, so you can be cited without controlling the company site. A listing is still rented: the directory decides who appears beside you, and many directories use your listing to generate leads you never get. LeadPops builds websites for loan officers; even so, in our sample the review and directory sites were cited more.
A similar pattern shows up outside mortgage: a University of Toronto preprint from September 2025 found, across 1,000 consumer ranking prompts in three product categories, that earned sources (independent media, review and comparison sites) made up roughly 69% to 92% of a search-enabled GPT model's citations, against roughly 32% to 54% of Google's results. It is a preprint and not about mortgage.
- Ask ChatGPT the questions from our study with your own city in them: "I am buying a home in [your city, ST]. Who should I use for a mortgage?" and "Who are the best mortgage brokers in [your city, ST]?" Then ask again for each loan type you do, and write down every site it cites.
- Take the review and directory sites from that list, make sure you have an accurate profile on each one, and link each profile to your own website; most profiles allow a link, paid or free. Across our study the most cited were Expertise.com, Birdeye, Experience.com, BBB, WalletHub, ThreeBestRated and Yably, but your market's list may differ.
- Ask your company whether it publishes a page for you or your branch, whether that page can be built around your city, and whether compliance allows you a website of your own.
- Wherever you can publish (a profile, a branch page, a personal site), build a page for one city and one loan type and answer the borrower's question in its first sentence.
- Re-run the same prompts every few months and compare the sites cited. ChatGPT varies from run to run, so look for the sites that keep appearing rather than reading much into one answer.
Do loan officers still need their own website if AI cites directories?
Yes. A directory profile can get you into the answer, but your own website is the only page AI can cite that you own, and the one place where every inquiry comes to you rather than to the directory.
What is the difference between a listing and your own website?
A listing is rented and your website is owned. On a directory or review site, the site decides the layout, who is listed beside you and where the inquiry form sends the lead; competitors are often on the same page, and many directories use your listing to generate leads you never get. On your own website, every page, form and offer is yours.
Rented: your listing sits beside competitors, and the inquiry can go wherever the directory sends it.
AI cites these pages often: a directory or review site was cited in 56% of ChatGPT answers about lenders in a city.
Keep it accurate, and link it to your website.
Owned: only you are on the page, and every inquiry comes straight to you.
AI cites these too: a lender's own website was cited in 36% of those answers.
Build it to turn visitors into leads.
ChatGPT study of September 24, 2026: 388 answers to questions naming a city, one run per prompt. In our experience building loan officer websites, many directories use your listing to generate leads you never get, and most profiles, paid or free, let you link to your website.
AI does cite lenders' own websites. In our ChatGPT study, an actual lender's website was cited in 36% of answers to questions naming a city, and in 44% of answers to "I am buying a home in [city, ST]. Who should I use for a mortgage?" Each prompt ran once, and ChatGPT varies between runs.
How do you turn a directory listing into visits to your own website?
Link to your website from every profile. Most directory and review profiles, paid or free, let you link from the profile to your own website. Point each link at the page that matches the listing, such as your page for that city and loan type, so a borrower who finds you on the directory lands on a page that answers their question and lets them reach you.
A borrower asks AI for a lender
The answer cites a directory profile, or your own website directly.
They land on a directory page
Your listing sits beside competitors. Many directories use your listing to generate leads you never get.
Your profile link brings them to you
The link on your profile carries them from the directory's page to your own website.
Your website turns them into a lead
On a well-optimized site, 5 to 15 of every 100 visitors become leads.
The lead closes
20% to 30% of leads from your own website close. Buyers without an agent can go to your agent partners.
Rates and the directory point are our experience building loan officer websites, not a study.
Why does your website's conversion rate matter when AI sends so few clicks?
Because the clicks are scarce. Our blog post on SEO for mortgage brokers was cited in the AI Overview for all four of its head terms in one snapshot, and over 90 days it earned 9 clicks from 9,310 impressions. When a click is that hard to win, the page it lands on has to turn it into a conversation, whether the visitor came from an AI answer, a directory profile or anywhere else.
In our experience building loan officer websites, most turn fewer than 2 in 100 visitors into leads, and more often fewer than 1. A well-optimized site turns 5 to 15 in 100 visitors into leads, depending on the quality of the traffic, and leads from a loan officer's own website close at 20% to 30%. These are our estimates from building and running these sites, not a study.
What are 1,000 website visitors worth?
By our estimates, a typical loan officer website that turns 1 to 2 in 100 visitors into leads gets 10 to 20 leads and about 2 to 6 funded loans from 1,000 visitors, and many get fewer; on a well-optimized site, the same 1,000 visitors become 50 to 150 leads and about 10 to 45 funded loans. We applied the same 20% to 30% close rate to both, so the whole gap comes from how many visitors become leads. Every buyer lead who does not yet have an agent is also an introduction you can make to one of your agent partners.
1% to 2% of visitors become leads, more often under 1%
- Leads
- 10–20
- Funded loans
- 2–6
Most loan officer websites turn under 2% of visitors into leads, more often under 1%.
5% to 15% of visitors become leads
- Leads
- 50–150
- Funded loans
- 10–45
Same 1,000 visitors. Even the low end, 10 funded loans, beats the best case of a typical site.
Each dot is one visitor; lit dots show the low end of each range. Both assume 20% to 30% of website leads close. Our estimates from building loan officer websites, not a study.
LeadPops builds conversion-focused websites for loan officers, so weigh this section knowing that. The rates are our estimates and the arithmetic is shown, so you can run it with your own numbers.
How should loan officers set up schema and website for AI search?
Make the site readable to AI crawlers, put your answers on pages that already rank, and treat schema markup and llms.txt files as upkeep: the best public test of schema we found, published by Ahrefs in May 2026, showed no citation gain.
Can AI crawlers read your pages?
Only if the text is in the page your server sends: AI crawlers that do not run JavaScript cannot read text that only appears after scripts run in the browser. In July 2026 we found our long guides were rendered only in the browser, so those crawlers saw a "Loading guide..." placeholder instead of the article. Fixing that one setting made every guide's headings and tables readable to them. One pillar guide went from 0 to 13 visible H2 headings and from 0 to 3 tables.
Then check robots.txt. Ours explicitly allows the AI crawlers, among them GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Extended. OpenAI's crawler documentation treats OAI-SearchBot, which serves ChatGPT search, separately from GPTBot, which collects training data, and a site that blocks OAI-SearchBot drops out of ChatGPT search.
Which pages should carry the answer?
Put the answer on pages that already rank: the uncited posts we checked barely appeared in search, so there was little or nothing for an AI Overview to cite. We tested the other explanation, that our uncited pages were less citable, and rejected it. We scored passages on a citability rubric: our most-cited sales page scored lowest, and an uncited post scored highest. One of the uncited posts had 0 impressions in 90 days when we checked. So we added how-to answers to pages that already had authority and rankings instead of polishing posts nobody saw.
How should each section be written?
Make each heading the question a reader would type and answer it in the first sentence, so the section still makes sense when an AI assistant lifts it out alone. On our main sales pages we use question-shaped H2 headings such as "Is Buying Mortgage Leads Worth It?", a direct answer in the first sentence, a matching FAQ entry, and figures that trace to one canonical source. Microsoft Advertising wrote in October 2025 that AI assistants split pages into chunks and can lift question-and-answer pairs "word for word". Keep the answer in the visible text rather than in tabs, accordions, images or PDFs.
Source every number so the page is checkable, without expecting it to move citations. Every number we publish has to trace to our own measured data, a named public dataset or a test we documented; if it cannot be traced, we cut it rather than hedge it.
Do internal links matter?
We do not know yet. We added internal links to under-linked pages on September 23, 2026, and have not measured a ranking change from it. Link your important pages from the pages people already visit anyway.
Does schema markup or an llms.txt file help?
Not for getting cited, according to the best public evidence we found (one controlled test of schema, and no major engine that has confirmed reading llms.txt), though we publish both. Ahrefs tracked 1,885 pages that added schema markup against 4,000 similar pages that did not, from August 2025 to March 2026. In May 2026 it reported that, relative to those controls, AI Overview citations fell 4.6%, while the changes for AI Mode (+2.4%) and ChatGPT (+2.2%) were indistinguishable from zero.
How do we optimize loan officer and branch pages for generative AI search?
Give each page one city and one loan type, use the borrower's question as its title, and answer it at the top. That is the shape of saferate.com, the most-cited of the data and ranking sites in our ChatGPT study (one run per prompt), cited in 113 of the 388 city answers. Its pages are built around the question: a city, "best mortgage lenders" and a loan type. They are ranked lists, which suits a "who should I use" question. Because they rank other lenders from public HMDA data, each page reads more like a reference than an ad.
Branch pages work the same way: one page for the branch's city and, where your company allows it, one for each loan type the branch does. In the same study, one large lender's branch pages were the only major-brand branch structure that showed up more than a handful of times in the city answers (13 of 388 city prompts).
Copying saferate.com fully means publishing rankings of other lenders, which is an editorial and competitive decision before it is a technical one. For a loan officer, it is a question to take to your compliance team first.
Which mortgage questions are hardest for a loan officer to win in AI search?
Learning questions about credit scores, loan types, eligibility and definitions are the hardest: ChatGPT sent them mostly to the CFPB and other government sources, and in our own Google tracking of mortgage-marketing keywords in September 2026 we were cited on none of the 10 definition questions. In the control group of our ChatGPT study, which was only ten questions, government and government-sponsored sources (including Fannie Mae and the NMLS registry) took 12 of the 17 citation slots (71%). In our separate 50-question pilot, ChatGPT cited consumerfinance.gov on 31 of 50 answers, or 62%.
On 4 of the 11 pilot questions where we could see its searches (7 of the 15 searches), ChatGPT had already written an agency into the query before it looked: the CFPB, the VA, the FHFA, or Fannie Mae and Freddie Mac. The searches it ran included "jumbo mortgage definition CFPB jumbo loan" and "VA home loan eligibility official VA 2026 entitlement funding fee occupancy requirements". A loan officer writing "what is a jumbo loan" is competing with the CFPB and other government and government-sponsored sources, which made up 6 of the 12 sites that answer cited; the same answer also cited two big lenders, and one individual broker was cited on a HELOC question.
Our Google data points the same way: we were cited in 43% of the AI Overviews on buying and product questions, 38% on vendor comparisons, 21% on how-to questions and 0% on definitions. Public data agrees: Pew, in a study published in July 2025 of Google searches from March 2025, found that .gov sites made up 6% of the links in AI summaries, against 2% in standard results. Write explainers for your own borrowers if they help them, and do not expect AI to cite them.
Does being cited by AI bring traffic?
Very little, in our data and in the public studies we found, so aim to have your name, your company or your page inside the answer the borrower reads. Our blog post on SEO for mortgage brokers was cited in the AI Overview for all four of its head terms in a September 23, 2026 snapshot. Over 90 days, in Search Console data pulled on October 14, 2026, it earned 9,310 impressions at an average position of 9.0 and 9 clicks, about one click per 1,000 impressions. We saw the citation in one snapshot, so we cannot say on how many of those 90 days it was cited.
- Clicked a result, no AI summary shown15%
- Clicked a result, AI summary shown8%
- Clicked a link inside the AI summary1%
Our post was cited for all four of its head terms in one snapshot, on September 23, 2026, so we cannot say on how many of the 90 days it was cited. Pew: 900 adults and 68,879 searches, published July 2025.
In our snapshot of the mortgage-marketing keywords we track, a Google AI Overview appeared on 809 of the 845 (96%), and on some of them we ranked near the top and still earned no clicks. For "mortgage leads cost", our page showed 7,159 times at an average position of 4.5 over 90 days and earned zero clicks, even after the AI Overview began citing us (the full story is below).
In a Pew Research Center study published in July 2025, of 900 adults and 68,879 searches in March 2025, people clicked a result on 8% of visits that showed an AI summary and on 15% of visits that did not, and on visits that showed an AI summary, they clicked a link inside it 1% of the time. Seer Interactive measured organic click-through rate on 3,119 informational queries from 42 client organizations, and in an update published in November 2025 put the Q3 2025 average at 1.45% with no AI Overview, 0.70% when the brand was cited in one and 0.52% when it was not.
With clicks this scarce, the page a click lands on has to turn it into a lead.
Do you need a page-one ranking to be cited by AI?
Not necessarily. A large public study published in March 2026 found most Google AI Overview citations came from outside the top 10, though in our own data the uncited posts we checked barely appeared in search at all. Ahrefs studied 863,000 results pages and about 4 million AI Overview URLs. In that study, 37.9% of cited URLs ranked in the top 10, 31.2% ranked 11 to 100 and 31.0% ranked beyond 100.
Both can hold at once: ours is a check of a few of our own posts, and theirs covers 863,000 results pages. Google's AI features documentation says there are "no additional requirements to appear in AI Overviews or AI Mode" beyond being indexed and eligible for a snippet. So get the page indexed and ranking for the question somewhere, and do not assume page two locks you out.
What has no good evidence of helping you get cited by AI?
An llms.txt file, schema markup, word-count targets and changing a page's date have no good evidence behind them as ways to get cited, and rewriting a page for AI models gives small, inconsistent gains. We publish llms.txt files and schema markup ourselves anyway, as upkeep; a live guide page on leadpops.com carried 6 blocks of schema markup when we checked. For schema, the best public test we found, an Ahrefs study published in May 2026, showed no citation gain.
The llms.txt format is a proposal, and no major engine has confirmed using it. In September 2026 we found ours still carried lead-cost figures we had already corrected on 27 pages, so the file meant to tell AI what is true was still wrong after the pages were fixed. We now generate it from one source so it cannot drift.
Write FAQ sections for readers; Google has phased out the search feature they used to earn. It limited FAQ rich results to well-known government and health sites in September 2023 and said they "will no longer appear in Google Search starting May 7, 2026", as its documentation change log records. We still write them, this guide included.
Rewriting a page to suit AI models gives small and inconsistent gains. The GEO study (KDD 2024) found that adding quotations, statistics and cited sources raised visibility by 43%, 33% and 28% on one of its measures, in a test engine the authors built. The gains went mostly to low-ranked sources, the top-ranked source lost 21% to 30%, and the engines tested were 2023 versions.
A later benchmark found less. C-SEO Bench, a NeurIPS 2025 benchmark of about 1,900 queries across four models, found a significant effect in only 3 of 54 "conversational SEO" cases.
Google's helpful content guidance warns against word-count targets and against changing a page's date without substantially changing the page. AI assistants do lean slightly toward newer pages: Ahrefs, in a study published in July 2025, found across about 17 million citations that AI-cited pages averaged 1,064 days old against 1,432 days for organic results. Google's AI Overviews, though, cited slightly older pages than organic results.
Whether backlinks drive ChatGPT citations at all is unproven in either direction.
How do you know if AI is citing you?
Check four places: Google Search Console, Bing Webmaster Tools, Google Analytics 4 and ChatGPT itself.
- In Google Search Console, export your queries and read the ones eight words or longer. From July 12 to October 10, 2026, 780 queries of eight words or more showed leadpops.com 13,998 times and earned 0 clicks. Their impression-weighted average position was 12.4, against 16.8 for queries of one to three words. Search Console hides rare queries, so the true count of long questions is higher.
- In Bing Webmaster Tools, open the AI Performance report. It shows how often Microsoft's AI (Copilot and Bing's AI answers) cites your site, and for which questions; it was in beta when we used it. Ours showed 1,872 citations across 46 questions, which the report calls "grounding queries", in the three months to October 14, 2026.
- In Google Analytics 4, open the default channel report and find the AI Assistant channel, which covers visits referred by AI assistants; Google's help page lists ChatGPT, Gemini, Deepseek, Copilot and Grok, and the channel does not include clicks from Google's AI Overviews. On leadpops.com, from September 13 to October 12, 2026, it brought 96 sessions and 12 key events (GA4's name for conversions), 12.5 per 100 sessions, against 9.0 for organic search (1,272 sessions, 115 key events). That is a small sample from one 30-day window on one site.
- Ask ChatGPT the questions your borrowers ask, with your city and each loan type, and keep a dated log of who gets cited. Compare runs over time instead of trusting one answer.
What did we change on leadpops.com, and what happened?
We made ordinary search changes on leadpops.com in 2026, and in our weekly snapshots of September 21 and October 12, 2026, it was the second most-cited domain in Google's AI Overviews across the mortgage-marketing keywords we track, behind only YouTube. The changes were: our guides made readable to AI crawlers in July 2026; how-to answers added to pages that already ranked, under question-shaped headings, with every number sourced; and internal links added to under-linked pages. We have not measured which of these changes, if any, the citations came from. For the ranking work underneath these changes, see our guide to mortgage SEO for loan officers.
- youtube.com1,025
- leadpops.com279
- housingwire.com200
- kaleidico.com136
- reddit.com123
- setshape.com109
- agedleadstore.com96
Citations in Google AI Overviews across the mortgage-marketing keywords we track, one weekly snapshot, October 12, 2026: 758 keywords captured that week, 735 of them with an AI Overview; the set of keywords captured changes from week to week.
In the October 12 snapshot, leadpops.com had 279 citations across 184 keywords, and across the four weekly snapshots from September 21 to October 12 the AI Overviews we tracked cited it 1,149 times. The share of tracked AI Overviews that cited us fell from 32.8% to 25.0% over those four weeks. The keyword set we capture also varies from week to week: in the latest snapshot it was 758 keywords, 735 of them with an AI Overview.
Other AI answers cite us too: in a weekly sample of our tracked questions, in the week of October 12, Gemini cited us in 33 of 147 answers (22%) and ChatGPT in 17 of 137 (12%). Answers vary from run to run, so read these as weekly samples. All of these are marketing questions asked by loan officers and lenders, a different audience from borrowers.
What happened on "mortgage leads cost"?
The AI Overview for "mortgage leads cost" began citing leadpops.com on September 28, 2026, after we added a cost-by-type section, and our clicks for the query stayed at zero. On September 18 and 21, we ranked second and then first for it, and the AI Overview above our listing did not cite us. On September 21, two of the five pages the AI Overview cited were cost guides priced by lead type ("Mortgage Leads Cost: Aged vs Fresh vs Exclusive Pricing Guide (2026)" and "Mortgage Lead Cost in 2026: Real Prices by Type"), and our page had no section breaking cost down by lead type.
- Sep 18Ranked #2Not cited
- Sep 21Ranked #1Not cited
- Sep 22Section addedHow Much Do Mortgage Leads Cost in 2026?
- Sep 28Cited1 of 5 sources
- Oct 5Cited3 of 9 sources
- Oct 12Cited3 of 9 sources
- Impressions in 90 days
- 7,159
- Average position
- 4.5
- Clicks
- 0
Impressions, position and clicks: Search Console, 90 days to October 10, 2026. AI Overviews change from week to week, so this is what we observed, not proof that the section alone caused the citation.
On September 22 we added a section titled "How Much Do Mortgage Leads Cost in 2026?" that answers in its first sentence. On September 28 the AI Overview cited leadpops.com as one of its 5 sources, and on October 5 and October 12 it cited us 3 times among 9 sources. In the 90 days to October 10, 2026, the query showed our page 7,159 times at an average position of 4.5 and earned no clicks. AI Overviews change from week to week, so this is what we observed, not proof that the section alone caused the citation.
What happened on "mortgage lead generation companies"?
Four of our pages were cited for "mortgage lead generation companies" in late September 2026, and by October none were. The AI Overview cited 9 sources on September 21, and 4 of them were leadpops.com pages; on September 28 it was 3 of 9. On October 5 and October 12 the same AI Overview cited none of our pages. Citations move week to week, and one good snapshot is not a position you own. Merging those pages into one in September would have removed citations. Before you consolidate similar pages, check which of them the AI answer cites over several weeks, not once.
- Sep 21Cited4 of 9 sources were ours
- Sep 28Cited3 of 9 sources were ours
- Oct 5Not citedNone of our pages
- Oct 12Not citedNone of our pages
One good snapshot is not a position you own. Check which pages an AI answer cites over several weeks before you merge or remove any of them.
Does this guide follow its own advice?
Yes, in six ways you can check.
- Three of its chapter headings are questions people typed into Google, changed only in capitalization, and its opening answers a fourth. From July 14 to October 12, 2026, each showed leadpops.com and earned 0 clicks: "how can loan officers get found in ai search?" (average position 4.0), "how can a loan officer rank in ai search without controlling the company website?" (10.2), "how should loan officers set up schema and website for ai search?" (9.2) and "how do we optimize loan officer and branch pages for generative ai search?" (1.0). They were four of 49 AI-related mortgage questions that showed our site in that window.
- Every question heading is answered in its first sentence, and each section names the figure it relies on so it can stand alone.
- Every figure carries a source: our own data, our estimates from building loan officer websites, or a public study linked where we use it. Each study and data pull is dated where it first comes up, and every date is listed in "How we measured this".
- It is a guide on leadpops.com, and since July 2026 our guides are sent from the server with their headings and tables in place, so AI crawlers that do not run JavaScript can read it.
- It names what has no good evidence behind it, including llms.txt files and schema markup, which we publish anyway.
- Its charts are built from page text, not images, so every figure in them can be read and quoted.
Whether AI cites this page is something we will measure, not assume.
Questions loan officers ask about AI visibility
Should I block AI crawlers from my site?
Not if you want to be cited in ChatGPT search. Keep OAI-SearchBot allowed; OpenAI documents it separately from GPTBot, its training crawler. We allow both on leadpops.com, along with the other major AI crawlers.
Can I use AI to write these pages?
Yes, as long as the facts and the experience are yours. Google's spam policies treat pages mass-produced with AI and no added value as scaled content abuse.
Is it worth writing articles that explain loan programs?
Yes for your own borrowers, if the articles help them, but our data suggests AI will rarely cite them. Government and government-sponsored sources took 12 of 17 citation slots on the ten learning questions in our ChatGPT study, and Google cited us on none of the 10 definition questions in our tracked set.
Does my own website matter at all?
Yes, even though it is not where most citations went: in our ChatGPT study, directory and review sites took 40% of the citation slots on city questions and actual lenders' websites took 17%. A listing is rented and your website is the only page AI can cite that you own, so link every profile to it, build a page for one city and one loan type that answers the borrower's question first, and make the site turn visitors into leads. By our estimates from building loan officer websites, a well-optimized site turns 5 to 15 in 100 visitors into leads, against fewer than 2 on most.
How we measured this
The AI Overview share figure (809 of 845 keywords) and the citation rates by question type come from one snapshot of the mortgage-marketing keywords we track, taken in September 2026. The citation counts come from four weekly snapshots of the same tracked keywords, September 21 to October 12, 2026; the set captured varies from week to week (758 keywords on October 12).
The ChatGPT study ran 498 prompts through the live ChatGPT web interface on September 24, 2026: 194 cities across all 50 states, a 50-state "near me" test and a control group. Each prompt ran once, and ChatGPT varies between runs. The cities were the three to five largest by population in each state, which under-weights rural markets, and the study covered the US only. The near-me prompts were sent with a state set as the location. We analyzed the websites each answer cited, not the answer text. On October 14, 2026 we classified the lender, ranking and some directory domains by hand; the other labels come from the original rules that match domain names. Each website counts once per answer, and some sites could be argued either way, saferate.com most of all. The pilot, run the same day, was 50 borrower questions across eight categories, one run each. Quoted prompts are copied from the study's prompt file, and the pilot's searches from its results file.
The website figures (fewer than 2 in 100 visitors becoming leads on most loan officer websites, 5 to 15 on a well-optimized one, and 20% to 30% of website leads closing) are LeadPops's estimates as of October 2026, from building and running loan officer websites, not a measured study; the per-1,000-visitor figures are arithmetic on those estimates.
Bing figures come from an AI Performance report export covering the three months to October 14, 2026; the report is in beta. GA4 figures cover September 13 to October 12, 2026. Search Console figures are our site's own, from 90-day windows pulled on October 14, 2026: July 12 to October 10 for the long-question figures, and July 14 to October 12 for the questions used as this guide's headings.