AI Visibility for Mortgage Brokers: Rates, Licenses and What AI Gets Wrong

Mortgage buyers use AI assistants for three jobs: comparing rates, understanding loan types, and finding someone licensed where they live. Their prompts mix all three. Here are realistic examples a borrower might type:

• "What are 30-year fixed mortgage rates right now?"
• "Best mortgage broker for first-time buyers in Phoenix"
• "Is a mortgage broker cheaper than going straight to a bank?"
• "Which brokers near me do VA loans with no down payment?"
• "Is [broker name] licensed in North Carolina?"
• "Can a self-employed borrower get a mortgage with bank statements instead of tax returns?"
• "What fees does [broker name] charge?"

The rate and fee prompts carry the most risk. They ask for a number that changes daily, and the assistant may answer from memory. The licensing prompts carry regulatory weight: a wrong state list sends a borrower to someone who cannot legally help them. Write down the ten prompts your own borrowers would ask, because those are the ones worth checking first.

An AI answer comes from one of two places. Either the model searches the web at answer time, or it answers from what it learned in training. Training knowledge stops at a cutoff date. OpenAI's model documentation, as checked in August 2026, lists a February 16, 2026 knowledge cutoff for the GPT-5.6 family and August 31, 2025 for the Instant model behind chat-latest.

So a retrieval-off answer about "today's rates" reflects a rate environment from months earlier. You cannot change that frozen knowledge this quarter. Nobody can.

Whether the model searches is not up to you either. ChatGPT search is model-discretionary, Google AI Overviews often do not trigger, and Claude web search is a user toggle. Read how often ChatGPT searches the web for why published estimates disagree.

What you control is the retrieval side: when an engine does search, your page should be reachable and should state the current fact clearly, with a date next to it.

Borrowers spread across several assistants, and each one draws on a different index:

• Google AI Overviews and AI Mode use Google's index. A page must be indexed and eligible to show with a snippet to appear as a supporting link, per Google's AI features documentation.
• Gemini is a separate Google product, governed in part by the Google-Extended control.
• ChatGPT uses OAI-SearchBot for search inclusion.
• Microsoft Copilot grounds on Bing's index.
• Perplexity and Claude run their own crawlers.

These engines do not agree. In OtterlyAI's 2026 experiment, Copilot cited two test sites thousands of times after Google removed them, while ChatGPT, Gemini and Google's AI features almost never did. That was one pair of sites on a low-stakes topic, but it shows why a single blended number hides problems. Check engines one at a time. The same logic applies to banks and credit unions, where rate prompts behave the same way.

AI platforms get specific facts wrong about specific brands. For a mortgage broker, these are the facts worth checking first:

• Rates. A quoted rate with no date, or a rate presented as fixed when it was a daily figure.
• Licensed states. An old state list, or a state you never held a license in.
• NMLS ID. Missing, or attached to the wrong entity.
• Broker or lender. Assistants can describe a broker as a direct lender, or the reverse.
• Loan types. FHA, VA, USDA, jumbo, non-QM and bank-statement loans you do or do not offer.
• Fees and compensation. Old origination fees or broker fee structures.
• Branches and ownership. Closed offices, old addresses, or a merger the assistant has not caught up with.

If an assistant already has something wrong, our guide to fixing wrong information in ChatGPT covers what to change and where. For context on scale, see how often AI search gets facts wrong.

When engines cite other sites instead of yours, being named correctly on those sites is the route into the answer. The evidence shows a strong association here, not a guaranteed effect. For a mortgage broker, audit these:

• NMLS Consumer Access. Your license record and states should match your website exactly.
• State regulator license lookups. Same check, state by state.
• Google Business Profile and Bing Places. Addresses, hours and service areas.
• Review platforms your borrowers actually use.
• Lender and wholesale partner directories that list you as a broker.
• Local news, chamber and real estate partner pages that mention you.

One inconsistent source is enough to produce a confused answer. Fix the third-party record at the source, then make your own site the clearest statement of the same facts. Insurance agents face a similar licensing problem, covered in AI visibility for insurance agents.

Stating a specific fact on a page that can be fetched gives an engine a sentence it can quote. That is the one on-page approach with positive evidence. Write facts as plain sentences:

• "NMLS #000000. Licensed in Texas, Oklahoma and Louisiana." (Hypothetical values.)
• "We are a mortgage broker. We arrange loans through multiple wholesale lenders."
• "We offer FHA, VA, conventional and jumbo loans. We do not offer reverse mortgages."

Put these in the server-rendered HTML. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but do not execute them. A license list that only loads after a script runs may not exist for those crawlers. Googlebot renders JavaScript, so this matters for the other engines.

Rates need different handling. Never present a daily rate as fixed. Put a date and time beside it, and state that rates change and depend on the borrower. Should you publish rates at all? See publishing prices for AI answers for the trade-off.

If you do not want a daily table summarized, Google and Bing both document data-nosnippet. Bing says marked content is still indexed and available for ranking but excluded from snippets and AI summaries:

<section data-nosnippet>
  <h2>Today's rates (updated 9:00 a.m.)</h2>
</section>

The cost is real: the engine cannot quote that rate. Keep your static facts outside the marked block. Check rate disclosures with your compliance team.

Access is the strongest lever you have, and it works by removing blocks. Check these:

1. robots.txt per bot. OpenAI says each setting is independent. Blocking GPTBot stops training use. Blocking OAI-SearchBot removes you from ChatGPT search answers. Many sites block both by accident.
2. Retrieval bots. Confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot and bingbot are allowed. There is no robots.txt token for AI Overviews; Googlebot rules are the control.
3. Firewall and CDN. A firewall can return 403 to a crawler that robots.txt allows. OpenAI's help centre says eligibility needs both robots.txt permission and a host or CDN that allows its published IP ranges. Cloudflare began blocking training and agent crawlers by default on ad-displaying pages for some customers from September 15, 2026.
4. Snippet directives. Look for stray noindex, nosnippet or max-snippet:0 inherited from a template.
5. Rendering. Compare the raw HTML with the rendered page for your license and loan pages.

The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Week 1: baseline

Write your ten borrower prompts. Run them on each assistant, logged out, and record what each says about your rates, states, NMLS ID and loan types. The free checker on this site runs up to three buyer questions as a starting snapshot.

Week 2: access

Audit robots.txt per bot, test firewall responses for each AI crawler, and remove accidental snippet directives. Confirm your key pages appear in Google Search Console and Bing Webmaster Tools separately.

Week 3: facts

Rewrite your licensing, lending area and loan type pages as plain, server-rendered sentences. Date-stamp every rate. Add data-nosnippet only where a daily rate should not be summarized.

Week 4: third parties and recheck

Correct NMLS, state, directory and profile records. Rerun your prompts and log changes engine by engine. Expect some answers to stay stale, because retrieval-off answers will not change until a future model.