Skip to content

Lesson 03 · 14 min read

AI for Prospecting and Owner Research

Use AI to clean, standardize, dedupe and classify public-record owner lists, plan Sunbiz research, and draft outreach from facts you supply, without inventing facts or skipping the calling and advertising rules.

Prospecting is mostly list work: pulling parcels, fixing owner names, merging duplicates, finding who is behind an LLC, then writing to real people about real buildings. AI is good at the messy middle. It is bad at knowing facts about a specific owner, and it changes no rule about how you may contact them.

This lesson shows where AI fits in the owner-list workflow from Building the Owner List from Public Records: cleaning and deduping, classifying owners, planning entity research, and drafting letters and call notes. It closes with the compliance points that still apply, linked rather than repeated, and a worked example on a Seminole County flex list.

Education, not legal advice. Calling, texting, email and advertising rules change and their reach to commercial outreach is not always settled. Confirm compliance points with your broker of record and a Florida attorney before you run a campaign.

Start from public records, not from the model

The list starts with data you pulled. Asking an assistant "Who owns industrial buildings in Sanford?" produces names that may be real, outdated or invented, and you cannot tell which.

Your raw material is the same as in Course 23:

  • County property appraiser exports or advanced search results for your farm
  • Florida DOR land-use codes to filter by property type, such as 048 for warehousing and distribution or 011 for one-story stores
  • Florida DOR NAL and SDF files for owner name, mailing address, building size and sale history
  • Clerk official records for deeds and mortgages on your top-tier owners
  • Sunbiz for the people behind entity owners

AI enters after you have a file. Its job is to clean the file and speed the research, not to supply facts.

Cleaning and standardizing the list

Public-record exports are messy. The same owner appears as "BAY HOLDINGS LLC," "BAY HOLDINGS, L.L.C." and "BAY HOLDINGS LLC ATTN ACCOUNTING." Mailing addresses mix "STE," "SUITE" and "#." This is exactly the kind of pattern work a model or a spreadsheet copilot handles well.

Cleanup taskWhat AI can doWhat you check
Normalize entity namesUppercase, strip punctuation, standardize LLC, INC, LP, TRA sample of 20 rows against the original
Split care-of linesMove "C/O" and "ATTN" text into a separate columnThat no owner name was cut off
Standardize addressesConsistent suite, street type and ZIP formatThat ZIPs did not change
Flag missing dataMark blank owner, blank mailing address, PO boxesBlanks may be confidential-exempt owners, not errors
Build an owner keyCombine normalized name, mailing line one and ZIPMerged groups make sense

Two ways to do it:

  1. Spreadsheet copilot or formulas. Ask the AI to write cleanup formulas, then run them in your own file. The data stays put, and the formulas are visible and testable.
  2. Paste into a chat tool. Faster for small lists, but check that no rows were dropped or changed, and strip any skip-traced phones or notes first.

For anything over a few dozen rows, use formulas. A model rewriting a 400-row table in a chat window can silently drop or alter rows.

I have a Google Sheet. Column A = owner name, B = mailing address line 1,
C = mailing city, D = state, E = ZIP.

Write formulas for new columns:
F = owner name uppercase, punctuation removed, and these replacements:
"L.L.C." and "L L C" -> "LLC", "INCORPORATED" -> "INC", "TRUSTEE" and "TRS" -> "TR".
G = mailing line 1 uppercase with "SUITE" and "#" both -> "STE".
H = owner key: F, G and the first 5 digits of E joined with "|".

Then give me three test rows with messy input and the expected output so I can check.

Deduplicating into owner keys

One owner with five buildings is one conversation. Course 23 uses an owner key built from name, mailing address and ZIP. AI can build the key and the grouping, but the merge decisions are yours.

  • Ask AI to group, not to delete. Have it add a group number column so every original row survives.
  • Review groups with different names at one address. A law firm or accounting firm mailing address may serve dozens of unrelated owners. That is a shared advisor, not a portfolio.
  • Review same names at different addresses. "SMITH JOHN" in Orlando and "SMITH JOHN" in Ohio are probably different people.
  • Count and sum. Once grouped, have formulas count parcels and sum building SF per owner key, then check one group by hand.

Classifying owners

Owner type changes how you research and how you write. A model can classify thousands of names quickly. It will also get a predictable set of them wrong.

Owner typeCommon signals in the nameWhere AI tends to miss
IndividualLast name, first name, middle initialConfuses "SMITH JOHN TR" (a trustee) with an individual
LLCLLC, L.L.C., LCMisses older names without a suffix
CorporationINC, CORP, COTreats "CO" in a person's name as company
Limited partnershipLP, LTD, LLLPLabels as LLC
TrustTR, TRUST, TRUSTEE, REV TRMisses abbreviations like "TRS" or "TTEE"
EstateEST, ESTATE OF, PRMisses personal representative wording

Ask the model to add confidence and reason columns. Review every low-confidence row and a random sample of the rest. A misclassified estate gets the wrong letter.

Planning entity research on Sunbiz

Sunbiz unmasking is where most of your research time goes. AI can help you plan it. It should not be your source for what Sunbiz says.

What AI does well here:

  • Build the research queue. Sort entity owners by tier, then list exact names to search.
  • Write a research checklist for each entity: status, filing date, principal address, mailing address, registered agent, managers or officers, last annual report date.

What you do by hand:

  • Run the searches yourself on the Division of Corporations site and copy results into your sheet with the date you checked.
  • Confirm the exact entity. Similar names are common. "Lakeside Flex LLC" and "Lakeside Flex Holdings LLC" are different companies.
  • Read the annual report, not a summary of it.

A general assistant cannot see Sunbiz without a browsing feature, and even then it can open the wrong entity, read an old annual report, or blend two entities. Treat every name a browsing tool returns as a lead to confirm, and log the entity document number so you know you matched the right record.

Drafting letters and call notes from facts you supply

Once you know who you are writing to, AI is a good drafting partner. The rule is simple: the model writes from a fact sheet you built, and nothing else.

A fact sheet for one owner might hold:

  • Owner of record and entity type
  • Decision-maker name and title from the annual report
  • Property address, building SF, year built, DOR use
  • Acquisition year from the appraiser record
  • Mailing address location, such as out of state
  • One market fact you can source, with the source named
  • Your reason for writing, in one sentence
ROLE: Act as a commercial real estate agent writing a first letter to a property owner.

FACTS (use only these):
[paste fact sheet]

TASK: Draft a one-page letter.

RULES:
- Use only the facts above. Do not add facts about the owner, the building,
  their plans, their health, age, finances, or family.
- Do not claim we have a buyer unless the facts say so.
- Do not state a value, cap rate, or price.
- Include a placeholder for my signature block with the licensed brokerage name.
- Plain language, under 250 words, one clear call to action.

After the letter, list every factual claim in it with the fact-sheet line it came from.

That last instruction is your audit trail. Any claim without a matching line gets cut.

Call notes work the same way. Ask for a 60-second pre-call brief built from the same fact sheet: who you are calling, the property, one question to open with, and what you will offer. Course 23 has the scripts in Cold Calling Commercial Owners.

The no-invention rule

Personalization is only good when it is true. AI is very good at writing sentences that sound personal. That is the risk.

Allowed in outreachNot allowed
"Public records show you have owned the building since 2004.""I know you have been thinking about selling."
"Your mailing address is in Georgia, so you may not see the property often.""Since you are getting close to retirement..."
"We are seeing buyer inquiries for small-bay flex in Seminole County." (if true and documented)"I have a buyer ready to pay top dollar." (unless true and you can show it)
"Your building is 14,400 SF on Example Road.""Your building's new roof adds value." (not in your facts)

Never guess age, health, family situation, financial distress or motivation. Even true inferences can feel invasive. A distressed owner is a reason to talk to your managing broker before any contact, as Course 23 explains.

Compliance still applies, and AI can make it easier to break

An AI-written letter is still an advertisement, an AI-personalized campaign is still commercial email, and skip-traced cell numbers may still be protected however the list was cleaned. Some AI tools make it easier to break these rules at scale.

The short list follows. The details live in other lessons, so read them rather than relying on this summary:

  • Calls and texts: the TCPA, the national Do Not Call registry, and Florida's Telephone Solicitation Act (F.S. 501.059). Hand-dial, scrub personal or unknown numbers, never use autodialers, prerecorded messages or bulk texting. See Course 23, Lesson 4.
  • Email: CAN-SPAM applies to business email. Honest subject lines, a postal address and a working opt-out. See Course 23, Lesson 5.
  • Advertising: FREC rules require your brokerage's licensed name on every ad, and no ad may be false or misleading. See Course 27, Lesson 5.

AI-specific traps to watch:

  • AI-generated or prerecorded voices on calls are a clear TCPA risk. Do not use them for outreach.
  • Automated outreach platforms may send texts or emails on a schedule with no human review. Know exactly what goes out, to whom, and under whose name.
  • Mass personalization can produce hundreds of letters with small factual errors. Review a sample of every batch, and every letter in a small one.
  • Stop requests reach your do-not-contact log the same day, whatever tool sent the message.

Respect website terms

Getting data the right way matters as much as using it the right way.

  • Use official downloads where offered. Many county property appraisers post data files or allow exports. Follow their stated terms.
  • Do not scrape sites that prohibit it. Listing marketplaces and paid data platforms commonly forbid automated collection and bulk export. An AI browsing agent collecting pages for you is still automated collection.
  • Respect licensed data agreements. Paid-platform data is usually licensed for your use, not for copying into another tool. Check whether your license allows pasting it into an AI service at all.
  • Use the bulk file, not thousands of searches, when a government site offers one.

If you are not sure a source allows a use, ask your broker of record or skip it.

A worked example: the Seminole flex list

Illustrative numbers and fictional owners. Not market data.

An agent farming small-bay flex in Seminole County pulls 40 parcels with DOR code 048 and 10,000 to 40,000 SF of building area from the property appraiser export.

Step 1: clean and key. The agent asks a spreadsheet copilot to write normalization and owner-key formulas, then tests them on five messy rows. The formulas group the 40 rows into 34 owner keys: one owner with three parcels, two owners with two parcels each, and two exact duplicate rows from the export. That is six extra rows (2 + 1 + 1 + 2), and 40 − 6 = 34.

Step 2: classify. The agent pastes only the 34 normalized names into a chat tool and asks for owner type, confidence and reason. The model returns 21 LLCs, 10 individuals, 2 trusts and 1 corporation. The agent reviews every low-confidence row and a sample of the rest and finds two errors:

  • "MARTIN R TR" was labeled an individual. It is a trustee, so it moves to trust.
  • "CYPRESS PARTNERS LP" was labeled an LLC. It is a limited partnership.
Owner typeAI countCorrected count
LLC2120
Limited partnership01
Individual109
Trust23
Corporation11
Total3434

Step 3: plan Sunbiz. The 22 entity owners (20 LLCs, 1 LP, 1 corporation) go into a research queue sorted by tier. The agent tries a browsing assistant on the first few and finds it matched "Lakeside Flex LLC" to "Lakeside Flex Holdings LLC," a different company with a different manager. From then on, the agent searches by hand and uses AI only to organize the results.

Step 4: draft. For a tier A owner, the fact sheet says: 14,400 SF warehouse-flex building, built 1998, acquired 2004 per the appraiser, mailing address in Georgia, manager name from the 2025 annual report. The AI draft is clean except for two sentences: "As you consider your next chapter after many years in business" and "Your recently updated roof will appeal to buyers." Neither is in the fact sheet. The audit list at the bottom of the draft shows no source line for either one, and both are cut.

Step 5: send compliantly. The letter goes out with the brokerage's licensed name in the signature block. The call follows by hand, on a number verified against the annual report, after the DNC scrub.

AI handled the cleanup, the first classification pass and the draft. Every fact that reached the owner came from a record the agent checked.

Common mistakes

  • Asking AI for owner names. Pull records instead.
  • Letting a chat tool rewrite a large table. Use formulas so rows cannot vanish.
  • Trusting a browsing tool's Sunbiz match. Similar names are different entities.
  • Personalizing with guesses. Age, health, motivation and finances are off limits.
  • Automating sends without review. You are responsible for every message.
  • Scraping sites that forbid it, or pasting licensed data into tools your license does not allow.

What to take away

  • Build the list from property appraiser and DOR data, never from a chat answer
  • Use AI or spreadsheet copilots to normalize names and addresses and build owner keys, then check a sample
  • Have AI group duplicates, not delete them, and review mixed groups by hand
  • AI classification is a first pass; trusts, estates and partnerships are where it errs
  • Use AI to plan and organize Sunbiz research, but run the searches and confirm the exact entity yourself
  • Draft letters and call notes only from a fact sheet, and ask the model to list the source of every claim
  • Never let AI invent or infer facts about an owner's age, health, finances or plans
  • TCPA, DNC, CAN-SPAM, Florida telemarketing and FREC advertising rules apply to AI-assisted outreach exactly as they apply to yours
  • Use official downloads and respect site terms and data licenses; no prohibited scraping

Next lesson: Summarizing Offering Memorandums and Market Reports.

Get Market Insights Delivered

Weekly Central Florida CRE updates — cap rates, new listings, market trends, and investment opportunities. No spam, unsubscribe anytime.