Data Sources and Methodology

LenDb is a free, lender-neutral directory of US mortgage loan officers. Every fact on LenDb carries where it came from and when it was last verified, so people and AI assistants can check it and cite it. This page explains where each kind of information comes from, how we check it, how fresh it is, and how ranking works.

1. Where each field comes from

Field	Source	Notes
Name, NMLS ID	State regulator public records; the loan officer's page at their lender	The NMLS ID is the key that ties records together.
Licensed states, license numbers, status, dates	State regulator public records (for example the Florida Office of Financial Regulation's published loan officer data) and other public records	Always verify at NMLS Consumer Access.
Current employer and branch	The loan officer's page at their lender; state records
Employment history	Changes we observe over time across sources
Title, languages, loan programs, property types, borrower situations	The lender's public page about the loan officer; voice or video intro transcripts the lender publishes; the loan officer's own claimed profile	Each specialty shows its evidence level.
Bio	A short summary of the lender's published bio, or the loan officer's own text if they claimed the profile	We summarize rather than copy.
Headshot	Displayed from the lender's own website	Removed on request.
Voice or video intro	A one-sentence summary in our words and a link to the lender page where it plays. We do not embed the recording or republish its transcript
Office phone, business email	The lender's public page about the loan officer	Shown on the website only after a human check. Never in the API or MCP server.
Regulatory actions	Public orders from state regulators and federal agencies	Shown with a link to the source document.
Lender details	The lender's own website and public regulatory records

What we do not use. We do not scrape NMLS Consumer Access, whose terms forbid it; we link to it so you can verify licenses yourself. We do not scrape LinkedIn (we store a LinkedIn link only if a lender or the loan officer publishes it). We do not log in to any site, create accounts on lender sites, or bypass blocks.

2. How we check facts

• Each lender page is read by our crawler, LenDbBot, and an AI model (currently Claude, from Anthropic) extracts the facts into a strict structure.
• Automatic checks: the NMLS ID must appear on the page, the name must match, and phone and email formats must be valid. Anything that fails goes to a review queue instead of the site.
• A stronger AI model spot-checks a sample of each lender's results when we first add the lender and whenever the lender's page layout changes.
• Specialties are tagged by keyword rules first, then by AI classification against our fixed taxonomy, with a short evidence snippet and a confidence score.
• Lender-level programs ("this lender offers non-QM") are kept separate from loan-officer-level focus ("this person focuses on non-QM").
• Loan officers can claim their profile and correct or add to it. Their additions are labeled self-reported.

3. Freshness

• Every fact shows a last verified date.
• Lender directory pages are re-crawled about weekly, and individual loan officer pages about monthly, or sooner when they change.
• State roster files are re-loaded when the regulator publishes a new one (Florida publishes monthly).
• A loan officer who disappears from all sources for 90 days is marked status unknown, not deleted.

4. Our crawler: LenDbBot

LenDbBot identifies itself as LenDbBot/1.0 (+https://lendb.ai/bot).

• Obeys robots.txt. LenDbBot is a directory and search crawler, not an AI training crawler. It honors rules aimed at LenDbBot, at all bots, and at AI search bots (such as ClaudeBot, PerplexityBot or OAI-SearchBot). Rules aimed only at AI training crawlers (such as GPTBot, CCBot or Google-Extended) do not apply to it, and we say so openly.
• Honors Content-Signal. If a site's robots.txt says Content-Signal: ai-input=no, we do not crawl it, and we check for this on every lender.
• Polite. One request at a time per site, with a delay between requests, backing off on errors and Retry-After. It caches with ETags and conditional requests.
• Public pages only. No logins, no forms.
• Opting out. To block LenDbBot, add User-agent: LenDbBot with Disallow: / to your robots.txt, or email bot@lendb.ai. If a lender opts out, its loan officers can still appear through public state records or by claiming their own profile, and we are happy to take a direct feed instead.

More detail for site owners and AI agents is on /bot.

5. How LenDb can be used by AI

LenDb is built to be read by machines. Our robots.txt welcomes search engines and AI crawlers and states Content-Signal: search=yes, ai-input=yes, ai-train=yes. Phone numbers and email addresses are kept out of the API, MCP server and machine-readable twins, and are only shown on the website after a human check.

6. How ranking works

For a situation (a specialty, a place, and optionally a language), LenDb scores each active loan officer licensed in the property's state:

1. Specialty evidence carries the most weight: production data, then the lender's own page, then self-reported claims, each scaled by confidence.
2. A license in the property's state is required.
3. Branch location near the place (a soft factor, since many loan officers work remotely).
4. Experience: years licensed.
5. Profile completeness and freshness.
6. Clean record: public regulatory actions lower the score and are always shown.

Every ranked list explains in plain words why each person is on it.

Nobody pays for rank. No lender or loan officer can pay to appear, to rank higher, or to have negative information removed. Claiming a profile is free and does not change the ranking rules; it only lets the loan officer add self-reported information, which is weighted as self-reported. If we ever offer paid features, they will be clearly labeled, will not affect ranking order, and will be described on this page and on our Disclosures page first.

Ranking never uses protected characteristics. See our Fair lending statement.

7. Mistakes

If something is wrong, tell us. See Corrections, removal and claims. We aim to resolve requests within 7 days.