AI does the research. I make the calls.
This is the full explanation of how artificial intelligence is used in my practice: what it does, what it is never allowed to do, and why you end up with better evidence and more of my time. No vendor pitch. Just the rules I work by, written down so you can hold me to them.
It reads, drafts and counts. I verify, decide and sign.
Every pricing analysis and every marketing step in my practice uses AI the same way. The software reads ten years of sales, drafts the first version of the copy and counts what needs counting. A licensed REALTOR® checks the work, makes the decision and puts his name on it. The judgment stays human. The speed is new.
- Reads the data and the documents
- Drafts the first version
- Counts, sorts and charts
- Re-runs the analysis for a second opinion
- Verify every figure against the source
- Choose the comps and the strategy
- Approve every word that goes out
- Show, negotiate, answer the phone
I did not add these tools to look modern. I added them because the research that makes a pricing analysis worth reading takes more hours than most agents give one address. What AI cannot do is tell you what your home is worth, or decide what goes in front of a buyer. That part stays mine.
Six rules. No exceptions.
The same six rules appear on the Sell page. Here is what each one means in practice, with an example.
- 01
Nothing posts, sends or spends without approval
AI prepares. A person approves. Every ad budget, mailing list, social post and report leaves my account after I have read it, not before. The tools can draft a week of content in an afternoon; none of it is live until I have pressed the button myself.
In practice. Say a listing carousel calls the kitchen "newly renovated" when the record says it was done years ago. I change it to "updated", approve the set, and it posts from my account. The ad behind it arrives with a proposed audience and budget; I set the dollar amount. The mail list is a radius pull; I read the addresses first.
- 02
Every number is checked against the source
Prices, square footage, sale dates and financing terms come from Bright MLS and the county record, which in Maryland means the State Department of Assessments and Taxation. A model can summarize a sale beautifully and still get the number wrong. If a figure cannot be traced to a source, it does not go in. "Looks right" is not a source.
In practice. Say the MLS counts a comp's finished area including the basement and the tax record counts above grade only. Both are real, and the difference is the whole story of that comp. The report shows both, labeled, and adjusts on the right one. That is why the fair market lines in my Seller Pricing Analysis are measured two ways.
- 03
Three models, one answer
Before a pricing analysis or any client deliverable reaches you, independent AI models re-run the work. They are built by different companies and share no memory, so they make different mistakes. Where they disagree, I dig in. Where they agree, you get a cleaner answer faster.
In practice. Say one model weights a finished basement more heavily than the other two. That sends me back to the sold comps to see what finished basements actually fetched on that street, and the report shows the answer and the reasoning. Without the disagreement, I might not have looked twice.
- 04
Honest images, always
Virtual staging helps an empty room and is the easiest place for a listing to mislead. So every virtually staged photo is labeled, and the original stays in the set so a buyer sees the room as it is. Permanent features and defects are never edited out: not a power line, not a crack, not the neighbor's shed. Inspiration is fair. Deception is not, and it tends to come back at the inspection or the appraisal.
In practice. Say an empty family room is restaged in two styles for an inspiration post. Both carry a "virtually staged" label. The empty-room photo stays in the MLS set. The radiator under the window stays in all three.
- 05
Fair housing is built into the system
AI never decides who sees a home based on a protected characteristic, and neither do I. Audiences for ads and mailings are chosen by location, price band and interest: people looking at homes in a ZIP code, in a price range, with a garage or a flat yard. The buyer personas in the Listing Distribution System describe how someone would use the house, never who they are. Copy describes the home, never the buyer.
In practice. A draft that says "perfect for a growing family" becomes "four bedrooms on one level and a fenced, flat yard". The second version is more useful to every reader, and it is the only one that goes out.
- 06
You get the time back
The point is not the software. It is the hours. Pulling sales, charting demand and adjusting comps line by line now happens in the background, and the time goes where it is worth the most: showings, negotiation, and picking up when you call.
In practice. Say you text me on a Tuesday night about the offer that just came in. I can walk you through it, because I am not up late formatting a chart. It was done that afternoon, and I had already read it twice.
Where the machine works, and where I do.
Step by step, who does what. Read the right-hand column closely. It is the one you are hiring.
| Step | AI does | Michael does |
|---|---|---|
| Pricing | Pulls ten years of sales and demand data, builds the comp grid and the charts. | Verify every figure against Bright MLS and county records, choose the comps, set the price strategy. |
| Validation | Independent models re-run the analysis from the same files. | Resolve the disagreements, then re-check the whole report twice by hand. |
| Marketing | Drafts the copy system, the carousel text, six video scripts and the mail cards. | Approve every piece, record the video, decide what runs where. |
| Images | Restages a room in a new style for inspiration posts. | Label every one "virtually staged", keep the original, never hide a defect. |
| Replies | An assistant texts new inquiries back within minutes, around the clock. | Take over the conversation the moment I reply. |
| Buyer search | Screens every listing in an area against your criteria and reads the photos for condition. | Tour, advise, write the offer. |
| Inspections | Reorganizes a 60-page report into priorities, with public price ranges labeled as Google results, not estimates. | Decide the ask with you, then negotiate it. |
| Deadlines | Builds the transaction calendar from the contract dates. | Review every date before an invite is sent. |
Who confirms what. A pricing analysis is not an appraisal. A price range on an inspection item is a public Google result, not my estimate; a contractor quotes the real number. Lending questions go to your lender, contract questions to your attorney or title company, tax questions to your CPA.
What AI never does here.
Five things that are off the table, in any tool, on any day. And the one thing it is for.
Decides who sees a home
Audiences are built on location, price band and interest. No tool I use selects or excludes people by a protected characteristic, and no draft that implies it survives my read.
Edits out a defect or adds a feature
Nothing permanent leaves a photo, and nothing that is not there goes in. Staging is labeled. The original stays in the set.
Sends anything unread
No message, offer, report or ad goes out without a human reading it first. That human is me.
Gives legal, lending or tax advice as fact
It can organize the question. The answer comes from your attorney, lender or CPA, and I will tell you which one to call.
Replaces a showing, a negotiation or a phone call
The part of this job that happens in a kitchen, across a table or on the phone at night is the part you hired. It is not delegated.
Reads, drafts, counts, re-checks
That is the whole job description: a research assistant with a very fast reading speed, treated like one.
How a report earns the right to reach you.
Every pricing analysis, buyer analysis and market comparison goes through this sequence before delivery. Five passes. The first and the last are mine.
- 01
I build it first
I pull the Bright MLS exports and county records, run the numbers and build the analysis myself. The models never start from a blank page; they start from my files.
- 02
A first model re-runs it
An independent AI model runs the same analysis from the same files without seeing my conclusions. If its grid matches mine, good. If not, we have found something.
- 03
Two more run comparisons
A second and a third model run comparison analyses. Three passes, three sets of assumptions, one set of source data.
- 04
Disagreements go to the source
Where the passes disagree, I trace the line to the MLS or tax record and settle it there. Not by majority vote. By the document.
- 05
I re-check it twice
Then I read the entire output two more times by hand, line by line, before it goes to you.
Why consensus, and a human at both ends
Independent models make different mistakes. When three land on the same number from the same files, a systematic error is much less likely to be hiding in it; when they do not, the disagreement points at the line that needs a human. A human check at the start and at the end keeps ground truth where it belongs.
Why auditability over polish
The output is built to be checked, not admired: tight numbers, a clear source for each, flagged assumptions, and a layout you can spot-check line by line.
Better evidence, more of my time.
The rules are the same whether you are selling or buying. What changes is where the hours land.
You see the numbers before we pick a price
Your Seller Pricing Analysis arrives with ten years of local sales, buyer demand by ZIP code and comps within a mile adjusted line by line, every figure verified against Bright MLS and county records and cross-checked by three independent models. Every ad, post and mail card is read by me before it runs. Staged images are labeled. The weekly Seller Report is read before it is sent. The hours the research used to take go into the pricing conversation, the showings and the negotiation.
Every listing screened, every tour in person
Every active listing in your area is screened against your criteria, and the photos are read for condition flags so the homes worth a drive rise to the top. I tour with you, advise, and write the offer. The inspection report is reorganized into priorities, with any price ranges labeled as public Google results rather than estimates, so we decide the ask together. The transaction calendar is built from the contract, and I review every date before an invite reaches you.
Your file is your file.
I keep these promises to what I control. There are three.
- Your documents are used for your transaction and nothing else. Your contract, inspection report, disclosures and numbers are used to do your deal. They are not reused as examples, posted, or folded into marketing for anyone else.
- Nothing about your home is posted after the Stop Rule. The Stop Rule from the Listing Distribution System applies to every tool I use. Once it is in effect, nothing about your home goes out, from me or from anything running on my behalf.
- I read everything that goes out. Every message, report, post and offer has been read by me first. That is rule one, and it is also a privacy rule.
I use paid, business-grade tools rather than free consumer apps, and I keep the number of places a client file lives as small as the job allows. I will not make claims about what any software company does behind the scenes, because I cannot control it. I can control what I upload, keep, delete and send.
AI can be confidently wrong.
This is the part most explanations leave out. It is the reason rule two exists.
An AI model does not know when it is wrong. It writes a fluent paragraph around a bad number with the same confidence it uses for a good one. That is how the tool works, and it is why every figure has to trace to a source document before it goes in.
In my experience the two most common errors are superlatives and counts written by eye. A draft will call a lot "the largest on the street" because it sounds right, or say "seven of the ten comps sold above list" because seven looked about right. So counts in my reports are computed from the data table, not eyeballed, and a superlative has to be provable from the record or it gets cut.
What this means for you. If a number in one of my reports looks wrong, say so. It will either trace to a source in under a minute or it will be fixed. Both are the system working.
Questions I get asked.
Will I be talking to a bot?
Sometimes, for a few minutes, at the very start. If you send a new inquiry late at night, an AI assistant may text back within minutes to answer the basics and let me know you reached out. The moment I reply, I take over and it stops. Once you are my client, you are texting me.
Do you use AI to write the price?
No. AI pulls the sales, builds the comp grid and charts, and re-runs the analysis for a second and third opinion. I verify the figures, choose the comps and set the strategy, and we pick the price together after you have seen the evidence. It is not an appraisal; if a lender needs one, a licensed appraiser does it.
Is my information safe?
I keep that answer to what I control. Your documents are used for your transaction and nothing else. I use paid, business-grade tools, I read everything before it goes out, and nothing about your home is posted after the Stop Rule. I do not make promises about any software company's internal practices, because I cannot enforce them.
Does AI write the listing description?
It drafts it. From the photos, the disclosures and the record, it produces a first version and a few alternatives. I rewrite, cut anything I cannot verify, make sure the copy describes the home and not the buyer, and approve the final. Every description under my name in Bright MLS has been read and edited by me.
Why not just do it all by hand?
Because by hand, the research hours came out of showings and phone calls. AI makes the research affordable. It does not make the judgment cheaper, so the judgment is still done by hand.
What if the AI gets something wrong?
It does, sometimes, and the system assumes it. Three independent models cross-check one another, every figure is traced to the MLS or the county record, and I re-read the whole output twice before it reaches you. If something slips through, tell me. It gets traced and fixed, and the check that missed it gets tightened.
See the evidence before you decide.
Start with your home's value, or book fifteen minutes and ask me anything on this page.
Equal Housing Opportunity. Michael Wade, REALTOR®, Samson Properties, licensed in Maryland and Washington, DC. 443-812-9980. MD license #664139 · DC license #SP40000356. Pricing analyses and market comparisons are not an appraisal. Nothing on this page is legal, lending or tax advice; confirm those questions with your attorney, lender or CPA. Any cost figures referenced in inspection summaries are public general ranges, not estimates.