The listing says, "charming home in a quiet neighborhood, perfect for a starter." It doesn't mention that the lot is in a flood zone, that the tax after purchase will be recalculated, or that the community has been raising funds for a new roof for two years. This guide shows how to use AI to conduct a property review, akin to what a good inspector, accountant, and skeptical friend would do together: eight prompts in one conversation, each in a copyable box, with citation rules and a checklist of things to verify in the response. It works for a home in the USA and for an apartment in Poland purchased from abroad.
Last verification: September 6, 2026. Informational and educational material – not legal, tax, or investment advice. Regulations, tax rates, and insurance vary between states, counties, and municipalities; confirm them with a licensed specialist and in an official source for a specific address.
In Brief
- AI does not know your address. It knows general rules and what it can find online. Therefore, each prompt asks it to specify the source for a particular county, municipality, or parcel number, rather than "typically in the USA."
- Order: context, location, offer, numbers, documents, pre-mortem, negotiation, sheet. Numbers come before documents because inspection changes the numbers, and pre-mortem comes after both, as it has something to work with.
- The most valuable documents are those you paste yourself: inspection report, seller's disclosure, community documents, appraisal. Claude and ChatGPT accept PDF files (Anthropic documentation, OpenAI documentation).
- Three numbers that AI often gets wrong: property tax (it takes the old one, prior to recalculation), insurance (it guesses), and square footage (it mixes data from the listing with the registry). You check these yourself at the county office, with an insurance agent, and in the registry.
- Do not ask AI, "what kind of people live there." The Fair Housing Act prohibits discrimination based on race, religion, family status, and other protected characteristics (HUD, Fair Housing Act). Ask for data: schools, commute, noise, zoning plans, flood history.
- The result has an expiration date. Rates, prices, and offers change weekly. The decision sheet with thresholds from the last prompt is more important than all the tables before it.
Why AI Makes Different Mistakes with Real Estate than with Stocks
A publicly traded company has a quarterly report, one document with numbers that everyone reads. A house has twenty scattered sources: the county assessor's registry with tax and square footage, a flood map, zoning plan, building permit records, community documents, price history, inspector's report. None of these are written for search engines. A language model that cannot find this will fill in a "typical" value and present it with certainty.
The second problem is the language of listings. A property description is marketing text, and language models are exceptionally well-trained in it: they can summarize a listing so that it sounds even better. Therefore, prompt 2 instructs the model to read the listing like a lawyer for the opposing party, not like an agent.
The third problem is local numbers. Property tax in California is recalculated to the market value at the time of sale (California tax authority on change of ownership), so the amount from the listing that the previous owner paid for twenty years has nothing to do with what you will pay. In other states, it works differently. A model that you do not force to check the rules in a specific county will take the number from the listing.
The proper role of AI – it does not appraise or advise. It gathers scattered data in one place, reads long documents faster than you, calculates scenarios, and asks questions you wouldn't think of. The decision and verification of key numbers remain your responsibility.
Before You Paste the First Prompt
Tool, Search, and Files
Enable web search from the “+” menu under the text box (instructions). You can add PDF files, such as the inspection report, from the same menu (documentation). It handles long community documents well.
Web search is planned for both free and paid versions (OpenAI help). You attach PDF files, spreadsheets, and documents with a paperclip (OpenAI help on files). You can use Deep Research mode for step 1.
Deep Research mode builds a report with sources (Google description); it is useful for evaluating the area. Subsequent prompts are pasted in the same conversation, in the same window.
One Calibration Question
Before you start, check if the model is searching the web and if it is reaching the correct office. Paste:
Before we begin: provide today's date. The property is located at [ADDRESS]. Write which county (or municipality) it is in, provide a link to the office's website that maintains the property tax registry for this address, and a link to the official flood hazard map for this location. If you cannot find something, write "I did not find it," do not guess.
If there is no date, the links lead to listing services instead of the office, or the model provides a "sample" office, the search is not working. Reply: "Search the web and provide official sites before we proceed further."
Where the Model Should Search
| What You Need | USA | Poland |
|---|---|---|
| Tax, square footage, ownership history | County assessor's registry and recorder; the model will find the address and link in the calibration question | Electronic Land and Mortgage Registers: owner, encumbrances, mortgages, easements |
| Flood and flood insurance | FEMA Maps, flood insurance program NFIP | Flood hazard maps on Geoportal |
| What can be built next door | City or county planning department (zoning), council meeting minutes | Local zoning plan: extract and drawing from the local zoning plan from the municipality office |
| Loan: documents and rates | CFPB Guide, including Loan Estimate and Closing Disclosure; Freddie Mac weekly rates | Bank offers; reports from NBP on the real estate market as price background |
| Health: radon and lead | EPA on radon, EPA on lead disclosure (homes built before 1978) | Energy performance certificate, inspection reports from the manager |
| Inspection standards | Standards from ASHI and InterNACHI: inspection is visual and non-invasive | Technical inspector or building appraiser; no single standard |
| Taxes after purchase | IRS Publication 527 (rental), Publication 523 (home sale) | PCC when purchasing on the secondary market (Civil Law Transactions Tax Act), rental tax |
| Neighborhood data | Census: income, age of buildings, percentage of renters | GUS, Local Data Bank |
Brackets in Prompts
In the boxes, there are brackets: [ADDRESS], [PRICE], [CURRENCY], [AMOUNT]. Replace them in the first line; the rest works unchanged. Each prompt has a header indicating whether to paste it in the same conversation, and a "Copy Prompt" button.
Framework: Eight Steps in One Conversation
- Buyer's Context
Why you are buying, for how long, for what, in which currency you are calculating, and what cushion you have. Without this, the numbers in step 3 will be calculated for someone else.
- Location
Flood, taxes, zoning plans, schools, commute, noise, price trend. With links to offices, not to listings.
- Offer
Read the listing as a document from the opposing party: price history, days on the market, discrepancies with the registry, permits, alarming phrases.
- Numbers
Total monthly and yearly ownership cost, and for rental: net income, cash-on-cash return, debt coverage, sensitivity to rate and vacancy.
- Documents
Inspection report, seller's disclosure, community documents. The model extracts defects, estimates repair ranges, and writes questions.
- Pre-mortem
The model assumes that in three years you will regret it and looks for reasons in what is visible today. It ends with one condition after which the black thesis falls apart.
- Offer and Negotiation
Maximum price, contingent conditions, list of demands after inspection, what to concede, what not to.
- Decision Sheet
One page with thresholds at which you exit the transaction and a checklist for closing day. For rental: review once a year.
Why in this order. Location before the offer, because a bad location invalidates the best house. Numbers before documents, because documents improve numbers (a roof replacement is a specific amount in step 3). Pre-mortem after everything, because only then does it have something to draw from. Negotiation after pre-mortem, because red flags are your arguments. The sheet at the end, because it summarizes everything.
Prompt 0 – Buyer's Context
This step changes the outcome of the next three. A house for ten years and a house for three years yield different numbers: transaction costs upon sale are spread over a different number of years. For the Polish diaspora, currency comes into play: if you earn in dollars but buy in zlotys, the exchange rate is a separate risk that the model must calculate. There is also the issue of tax residency; we describe it in the guide on tax residency and PIT.
We begin the property analysis at [ADDRESS], listing price [PRICE]. Before you assess anything, remember my context and apply it throughout the conversation:
– Purpose: [I live alone / I live and rent part / I rent entirely / renovate and sell].
– Horizon: [e.g., at least 7 years].
– Financing: [cash / loan; down payment X percent; pre-approval for amount Y; rate from bank offer Z percent].
– I calculate in currency: [CURRENCY]. If the property is in a different currency, convert at the rate with the date and show how a 10 percent change in the rate affects the outcome.
– Financial cushion after purchase: [AMOUNT]. This cannot drop below [e.g., 6 months of costs].
– I settle taxes in: [country / state]. Do not give tax advice; indicate where tax changes the outcome and ask me to confirm it with an advisor.
– Hard constraints: [e.g., commute to work up to 40 minutes, two bedrooms, no flood zone].
Confirm in three sentences how you understand this context. Do not assess the property yet.
Check in the response: whether the model has already started praising the area or the price. If so, reply: "Do not assess yet. Wait for the next prompt."
Prompt 1 – Location
A house can be renovated, but the street cannot. This prompt gathers what is not in the listing: flood map, tax rules in this county, zoning plans in the neighborhood, noise, commute, and price trend. Questions about schools concern results and zoning, not "what children go there": fair housing rules apply to your research as well (HUD).
You are a real estate market analyst. Evaluate the location of the address [ADDRESS]. Use web search.
DATA RULES
– Each piece of information must have a source with a link and date. Order of sources: office (county, city, municipality, federal agency), then statistical data, finally listing services. Mark each item: [U] office, [S] statistics, [O] listing service.
– What you cannot find for this specific address, do not generalize from the city or state level. Write "I did not find it" and indicate where to check.
– Do not describe the residents. I am interested in data about the place, not about people.
DELIVER IN THIS ORDER
1. Flood and natural hazards: zone on the official flood map, whether flood insurance is required with the loan, history of floods and other events (fires, hurricanes, landslides) within a reasonable radius for this area. Link to the map.
2. Property tax: how it is calculated in this county or municipality, whether it is recalculated upon sale, what the effective rate is, what exemptions apply to the owner living on-site. Provide an estimate of the tax after purchase at the price [PRICE] and state the basis for it.
3. What will change around: zoning plans for neighboring parcels, building permit applications in the area, planned roads, rail lines, major investments, changes in the local plan. Source: planning department, council minutes.
4. Noise and nuisances: airport, highway, railway, factories, bars, stadium. Distances in meters or minutes.
5. Commute: time to [WORKPLACE] during peak hours and off-peak, public transport.
6. Schools: zoning for this address and public results, with the source. No qualitative assessments.
7. Price trend: median sale price in this area over the last 12 and 36 months, number of listings, median days on the market. Provide the source and how large the area is that you are referring to.
8. Age of buildings and structure: median year built, percentage of rented homes, from census data. If the house is from before 1978, note the obligation to disclose lead.
9. Three things you would check on-site that cannot be verified online.
10. LOCATION BRIEFING: five sentences, just facts with numbers. We will return to this.
FORMAT
Tables and short paragraphs. No introduction, no summary, no assessment of "whether it is worth it."
Check in the response:
- Point 1 must have a link to the official map, not to an article. In the USA, this is FEMA maps. Enter the address yourself and compare the zone with what the model provided.
- Point 2 must explain how the tax is calculated in this county, not just provide an amount. If the model took the amount from the listing, ask it to calculate based on the rule.
- Markings [U], [S], [O]. If most are [O], reply: "Replace data from listing services with data from the office or state that they do not exist."
- Point 3 often returns empty. In that case, open the planning department's website and paste the link to the model: "Search this page for [street]."
Prompt 2 – Offer
The listing is sales text. This prompt instructs the model to read it like a lawyer for the opposing party: compare it with the registry, trace the price history, catch phrases that conceal something, and build a list of questions for the agent. Paste the content of the listing and the link.
Same conversation. Below, paste the content of the listing and the link. Read it as the buyer's lawyer, not as the selling agent.
RULES
– Compare each number from the listing with the office registry: square footage, year built, number of rooms, lot size, tax. List discrepancies separately, with both values and sources.
– Do not assess whether the house is nice. Assess what the listing says, what it does not say, and what it suggests.
DELIVER
1. Offer history: when it was listed, price changes with dates, how many days on the market, whether it was previously listed and withdrawn, when and for how much it was last sold. Sources.
2. Price per square foot or meter next to the median for the area from step 1. Percentage difference.
3. Alarm phrases and what usually stands behind them: "as is," "motivated seller," "potential," "possible additional dwelling," "new installations" without a date, "part finished without a permit." List each from the listing and add what to ask about.
4. Permits: check with the city or county registry what building permits have been issued for this address and compare with what the listing describes as "new" or "renovated." Work done without a permit should be listed separately.
5. What is missing: which rooms are not shown in the photos, what the listing is silent about (roof, furnace, windows, basement, foundation, installations, age of appliances).
6. List of 12 questions for the selling agent, from the most important. Each formulated so that the answer is a fact, not an opinion.
7. OFFER BRIEFING: five sentences of facts.
LISTING:
[paste content and link]
Check in the response: whether point 1 has dates and sources, not "the listing has been on the market for some time." Whether the square footage in point 2 comes from the registry, not from the listing; a difference of several percent between them is common and changes the price per square foot. If point 4 returns as "I did not find it," enter the address in the county's permit search yourself, if it has one, and paste the result to the model.
Prompt 3 – Numbers
Here, most buyers only calculate the payment. The total cost of ownership includes the payment, tax after recalculation, insurance, community fees, utilities, maintenance, and reserve for major repairs. For rental, vacancy, management, tenant turnover, and income tax come into play. Internet thumb rules ("one percent of value per year for maintenance," "five percent vacancy") are a starting point, not data; the prompt instructs the model to replace them with local numbers wherever possible.
Same conversation. Calculate the total cost of this property for my context from prompt 0.
RULES
– Each item must have a source or be marked as an assumption. List assumptions separately at the end, with the value and justification, so I can replace them.
– Take the tax from step 1 (after recalculation, not from the listing). Mark insurance as an assumption to be confirmed with the agent's offer; if the house is in a flood zone, add flood insurance separately.
– Interest rate: from my bank offer; next to it, show the current market average with a source and date.
– Everything in [CURRENCY], with the rate and date if conversion is needed.
DELIVER
1. Monthly table: payment (principal and interest separately), tax, insurance, flood insurance, community fees, utilities, current maintenance, reserve for major repairs. Total. If the down payment is below 20 percent, add mortgage insurance and state when it disappears.
2. One-time costs: closing costs broken down, inspections, appraisal, moving, repairs from day one. Total and how much remains from my cushion after purchase.
3. Comparison with renting a similar unit in this area for my horizon: total cost of buying (including selling costs at the end) against total cost of renting, under three assumptions of price growth: zero, historical average for the area, half of that average. Provide a threshold: at what annual price increase does buying break even.
4. If rental is in context: rental income with a source (comparable offers from the last 60 days, with links), vacancy, management, turnover, net operating income, capitalization rate, cash-on-cash return in the first year, coverage of payment by net income. Separately: cash flow after tax noting that depreciation and tax must be confirmed with an accountant.
5. Sensitivity: a table where you change the interest rate by one percentage point, vacancy by five points, rent by ten percent, tax by twenty percent. Show how the monthly result changes and which variable hurts the most.
6. Critical point: at what monthly cost my cushion falls below the threshold from prompt 0 within 12 months if income (or my salary) drops by 20 percent.
7. LIST OF ASSUMPTIONS for replacement.
Check in the response:
- Insurance and tax are two numbers you check yourself: tax in the county registry according to the rule from step 1, insurance from an agent's offer. The model guesses both.
- Is the reserve for major repairs in point 1 based on the age of the roof, furnace, and installations, not on "one percent"? If step 4 (documents) has not been done yet, return to this table after the inspection.
- Does the market rate in point 1 have a source and date; in the USA, the benchmark is Freddie Mac weekly data, loan documents are explained by CFPB.
- Is the rent in point 4 from specific offers with links, not from "the average for the city"?
Prompt 4 – Documents
The most valuable step and the only one where the model works on what you provide, not on what it finds. Attach the inspection report, seller's disclosure about the property's condition, and for an apartment or house in a community: regulations, budget, status of the repair fund, and meeting minutes from the last two years. Inspection in the USA is by definition visual and non-invasive (standards from ASHI and InterNACHI), so what the inspector did not see is a separate category in the prompt.
Same conversation. I am attaching: [inspection report / seller's disclosure / community documents / title report]. Work only on the content of these documents; if something is missing, write that it is not there.
DELIVER
1. Defects from the inspection report grouped: safety, structure and water (roof, foundation, drainage, moisture, mold), installations (electrical, plumbing, heating, air conditioning), others. For each: page number of the report, quote from the inspector, weight (high, medium, low), cost range for repairs with indication of what you base it on, and whether a specialist is needed for the estimate.
2. Age and remaining lifespan: roof, furnace, air conditioning, water heater, windows, electrical installation. Value from the report, typical lifespan with a source, how much is left.
3. What the inspector did not check: list every sentence like "not accessible," "not tested," "recommended evaluation by a specialist" and add what kind of testing it requires (sewer camera, radon, asbestos, lead, oil tank, chimney, septic tank, termites).
4. Seller's disclosure versus report: what the seller marked as "I don't know" or "no," and the inspector found. Each discrepancy separately.
5. Community, if applicable: amount of the repair fund per unit, planned major repairs, whether the minutes mention additional fees, lawsuits, owner arrears, rental restrictions, pets, renovations. Quote with page number.
6. Title report, if attached: easements, liens, restrictions, boundary disputes.
7. Updated repair reserve to the table from step 3: amount on day one and amount for the first five years, broken down.
8. List of things to request from the seller before closing: repairs, price reduction, credit for repairs. Organize by what has the highest value for me and the least difficulty for the seller.
Check in the response:
- Each defect has a page number and quote. Without this, the model might have added it from general knowledge. Open two random pages and compare.
- Cost ranges in point 1 are guessed. Treat them as an order (what is expensive, what is cheap), not as an estimate; before negotiation, get an estimate from a contractor.
- Radon: EPA recommends testing every home and taking corrective action at levels of 4 pCi/L (EPA guide for home buyers). If the inspection report does not include a radon test, it is an item for point 3.
- Homes built before 1978: sellers are required to disclose known information about lead (EPA). Check if the form is among the documents.
Prompt 5 – Pre-mortem
Instead of asking "what are the risks," you assume that in three years you will regret it and instruct to find reasons in what is visible today. The technique comes from project management (Gary Klein's article in Harvard Business Review) and works particularly well in real estate because buyers are already emotionally invested in the home after several visits.
Same conversation. Now you are playing against me. Assume I bought this property for [PRICE] and three years later I regret it: I lost money or peace. Write why.
RULES
– Do not soften. Not a single sentence starting with "on the other hand," "it is worth noting," "nevertheless."
– Each accusation based on something from steps 1 to 4: a number, a quote from a document, an entry in the registry. An accusation without evidence is discarded.
– No catastrophes. I am interested in what is visible today.
DELIVER
1. Five red flags. For each: weight, evidence with source, mechanism (how exactly it translates into cost or lack of ability to sell), and the quarter in which it will hurt.
2. Costs that rise on their own: tax after recalculation and its trend in this county, insurance in this zone (trend of premiums over the last three years, with a source), community fees (trend from budgets), planned special assessments. Table: today, in three years maintaining the trend.
3. Scenario "I cannot sell": what would have to happen to prices in the area, rates, and my situation for selling in three years to mean a loss after transaction costs. Provide a selling price below which I lose, with a breakdown of costs.
4. If rental: tenant scenario who does not pay. How long does the procedure take in this state or country, how much does it cost, what are the local rent and eviction restrictions. Source: state regulations or law, with a link.
5. Silent cost: how much I lose by keeping the down payment and closing costs in this property instead of in treasury bonds in my currency if the price stays flat for three years. Interest rate with source and date, pre-tax result.
6. One condition after which this black thesis no longer applies. Verifiable: name of the document, test result, or number in the registry.
7. Three questions you do not know the answers to that would change the picture. For each: who to ask.
Check in the response: softening sentences despite the ban; reply "remove all softening sentences." Point 4 must quote a state regulation or law, not "it usually takes a few months." Point 6 must be a condition, not a mood: "sewer camera test without damage" is a condition, "the area will develop" is not.
Prompt 6 – Offer and Negotiation
Only now, having the numbers, documents, and red flags, do you set the maximum price and a list of conditions. The model should not provide a "fair price," only three levels of offers with justification, a list of contingent conditions, and what to demand after inspection and what to concede.
Same conversation. Prepare me to make an offer. Do not provide one "correct price."
DELIVER
1. Three levels of offers: cautious, base, maximum. For each: price, justification with numbers from steps 1 to 5 (comparable sales from the last 90 days with links, days on the market, repair costs, tax recalculation), and what I do if the seller refuses.
2. Price above which I walk away, with one sentence why. This should be the number I will write on the sheet in step 7.
3. Contingent conditions for the offer: inspection, appraisal, financing, title, sale of my current property, review of community documents. For each: what it gives me, how many days are standard in this market, what I lose if I concede it.
4. After inspection: list of demands from step 4 divided into three baskets: I demand repairs, I demand a reduction or credit for repairs, I concede. Amounts and order.
5. Questions for my agent and lawyer before signing: ten, from the most important.
Check in the response: whether comparable sales in point 1 are closed (sold), not listed; listing prices are not data. Whether they are from the last 90 days and from the same area, not from the whole city. Loan: Loan Estimate and Closing Disclosure documents and their deadlines are explained by CFPB guide to the loan process; the model can help read them, but you check the deadlines in the document.
Prompt 7 – Decision Sheet
The entire analysis fits on one page: thresholds at which you exit, a checklist for closing day, and, for rental, a review once a year. The sheet written today protects you from yourself in three weeks when you will already be emotionally invested in this home.
Wrap these steps into one sheet that I will paste into my notes.
– Thesis in three sentences: why I am buying and what must be true for it to be a good decision.
– Maximum price from step 6 and maximum monthly cost from step 3.
– Five exit conditions: an event or outcome of the inspection after which I withdraw regardless of emotions (e.g., confirmed flood zone, repair fund below threshold, sewer needs replacement, appraisal below price, no permit for extension).
– Checklist before closing: documents to read, tests to do, amounts to confirm with a person (tax, insurance, closing costs), final walkthrough.
– If rental: three numbers to check every year (market rent for comparable listings, total costs, coverage of payment) and values at which I consider selling or refinancing.
– One sentence: what I still do not know.
Without introduction and summary. Just points.
Check in the response: thresholds are numbers and document names, not adjectives. Save the sheet outside the chat and return to it before signing the offer, after the inspection, and before closing.
Appendix: Apartment in Poland Purchased from Abroad
The framework is the same, but the sources are different. For prompts 1, 2, and 4, add the block of rules below. Three differences that change the analysis:
- The land and mortgage register is public and online (Electronic Land and Mortgage Registers): owner, mortgages, easements, claims. The register number should be in the listing or from the seller; without it, there is no analysis.
- Local zoning plan instead of zoning. An extract and drawing from the local zoning plan is issued by the municipality (procedure on biznes.gov.pl); if there is no plan, the decision on building conditions applies, which is a separate risk for the view from the window.
- Transaction costs are different than in the USA: on the secondary market, the Civil Law Transactions Tax (PCC; rate and exemption for the first apartment to be confirmed with a notary on the day of the transaction), notary fee, entry into the land and mortgage register; on the primary market, VAT is included in the price and the developer's escrow account. Price background in major cities is published quarterly by NBP. Rules for saving and investing in Poland from abroad, including the issue of the American PFIC regime for funds, are described in a separate guide.
The property is in Poland, I am buying from abroad. Additional rules:
– Land and mortgage register number: [REGISTER NUMBER]. List sections I to IV: owner, method of acquisition, encumbrances, mortgages, easements, mentions of pending applications. Each mention is a question for the seller.
– Check whether a local zoning plan applies to the parcel and what it allows on neighboring parcels; if there is no plan, state this directly as a risk.
– Secondary market: list transaction costs (PCC, notary fee, court entries, agent) with legal basis and indicate which require confirmation with a notary. Primary market: status of the escrow account, delivery date, information prospectus, developer's entry in KRS.
– Apartment: amount of administrative rent broken down, repair fund, planned repairs of common areas, community or cooperative debt, form of ownership (separate ownership, cooperative ownership, share in the land).
– All amounts in zlotys, alongside conversion to [CURRENCY] with the rate and date and the impact of a 10 percent change in the rate.
– If I plan to rent: form of the contract (occasional rental requires the tenant's declaration with a notary), rental tax in Poland, and the obligation to report income in the country of residence – indicate for confirmation with an advisor.
– Purchase through a proxy: what the power of attorney must include, where to certify it abroad, and whether apostille is needed.
Numbers that AI Often Gets Wrong
| Number | Typical Error | How to Check |
|---|---|---|
| Property tax | Amount from the listing that the previous owner paid; after sale, it is often recalculated | Rule from the county office; calculate from the purchase price |
| Insurance | "National average"; in flood and hurricane zones, premiums can be many times higher | Quote from an insurance agent for this address, before the purchase offer |
| Square footage | The listing counts the basement and garage, the registry does not; the difference changes the price per square foot | County assessor's registry or land and mortgage register |
| Comparable sales | Listing prices instead of closed transaction prices; listings from the entire city | Only sold, last 90 days, same area |
| Potential rent | Average for the city or upper limit from listings | Specific offers from the last 60 days with links, then median |
| Repair costs | Ranges "from the internet" without relation to local rates | Contractor's estimate; take only the order from the model |
| Community fees | Current fee without planned increases and special assessments | Budget and meeting minutes from two years |
| Loan rate | Value from the model's memory from a year ago | Bank offer plus current average with source and date |
| Closing costs | One percent "for simplicity" | Loan Estimate from the lender; in Poland, calculation from the notary |
Indicators for Rental Properties
The model uses these terms loosely; it is worth knowing what they mean and remembering that thumb rules are a starting point for your own numbers, not a norm.
| Indicator | What It Says | Trap |
|---|---|---|
| Net Operating Income (NOI) | Rent minus vacancy and all costs except loan payment | Omitted management and reserve for repairs inflate the result |
| Capitalization Rate (cap rate) | NOI divided by price; compares properties regardless of financing | Calculated on listing rent, not on signed agreement |
| Cash-on-Cash Return | Annual cash flow after payment divided by down payment and closing costs | First year without repair reserve looks better than subsequent ones |
| Debt Service Coverage Ratio (DSCR) | NOI divided by annual payment; investment banks look at this ratio | Below one, the property subsidizes itself from your salary |
| Gross Rent Multiplier | Price divided by annual rent; a quick filter, nothing more | Ignores costs, taxes, and technical condition |
| Thumb Rules (1 percent rent monthly, 1 percent for maintenance yearly, 5 to 8 percent vacancy) | Starting point for discussion | Not data; in prompt 3, replace them with local numbers |
| Depreciation | In the USA, a residential rental building depreciates over 27.5 years (IRS Publication 527) | Land is not depreciable; upon sale, depreciation is recaptured for tax; confirm with an accountant |
Safety and Privacy
- Pre-approval letters, bank statements, or documents with Social Security numbers. For numbers, the amount and rate are sufficient.
- Personal data of the seller from disclosures and deeds; before attaching, redact names and numbers.
- Transfer instructions for closing. Scams involving swapping account numbers just before closing are a real and costly phenomenon; confirm the account number by phone with a known person, never from an email (CFPB warning).
Check model training settings in your account: Claude (Anthropic article), ChatGPT (OpenAI data settings). We discuss how fake "investment opportunities" and groups look in the guide on fake investments. If you are financing the purchase with an FHA loan, check the guide on FHA for those without a long credit history.
Common Mistakes
- Starting with "is this a good price." You will get confirmation. Always. The price is in step 6, after five steps of facts.
- Tax from the listing. In many places, this is a historical number. After purchase, you pay according to county rules.
- Payment instead of total cost. Tax, insurance, community fees, maintenance, and reserves can add up to double.
- Analysis without documents. Steps 1 to 3 are from the web. Value starts in step 4 when the model reads the inspection report and community minutes.
- Trusting repair ranges. The model does not know local rates. Order yes, amounts no.
- Questions about people instead of data. Besides being against fair housing rules, the answer will still be fabricated.
- Skipping pre-mortem because "I already know I want it." That is when it is most needed.
- Sheet in the chat. In three weeks you will not find it. Notes, file, calendar.
This framework organizes research and forces a look at the other side. It does not replace an inspector, lawyer, loan advisor, or accountant. The model can provide a number that is not there, and it will do so with certainty. Key amounts (tax, insurance, closing costs, repairs) must be confirmed with people before you sign anything.
Frequently Asked Questions
Can AI replace an inspector?
No. The inspector sees the house; the model sees the inspector's report. However, the model is better at reading sixty pages of the report and a hundred pages of community minutes in a few minutes and turning them into a list of questions.
Can I use this framework before the first visit?
Yes, steps 0 to 3 work on data from the web and listings. Step 4 requires documents that you get after making an offer or upon request. It is worth doing steps 0 to 3 for several listings and only then going to see them.
Which tool is best?
The one where you have web search enabled and can attach PDFs. The differences between Claude, ChatGPT, and Gemini are smaller than the difference between step 4 with documents and without them.
Does this work for a lot or a house under construction?
Partially. Steps 1, 3, 5, and 7 do. Instead of inspection in step 4, you attach the zoning plan, building conditions, soil tests, and contract with the contractor or developer.
How often to return to the sheet when renting?
Once a year and after every event: tax increase, tenant change, major repair, rate change. Three numbers from the sheet will suffice to know whether to hold, refinance, or sell.
Can the model write an offer or contract?
It can draft questions and a list of conditions. The contract, offer, and power of attorney are prepared by an agent, lawyer, or notary; in many states and in Poland, the form of the document is specified by regulations.
Fact-Check Summary
- Definitely true: FEMA flood hazard maps are public; in California, sale triggers tax recalculation to market value; EPA recommends action on radon at levels of 4 pCi/L; a seller of a home built before 1978 is required to disclose information about lead in the USA; a residential rental building depreciates in the USA over 27.5 years; land and mortgage registers in Poland are available online. All with links in the text.
- Probably true: language models more often confirm the user's thesis and smooth out criticism; property tax listed is a historical value in many counties. Rules vary between states, which is why prompt 1 instructs to check them locally.
- What is uncertain: cost ranges for repairs and insurance premiums provided by the model; PCC rates and exemptions in Poland on the day of your transaction (confirm with a notary); the quality of search in a specific tool, which changes with the version.
- Common myth: "AI will appraise the house." It will not appraise. It can gather data, read documents, and show at what numbers the decision no longer makes sense.
Comments (0)
No comments yet. Be the first!