Disclosure
Pacific Software Ventures (PSV) builds AI agents for commercial real estate firms, so we have a stake in when a prompt library is enough and when it is not. The prompts below are free to copy and work without us.
What makes a good ChatGPT prompt for commercial real estate?
The best ChatGPT prompts for commercial real estate give the model a role, tell it exactly what to paste in, fix the output format and make it flag anything it can't source from the document. Use them to extract and structure: rent rolls into your model's columns, T-12s into your categories, lease terms, OM screens against your buy box, and first drafts of memos, letters and outreach. Let your spreadsheet do the arithmetic, and have a person check the output against the source before anyone relies on it. Below are 50 copy-ready prompts grouped by role: acquisitions and underwriting, brokers and leasing, asset management, lenders, development and research. Each one works in ChatGPT or Claude.
How do you write a reusable CRE prompt?
A prompt you can reuse on every deal has the same five parts every time. Write it once, test it on a deal you have already closed, and save it where your team can find it.
- Give it a role. One sentence that says who the model is and what job it is doing: an acquisitions analyst screening an OM, a lender's underwriter checking guidelines. The role sets the vocabulary and the level of detail.
- Name the inputs, in brackets. Say exactly what to paste or attach: the rent roll, the T-12, your buy box, your template headings. Brackets make the prompt a form anyone on the team can fill, and they stop the model filling gaps from general knowledge.
- Number the steps. Break the task into the steps a good analyst would take, in order. Numbered steps are easier to check, and when the output is wrong you can see which step failed.
- Fix the output format. Say what comes back: a table in your column order, a memo in your template headings, a list of exceptions. If the output goes into Excel, ask for the columns your model already uses.
- Close with the same rules. Every prompt here ends with one rules line: use only the pasted material, cite the page or tab for every figure, copy numbers exactly as written, do not calculate, and flag anything you can't source from the document. That one line does more for accuracy than any clever wording.
- Test it on a closed deal. Run the prompt on a deal whose numbers you already know. Where it misses, tighten the step that failed, then save that version with a date so the team knows which one to use.
50 ChatGPT prompts for commercial real estate, by role
Copy a prompt, replace each bracket with your own material, and run it. Every prompt names its inputs, its steps and its output, and ends with the same rules line. They are written for the documents firms we talk to handle every week: OMs, rent rolls, T-12s, leases, loan packages, budgets and research reports.
Acquisitions and underwriting
Screening OMs, setting up the rent roll and T-12 for your model, lease terms, broker questions and the path to IC.
Screen an OM against your buy box
You are an acquisitions analyst at a commercial real estate investment firm. Screen this deal against our written buy box. Inputs: - [paste our buy box: markets, asset types, size range, vintage, basis and hold criteria] - [attach or paste the OM or teaser] Steps: 1. Pull the asset name, address, submarket, size (units or square feet), year built, occupancy, price guidance and call-for-offers date. 2. Check each buy box criterion one at a time. Mark it Meets, Fails or Unknown and quote the line of the OM you relied on. 3. Recommend Pursue, Pass or Needs information in one sentence that names the deciding criterion. Output: A two-column table of deal facts, then a criteria table (criterion, OM evidence, page, verdict), then the one-line recommendation. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Set up a rent roll in your model's columns
You are an underwriting analyst preparing a rent roll for our Excel model. Inputs: - [attach the rent roll, PDF or Excel] - [paste the column headings of the rent roll tab in our model] Steps: 1. Map each rent roll field to our headings. Keep one row per unit or suite, in the rent roll's order. 2. Keep vacant units as rows and mark them Vacant. 3. List every unit whose lease dates, rent or square footage are missing, unreadable or inconsistent with the rest of its row. Output: A table in our column order that I can paste straight into Excel, with no total row, then a list of exceptions (unit, field, reason). Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Map a T-12 to your chart of accounts
You are an underwriting analyst normalizing a seller's trailing twelve-month operating statement. Inputs: - [attach the T-12] - [paste our income and expense categories] Steps: 1. Assign every T-12 line to one of our categories and keep the original line name beside it. 2. Keep the twelve monthly figures exactly as written. Do not add, average or annualize anything. 3. Flag lines that look one-time or unusual (a single large month, a refund, a settlement, a line that appears once) with the month and a short reason. Output: A table (original line, our category, the twelve months as written), then the flagged lines with your reason for each. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Abstract a lease for underwriting
You are a lease analyst abstracting a commercial lease for an acquisition underwriting. Inputs: - [attach the lease and every amendment, in date order] Steps: 1. Pull tenant, premises, square feet, commencement and rent commencement, expiration, the base rent schedule as written, escalations, renewal, expansion, contraction and termination options, rights of first offer or refusal and any purchase option, expense recovery and caps, free rent, TI allowance and landlord work, security deposit or guaranty, assignment and subletting, exclusive use, co-tenancy, and subordination or SNDA terms. 2. Where an amendment changes a term, show the latest term and cite the amendment that changed it. 3. List clauses that are ambiguous or unusual enough to need counsel's read. Output: A two-column abstract (term, value with section cite), then the list of clauses for counsel. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Lay out the broker's pro forma assumptions
You are an acquisitions analyst. Lay out the assumptions behind the broker's pro forma so our team can set its own. Inputs: - [attach the OM, including the pro forma and its footnotes] Steps: 1. List every assumption the broker states or implies: rent growth, vacancy and credit loss, concessions, other income, expense growth, reserves, capital plan, exit cap rate and hold period. 2. For each one, give the page and say whether it is stated outright or only implied by the numbers. 3. Write Not stated for any assumption the OM does not give. Do not judge, replace or recalculate any assumption. Output: A table (assumption, broker's figure as written, page, stated or implied) with an empty column headed Our assumption for the team to fill. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Draft the screening memo for approval to bid
You are an associate drafting a short memo asking our investment committee for approval to bid. Inputs: - [paste the deal facts table from the screen] - [paste the returns and key metrics from our model, exactly as the model calculated them] - [paste our memo template headings] Steps: 1. Write each section of the template using only the pasted facts and model outputs. 2. Quote every figure exactly as given. Where a section needs the team's judgment (the thesis, the bid, the key risk), write [TEAM TO CONFIRM] instead of an opinion. 3. List the risks the documents support, each with its source. Output: The memo in our template headings, no longer than two pages, then a list of every [TEAM TO CONFIRM] placeholder. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Find what changed between two versions of a deal document
You are an acquisitions analyst comparing a revised deal document with the earlier version. Inputs: - [attach the earlier version of the OM, rent roll or T-12] - [attach the revised version] Steps: 1. List every figure, date, tenant, term or assumption that differs between the two versions. 2. Show the old value and the new value side by side with the page or row of each. Do not compute the size of any change. 3. List anything that appears in one version and not the other. Output: A change table (item, old value, new value, location in each), then the added and removed items. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Write the question list for the broker
You are an acquisitions analyst preparing questions for the listing broker before we underwrite further. Inputs: - [attach the OM, rent roll and T-12] - [paste any exception lists from earlier prompts] Steps: 1. Find missing items, inconsistencies between documents and terms the documents leave unclear. 2. Write one plain question for each, grouped under Rent roll, Operating statement, Physical and capital, and Legal and title. 3. After each question, cite what prompted it. Output: A numbered question list ready to paste into an email, with the citation after each question in brackets. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Check the data room against your DD checklist
You are a due diligence coordinator checking a data room against our checklist. Inputs: - [paste our due diligence checklist] - [paste the data room index or folder listing] Steps: 1. Match each checklist item to the file or files in the index that answer it. 2. Mark each item Received, Partial or Missing, and say what is missing for a Partial. 3. List files in the data room that match no checklist item. Output: A table (checklist item, matching files, status, note), then the unmatched files. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Compare a tenant estoppel with the lease
You are a due diligence analyst comparing a tenant estoppel certificate with the lease abstract. Inputs: - [attach the signed estoppel] - [paste the lease abstract, or attach the lease and amendments] Steps: 1. For each term the estoppel states (rent, dates, deposit, options, defaults, prepaid rent, landlord obligations), show the estoppel's version and the lease's version side by side. 2. Mark each Matches, Differs, or Not in estoppel. 3. Quote any default, dispute or claim the tenant asserts, word for word. Do not judge whether a difference is material. Output: A comparison table, then the quoted tenant assertions, then the differences for our counsel to review. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Brokers and leasing
Prompts for brokers: BOVs, OMs, owner outreach, inbound inquiries, LOIs, tour books and lease comps.
Draft a BOV narrative from your own comps
You are an investment sales broker drafting the narrative sections of a broker opinion of value. Inputs: - [paste the subject property facts] - [paste the comps from our own closed deals and broker packages, with the source of each] - [paste the value range from our pricing model, as calculated] - [paste our BOV template headings] Steps: 1. Write the property description, the market narrative and the comp discussion using only the pasted facts. 2. Present the value range exactly as our model gives it. Do not propose, adjust or round a value. 3. Note which comps are closest to the subject on location, size, vintage and tenancy, and why, in a sentence each. Output: The BOV sections in our template headings, in plain owner-facing language. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Draft offering memorandum sections
You are a capital markets marketing associate drafting an offering memorandum. Inputs: - [paste the property fact sheet] - [paste the rent roll summary and financial summary from our model] - [paste a past OM of ours for tone and structure] Steps: 1. Write the executive summary, investment highlights, property overview, location overview and tenancy sections in the structure of our past OM. 2. Every figure must come from the fact sheet or the model summary, quoted as written. 3. Keep claims factual. Write [BROKER TO CONFIRM] wherever a highlight needs our judgment. Output: The OM sections with headings, then a list of every figure used and where it came from. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Write a first-touch email to a property owner
You are a commercial real estate broker writing a first-touch email to an owner. I review and send every message myself. Inputs: - [paste the owner name and property facts from public records] - [paste one recent relevant deal of ours from our files] - [paste two past emails of mine for voice] Steps: 1. Write an email of no more than 90 words about their specific building, with one concrete reason to talk now. 2. Make no claim about the building's value and no promise about price. 3. Match the voice of my past emails. Output: Two versions of the email, each with a subject line, then any fact you used that I should double-check. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Reply to an inbound listing inquiry
You are a leasing broker replying to an inbound inquiry about one of our listings. Inputs: - [paste the inquiry] - [paste the listing fact sheet] - [paste my available tour times this week] Steps: 1. Answer the questions the prospect asked, using only the fact sheet. 2. Offer two tour times from my availability. 3. Write one short qualifying question about their size, timing or use. Output: A reply under 120 words in a friendly, direct tone, then a list of questions you could not answer from the fact sheet. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Compare an LOI with your proposed terms
You are a leasing broker comparing a counterparty's letter of intent with the terms we proposed. Inputs: - [attach the LOI] - [paste our proposed terms] Steps: 1. Go term by term: premises, term, rent and escalations, free rent, TI, expense structure, options, security, contingencies and dates. 2. Show their version and ours side by side and mark each Matches, Differs or Missing. 3. Do not recommend a counter. List the terms where the two documents use different definitions. Output: A side-by-side table, then the definition differences. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Match listings to a tenant's requirement
You are a tenant representation broker matching available spaces to a client's requirement. Inputs: - [paste the tenant requirement: size range, use, location, timing, parking, budget and must-haves] - [paste the flyers or fact sheets for each candidate space] Steps: 1. For each space, list its facts as written beside each requirement. 2. Mark each requirement Fits, Does not fit or Unknown for each space. 3. List the questions to ask each landlord to close the Unknowns. Output: One table per space, then a short list ordered by how many requirements each space fits, then the landlord questions. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Draft a tenant RFP from requirement notes
You are a tenant representation broker drafting a request for proposal to send to landlords. Inputs: - [paste my notes from the client requirement meeting] - [paste our RFP template] Steps: 1. Fill every section of the template from my notes. 2. Where my notes do not cover a section, write [NEEDED FROM CLIENT] rather than inventing a requirement. 3. Keep the client unnamed if my notes say the search is confidential. Output: The finished RFP in our template, then the list of [NEEDED FROM CLIENT] items as questions for the client. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Prepare a pre-call owner brief
You are a broker's research associate preparing a one-page brief before a call with a property owner. Inputs: - [paste the public record ownership and sale history] - [paste our CRM notes and past emails with this owner] Steps: 1. Summarize the ownership history, the property facts and every past touch we have had, with dates. 2. List what the owner told us before about plans, timing or concerns, quoting our notes. 3. Suggest three open questions for the call based only on these materials. Output: A one-page brief with headings: Property, Ownership, Our history, What they told us, Questions for this call. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Turn call notes into CRM fields
You are a brokerage coordinator logging a call in our CRM. Inputs: - [paste the call notes or transcript] - [paste our CRM field list] Steps: 1. Fill each CRM field from the notes. 2. Capture the next step and its date exactly as agreed on the call. 3. Leave a field blank rather than infer it. Output: A list of field: value pairs ready to paste, then a two-sentence call summary. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Normalize lease comps from broker emails
You are a leasing analyst building a lease comp set from comps that arrived by email in different formats. Inputs: - [paste the comp emails or attachments from our own deals and other brokers] - [paste our comp template columns] Steps: 1. Create one row per comp in our template columns. 2. Record each rent exactly as stated with its basis (net or gross, per year or per month, per square foot or total). Do not convert between bases. 3. Mark duplicates that appear in more than one email. Output: The comp table, then the duplicates and any comp missing its date, size or rent basis. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Asset management and reporting
Variance commentary, LP letters, monthly property reports, lease and debt tracking.
Write variance commentary from budget vs actual
You are an asset manager writing the monthly variance commentary for a property. Inputs: - [paste the budget vs actual report from our spreadsheet, with the variances already calculated] - [paste our variance threshold: [amount] or a percentage] - [paste the property manager's notes for the month] Steps: 1. For each line over the threshold, write one sentence explaining the variance using only the property manager's notes. 2. Where the notes give no explanation, write PM EXPLANATION NEEDED. 3. Say whether each variance is timing or permanent only if the notes say so. Output: A table (line, variance as given, explanation, timing or permanent), then a three-sentence summary for the monthly report. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Draft the quarterly LP letter
You are an investor relations associate drafting the quarterly letter to limited partners. Inputs: - [paste the quarter's metrics from our reporting workbook, as calculated] - [paste last quarter's letter] - [paste the business plan updates from the asset team] Steps: 1. Write in the structure and tone of last quarter's letter. 2. Use every figure exactly as given. Make no forward-looking promise; describe plans as plans. 3. Write [IR TO CONFIRM] wherever a sentence would need information not in the inputs. Output: The full letter draft, then a list of every figure used with its source line. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Summarize the property manager's monthly package
You are an asset manager reviewing a property manager's monthly reporting package. Inputs: - [attach the monthly package] - [paste the reporting requirements from our management agreement, or our monthly package checklist] Steps: 1. Pull occupancy and leasing activity, delinquency, collections, capital projects, service requests and open issues, each as the package reports it. 2. List anything the package says is behind plan, and every item on our checklist that the package does not include. 3. Write three questions for the property manager. Output: A one-page summary with those headings, then the questions. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Build a lease expiration and option tracker
You are a lease administrator building an expiration and option tracker for a property. Inputs: - [paste the lease abstracts, or attach the leases and amendments] Steps: 1. For each tenant, pull expiration date, renewal options, termination options, notice periods and notice deadlines exactly as the lease states them. 2. Where a notice deadline depends on a date the lease does not state, write NOT IN SOURCE. 3. Sort the rows by expiration date. Output: A table (tenant, suite, expiration, options, notice terms as written, section cite) I can paste into Excel, where the deadline dates get calculated. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Pull covenants and maturity terms from a loan agreement
You are an asset manager setting up a debt covenant and maturity tracker. Inputs: - [attach the loan agreement and any amendments] Steps: 1. Pull every financial covenant with its definition, threshold, test date and reporting deadline, quoting the definition word for word. 2. Pull maturity, extension options and their conditions, rate terms, reserves and cash management triggers. 3. Do not test any covenant. Our spreadsheet tests them from the definitions you extract. Output: A covenant table and a terms table, each with section cites, then any definition that refers to another document I need to send you. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Answer an investor question from approved documents
You are an investor relations associate drafting a reply to an investor question. Our IR lead reviews every reply before it is sent. Inputs: - [paste the investor's question] - [paste the approved documents that may answer it: the latest report, the relevant operating agreement section, the last letter] Steps: 1. Answer only from the pasted documents, citing each one. 2. If the documents do not answer the question, say so plainly and suggest what the IR lead should check. 3. Do not discuss performance beyond what the latest report states. Output: A reply under 150 words, then the citations. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Summarize capital project status
You are an asset manager summarizing capital project status across a property. Inputs: - [paste the contractor and project manager updates] - [paste the capital budget tracker from our spreadsheet, as calculated] Steps: 1. For each project, give status, schedule against plan and open issues as the updates describe them. 2. Quote budget figures exactly as the tracker shows them. 3. List decisions the updates say are waiting on ownership. Output: A table (project, status, schedule, budget as tracked, issues), then the decisions waiting on us. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Review the property manager's draft budget narrative
You are an asset manager reviewing the property manager's draft budget for next year. Inputs: - [paste the draft budget narrative and assumptions] - [paste last year's budget narrative] - [paste the actuals report from our spreadsheet] Steps: 1. List every assumption the property manager states (rents, occupancy, concessions, payroll, contracts, utilities, taxes, insurance, capital). 2. Mark which assumptions changed from last year's narrative and quote both versions. 3. Write a question for each assumption the narrative does not explain. Output: An assumptions table (assumption, this year as written, last year as written, changed or not), then the questions for the property manager. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Abstract a service contract
You are an asset manager abstracting a property service contract. Inputs: - [attach the service contract] Steps: 1. Pull vendor, scope, term, renewal, termination rights and notice period, pricing as written, escalations, insurance requirements and what happens on a sale of the property. 2. List every date that requires action from us. 3. Flag terms that are unusual or one-sided. Output: A two-column abstract with section cites, then the action dates, then the flagged terms. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Lenders and debt
Loan intake, credit memos, sponsor reviews, document checklists and term sheets.
Screen a loan request against your criteria
You are a loan officer at a commercial real estate lender doing a first screen on a new loan request. Inputs: - [paste our lending criteria: geography, property types, loan size range, leverage limits, sponsor requirements] - [attach the sponsor's loan request package] Steps: 1. Pull the loan request facts: property, location, type, loan amount requested, purpose, term, sponsor and timing. 2. Check each criterion. Mark it Meets, Fails or Unknown and cite the package page. 3. Where a criterion needs a calculated figure (leverage, coverage), write MODEL and name the inputs our model needs. Output: A facts table, a criteria table, and a one-line recommendation: Proceed, Decline or Need information. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Draft the credit memo
You are a credit analyst drafting a credit memo for our loan committee. Inputs: - [paste our credit memo template] - [paste the underwriting outputs from our model, including coverage, leverage and debt yield, as calculated] - [attach the sponsor package, appraisal summary and third-party reports] Steps: 1. Write each section of the template from the inputs, quoting every figure exactly as given. 2. Write [CREDIT TO CONFIRM] wherever the memo needs a credit judgment: risk rating, structure, mitigants, recommendation. 3. List the strengths and risks the documents support, each with a citation. Output: The memo in our template, then the [CREDIT TO CONFIRM] list. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Summarize a sponsor's experience
You are a credit analyst summarizing a borrower sponsor's background. Inputs: - [paste the sponsor's resume, track record schedule and personal financial statement summary] - [paste our reference call notes] Steps: 1. Summarize experience by property type and market, exactly as the track record states it. 2. List any gaps in the track record, missing years or deals without stated outcomes. 3. Quote what references said. Do not rate the sponsor. Output: A one-page summary with headings: Experience, Track record as stated, Gaps, References. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Check received documents and name the files
You are a loan closer checking which documents we have and naming them to our convention. Inputs: - [paste our document checklist for this loan type] - [paste the list of file names received] - [paste our file naming convention] Steps: 1. Match each received file to a checklist item. 2. Suggest a new file name for each, following our convention. 3. List the checklist items still missing. Output: A table (current file name, checklist item, suggested name), then the missing items. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Fill a term sheet from approved terms
You are a loan officer preparing a term sheet from terms our committee approved. Inputs: - [paste the approved terms] - [paste our term sheet template] Steps: 1. Fill each template field from the approved terms, word for word where possible. 2. Do not add any term that was not approved. 3. Mark any template field the approved terms do not cover as NOT APPROVED. Output: The filled term sheet, then the NOT APPROVED fields. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Check a deal against your underwriting guidelines
You are a lender's underwriter checking a borrower's rent roll and operating statement against our written guidelines. Inputs: - [paste our underwriting guideline document] - [attach the rent roll and T-12] Steps: 1. List each guideline and what the documents show for it, with the row or page. 2. Mark each Meets, Fails or Unknown. 3. Where a guideline needs a calculated figure, write MODEL and list the inputs the model needs from these documents. Output: A guideline table, then the list of model inputs with their source rows. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Request missing items from a borrower
You are a loan officer writing to a borrower about items we still need. Inputs: - [paste the list of missing items] - [paste the borrower contact name and the date we need the items by] Steps: 1. Write a short, polite email listing each missing item in plain words. 2. Say why the closing timeline depends on them, without pressure. 3. Offer a quick call if anything is unclear. Output: An email under 150 words with a subject line. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Development and construction
Zoning, draws, construction contracts, entitlements, pro forma inputs and bid leveling.
Summarize the zoning for a site
You are a development associate summarizing the zoning that applies to a site. Inputs: - [paste the zoning code sections for the site's district and any overlay] - [paste the site facts: address, parcel size, current use] Steps: 1. Pull permitted and conditional uses, height, setbacks, lot coverage, floor area ratio, density, parking and open space requirements, exactly as written. 2. Cite the code section for each. 3. Do not calculate buildable area or unit count. Our massing study and model do that from the rules you extract. Output: A rules table with section cites, then a list of approvals or variances the code says a project like ours would need. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Map a contractor pay application to your budget
You are a development project accountant mapping a contractor's pay application to our construction budget. Inputs: - [attach the pay application and its schedule of values] - [paste our budget line items] Steps: 1. Map each pay application line to one of our budget lines. 2. Quote each amount exactly as written. Do not total or net anything. 3. List pay application lines that match no budget line, and any line billed for the first time this period. Output: A mapping table I can paste into our draw spreadsheet, then the unmatched and first-time lines. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Abstract a construction contract
You are a development manager abstracting a general contractor agreement. Inputs: - [attach the construction contract and its exhibits] Steps: 1. Pull contract sum or GMP as written, contingency and allowances, retainage, schedule and substantial completion date, liquidated damages, change order process, payment terms, insurance and bonding, warranty and termination rights. 2. List every exhibit the contract refers to and whether it was attached. 3. Flag terms that shift unusual risk to the owner. Output: A two-column abstract with section cites, then the exhibit list, then the flagged terms. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Build an entitlement timeline from city correspondence
You are a development associate tracking entitlements for a project. Inputs: - [paste the city letters, staff reports and meeting minutes, in date order] Steps: 1. List each step: submittal, comment, hearing, approval or condition, with its date and status as stated. 2. Pull every condition of approval word for word. 3. List the next deadline or hearing each document mentions. Output: A timeline table (date, step, status, source), then the conditions, then upcoming deadlines. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Fill pro forma inputs from a site package
You are a development analyst setting up the inputs tab of our development pro forma. Inputs: - [paste the land deal terms] - [paste the architect's program: unit mix or square footage by use] - [paste the contractor's cost estimate] - [paste the headings of our pro forma inputs tab] Steps: 1. Fill each input heading with the figure from the documents, exactly as written, with its source. 2. Leave blank any input the documents do not give, such as rents, cap rates and financing terms. Our team sets those. 3. List figures that disagree between documents. Output: An inputs table in our heading order, then the blanks and the disagreements. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Draft the monthly development update for investors
You are a development manager drafting the monthly project update for investors. Inputs: - [paste the schedule update] - [paste the budget report from our spreadsheet, as calculated] - [paste the leasing or pre-sales update] - [paste last month's update] Steps: 1. Write in the structure of last month's update: progress, schedule, budget, leasing, next month. 2. Quote every figure exactly as given. 3. Describe any delay or overrun plainly, with the reason the inputs give. Output: The update draft under 500 words, then a list of every figure with its source. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Level competing bids
You are a development manager leveling bids from several contractors or consultants for the same scope. Inputs: - [attach each bid] - [paste our scope of work] Steps: 1. Lay out each bid's inclusions, exclusions, allowances, assumptions and schedule against our scope. 2. Quote each price exactly as written. Do not total or adjust any bid. 3. List the scope gaps and qualifications that make the bids not directly comparable. Output: A side-by-side leveling table, then the gaps to clarify with each bidder. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Research and market
Submarket briefs, supply pipelines, research reports and your own deal memory.
Write a submarket brief from your own sources
You are a research analyst writing a submarket brief for an investment committee. Inputs: - [paste our past underwriting notes on this submarket] - [paste the public broker research we receive, with dates] - [paste public data such as permit filings and major employer news] Steps: 1. Summarize supply, demand drivers, rents and vacancy as each source states them, naming the source and its date for every figure. 2. Where sources disagree, show each figure separately. Do not blend them into one number. 3. List what we would need to know that the sources do not cover. Output: A one-page brief with headings: Supply, Demand, Rents and vacancy, Disagreements, Gaps. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Build a supply pipeline from permit records
You are a research analyst building a new-supply list from public permit records. Inputs: - [paste the permit records from the city or county open data portal] - [paste our asset type and target area] Steps: 1. Keep only projects that match our asset type and area. 2. Create one row per project: address, use, units or square feet, permit type, status and date, exactly as the records show. 3. Mark records that look like duplicates or phases of one project. Output: A table I can paste into Excel, sorted by status, then the possible duplicates. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Summarize a broker research report
You are a research analyst summarizing a public market report from a brokerage. Inputs: - [attach the report] Steps: 1. Pull the key figures (vacancy, rents, absorption, deliveries, under construction, sales volume) with the page, period and market definition for each. 2. Note how the report defines the market and the property set it covers. 3. Pull the report's own outlook statements, quoted. Output: A figures table with page and period, then the definitions, then the quoted outlook. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Compare two reports on the same market
You are a research analyst comparing two research reports on the same market. Inputs: - [attach report A] - [attach report B] Steps: 1. Put each report's figures side by side for the same measures and periods. 2. Note where the market definition, property set or period differs, which can explain a difference. 3. Do not average, reconcile or pick between the figures. Output: A side-by-side table with page cites, then the definition differences. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Summarize news on your tenants
You are an asset management analyst watching news about the tenants in our portfolio. Inputs: - [paste the news articles] - [paste our tenant list with property and suite] Steps: 1. For each article that names one of our tenants, give the tenant, the event, the date and the source. 2. Say which of our properties the tenant occupies. 3. Describe the event as the article does. Do not speculate about what it means for the lease. Output: A table (tenant, property, event, date, source), most recent first. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Pull what a public filing says about your markets
You are a research analyst reading a public company filing or earnings call transcript for what it says about our markets or tenants. Inputs: - [attach the filing or transcript] - [paste our markets and asset types] Steps: 1. Pull every statement about our markets, asset types, leasing, store or office plans, and guidance. 2. Quote each statement and give its page or timestamp. 3. Separate what the company reported as fact from what it described as plans or expectations. Output: Two lists, Reported and Plans or expectations, each item quoted with its cite. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Ask what your firm already knows about a submarket
You are a research analyst answering a question from our own deal history. Inputs: - [paste excerpts from our past memos, underwriting notes and broker correspondence on this submarket] - [paste the question, for example: what do we already know about this submarket?] Steps: 1. Answer only from the excerpts, citing the document and date for each point. 2. Put the most recent information first and say when an older note may be out of date. 3. List what the excerpts do not cover. Output: A short cited answer, then the gaps. Rules: Use only the material I pasted. Give the page, tab or section for every fact and figure. Copy numbers exactly as written. Do not calculate totals, averages, ratios or returns; my spreadsheet does that. Flag anything you can't source from the document as NOT IN SOURCE instead of guessing.
Why don't these prompts ask the AI to do the math?
Because a language model writes the most likely next words, not the result of a formula. It can read a T-12 well and still hand back an NOI that is off by a line it skipped, stated with the same confidence as a right one. So every prompt here reads and structures, and the spreadsheet calculates.
- A formula can be checked; a chat answer cannot. When a number comes out of your model, anyone can click the cell and see how it was built. A number typed in a chat reply has no formula behind it, so no one can audit it before it reaches a memo.
- Your assumptions stay in one place. Exit cap, rent growth, vacancy and capital belong in your model, set by your team. A prompt that calculates returns quietly brings in its own assumptions.
- Extraction is where AI saves the time. The hours go into re-keying the rent roll, mapping the T-12 and reading leases. Hand that to the AI, keep the math in the model, and you get the speed without the risk. Westlake Realty Group cut per-deal underwriting from 6-8 hours of senior time to a 20-minute review, in its own model with every number traced to source. Read the Westlake Realty Group case study
What should you never paste into ChatGPT or Claude?
The prompts above only ask for documents your firm already holds and is free to use: the seller's OM, rent roll and T-12, your own leases, your own past deals, public records and the public research you receive.
- Data vendors' exports. Keep data vendors' exports out unless you have their written permission. Several vendors' terms bar putting their data into outside AI tools, and that includes exports you downloaded under your license. For comps, use your own closed deals, the comps in broker packages and public records.
- Personal and banking details. Investor personal information, bank account details and tenant personal data do not belong in a general chat tool. Take them out of a document before you paste it.
- Confidential deal documents on the wrong plan. Check whether your plan uses your inputs for training, where files are stored and who can see them, and use a business or enterprise workspace with training turned off for deal documents. Is AI safe for CRE deal documents
How should you check what a prompt gives back?
Treat every output as an analyst's first draft: useful, fast, and not yet approved. These rules keep a prompt library safe to use on live deals.
- Read the flags first. Start with everything marked NOT IN SOURCE, PM EXPLANATION NEEDED or [TEAM TO CONFIRM]. Those are the places the model told you it did not know.
- Spot-check figures against the source. Pick a handful of figures from each output, including the largest ones, and find them on the cited page or tab. If a cite is wrong once, check the whole output.
- Let the spreadsheet tie it out. Paste extracted tables into your model and let the formulas total them. If the model's total does not match the document's stated total, the extraction missed a row.
- Keep a named person on the sign-off. A person who knows the deal approves every number that goes into a memo, every letter that goes to investors and every email that goes to an owner or borrower.
- Version your prompts. Save each prompt with a date and change it in one place. When your template or buy box changes, update the prompt and re-test it on a closed deal.
When does a prompt library stop being enough?
A prompt library is the right start. It stops being enough when the same person pastes the same documents into the same prompts every day, and the answer still has to be copied into your model by hand. That is when firms we talk to ask for an agent built on their own model, templates and files instead.
- The output has to land in your model or template. A chat reply is text. An agent fills your own Excel model with formulas intact, or your own BOV or memo template, with every number linked to its source.
- Deals arrive faster than anyone can paste them. An agent reads the inbox or deal folder as documents arrive and runs the screen without anyone opening a chat.
- Several people need the same answer. A prompt in one person's chat history is not a team process. An agent applies the same buy box, guidelines and template for everyone.
- You need it to remember. An agent can answer from your past deals, memos and broker history, cited to your own files, instead of only what was pasted today.
- Have it built around your firm (ours). We build the agent on your own model, templates and files, then stay on as your AI team as they change. Saunders Real Estate cut senior time per broker opinion of value from 4-8 hours to about 30 minutes, drafted in its own template off its own comparables. Read the Saunders Real Estate case study
Common questions
- What are the best ChatGPT prompts for commercial real estate?
- Prompts that give the model a role, name the documents to paste, number the steps, fix the output format and require it to flag anything it can't source from the document. The highest-value ones extract rent rolls, T-12s and lease terms into your own formats and screen OMs against your written buy box.
- What are the best ChatGPT prompts for commercial real estate brokers?
- Prompts tied to work brokers repeat every week: a BOV narrative from your own comps, OM sections, a first-touch email to an owner, a reply to an inbound inquiry, an LOI compared with your terms, a pre-call owner brief and call notes turned into CRM fields. Each one in the Brokers and leasing group names its inputs, fixes the output and flags anything it can't source.
- Do these prompts work in Claude as well as ChatGPT?
- Yes. They are plain instructions with no tool-specific syntax, so they work in ChatGPT, Claude or any capable chat assistant. Use a business workspace with training turned off for deal documents.
- Can a ChatGPT prompt replace my underwriting model?
- No. It can do the reading: extract the rent roll, map the T-12, pull lease terms and lay out the broker's assumptions. It should not produce the final numbers. Paste what it extracts into your model, let the formulas calculate, and keep your team on the assumptions.
- How do I stop ChatGPT from making up numbers?
- Tell it to use only the material you pasted, cite the page for every figure, copy numbers exactly as written and mark anything missing as NOT IN SOURCE. Then spot-check the cites and let your spreadsheet total the extracted rows.
- Is it safe to paste an OM or rent roll into ChatGPT?
- It can be, on a business plan with training turned off and the right access settings. Remove personal and banking details first, and keep data vendors' exports out unless you have their written permission.
- Can I use AI prompts for comps?
- Yes, with comps you are free to use: your own closed deals, the comps in broker packages and public records. Ask the AI to normalize them into your comp template without converting rent bases, and keep the adjustments and the value conclusion in your model.
- When should a CRE firm move from prompts to an AI agent?
- When the same documents go through the same prompts every day, when the output has to land in your own model or template, or when several people need the same process. An agent built on your own files runs the work as documents arrive.