Real Estate Contract Review Prompts

Six prompts, run in sequence, will get you from raw purchase agreement to a marked-up redline memo in one sitting — here’s each one verbatim, plus where the chain breaks.

Real estate transaction work is document-heavy, deadline-sensitive, and full of state-specific landmines that look identical across deals until they aren’t. A Texas option period is not a Florida inspection contingency. A New York contract of sale runs on a completely different skeleton than a California CAR form. Most AI tools treat a purchase agreement like any other contract — which means a generic review prompt returns generic output. This six-prompt sequence is built differently: each prompt assumes the previous one ran, narrows the task, and pushes toward a concrete deliverable. I assembled these working across residential and light commercial purchase agreements on GPT-4o and Claude 3.5 Sonnet. They won’t replace your judgment on a complex 1031 exchange or a commercial sale-leaseback — but for routine review work, they will cut your first-pass time significantly.

How these prompts were chosen

The sequence mirrors the order a careful reviewer actually works: understand the deal structure first, flag problems second, extract deadlines third, audit financial terms fourth and fifth, then produce the output the client or opposing counsel actually receives. Each prompt is self-contained enough to run solo, but they compound — later prompts reference outputs from earlier ones. I tested each prompt on GPT-4o (ChatGPT Plus, May 2025 build) and Claude 3.5 Sonnet (Anthropic console, same period). Performance notes reflect both. Paste the full contract text before running Prompt 1; for Prompts 2–6, you can either keep the conversation going in the same thread or re-paste relevant sections as indicated in each prompt’s notes.

1. The contract structure summary prompt

Run this first. Its job is orientation: purchase price, closing date, financing structure, earnest money amount, and the names and roles of every party. You want this in a tight table or numbered list, not prose. The model needs the full contract text in the same conversation window before you send this prompt.

You are a real estate contract analyst. I have pasted a real estate purchase agreement above. Extract and present the following information in a structured list. Do not summarize or interpret — pull exact figures, dates, and names as written in the document.

1. Buyer name(s) and entity type (individual, LLC, trust, etc.)
2. Seller name(s) and entity type
3. Property address and legal description (if included)
4. Purchase price (exact figure)
5. Earnest money amount and due date
6. Financing type (cash, conventional, FHA, VA, seller financing, other)
7. Loan amount and lender name (if stated)
8. Closing date (exact date or formula, e.g., "30 days from execution")
9. Possession date (if different from closing)
10. All contingencies listed in the contract (name each one)
11. Option period or due diligence period: duration and termination fee (if any)
12. Any addenda or exhibits listed or attached

If any of these items are missing or ambiguous in the document, flag that item explicitly with the note [NOT FOUND] or [AMBIGUOUS — see section X].

What to expect: Both GPT-4o and Claude handle this cleanly on standard residential forms (TREC, CAR, FAR/BAR). On hand-drafted commercial agreements, expect occasional misreads of closing date formulas — double-check those manually. The [NOT FOUND] flag instruction is important: without it, models tend to infer or omit rather than surface gaps.

Adaptation note: For commercial deals, add item 13: “Due diligence deliverables and deadlines (title commitment, survey, environmental, zoning, estoppels).” That bucket is often absent from residential forms but central to commercial review.

2. The red flag scan prompt

This prompt asks the model to act as an adversarial reviewer — looking for what’s unusual, missing, or seller-favorable. Run it in the same thread immediately after Prompt 1. The model now has both the contract text and your structured summary as context.

Review the same purchase agreement. Identify and list every provision that:

(a) Departs from typical market-standard language for this contract type (note what the standard typically says and how this contract differs);
(b) Benefits the seller at unusual expense to the buyer — including broad seller-favorable representations, narrow warranty language, or limitation-of-liability clauses;
(c) Is missing relative to what a standard residential [or commercial] purchase agreement would normally include — specifically look for: inspection rights, title contingency, survey contingency, financing contingency, seller disclosure obligations, casualty/damage provisions before closing, and assignment restrictions;
(d) Contains an inspection waiver, as-is clause, or release of seller liability for property condition — quote the exact language;
(e) Uses vague or undefined terms that could create dispute at closing (e.g., "reasonable," "promptly," "satisfactory condition" without definition).

Format your output as a numbered list. For each item, include: (1) the section or paragraph number where the issue appears, (2) a one-sentence plain-language description of the issue, and (3) a risk rating of High / Medium / Low from the buyer's perspective.

Do not recommend legal action or provide legal advice. Flag the issues only.

What to expect: This is where Claude 3.5 Sonnet consistently outperformed GPT-4o in my testing — it caught more “missing provision” issues and was less likely to call something a red flag when it was actually market-standard in a given state. GPT-4o tends to over-flag on Texas TREC forms because the form’s option period mechanism looks unusual to a model trained on generic contract patterns. The risk ratings are useful for triaging your review, but treat “Low” flags as “still needs your eyes.”

State-specific note: In New York, the attorney approval contingency is standard and its absence genuinely is a red flag. In California, the CAR form’s liquidated damages clause and arbitration provision are both standard — a model may flag them unnecessarily. In Florida, an as-is rider is extremely common and not inherently problematic. Add a line to the prompt specifying the state so the model calibrates accordingly: “This is a [STATE] residential purchase agreement.”

Close-up detail shot of two hands resting near a keyboard, a blurred contract document visible as abstract cream shapes

3. The contingency timeline extraction prompt

Missed contingency deadlines are where real estate transactions collapse and malpractice claims are born. This prompt produces a chronological deadline list suitable for dropping into your docketing system. It works best when you give it a contract execution date — paste that into the prompt.

Using the purchase agreement already reviewed and the contract execution date of [INSERT DATE], extract every deadline and time-sensitive obligation in the contract. Present them in chronological order from execution to closing.

For each deadline include:
- The calendar date (calculate from execution date if the contract uses a day-count formula)
- The event or obligation due on that date
- Which party bears the obligation (Buyer, Seller, Both, or Third Party)
- The consequence of missing the deadline as stated in the contract (e.g., contract terminates, deposit forfeits, right waived)
- The section or paragraph reference

If the contract uses "business days," flag that explicitly and note that your calculation assumes Monday–Friday, excluding no holidays — the reviewing attorney should verify applicable holiday exclusions under state law.

Present the output as a table with columns: Date | Event | Party | Consequence of Miss | Section.

What to expect: Both models handle simple day-count math reliably. Where they stumble is nested conditions — e.g., “within 5 days of lender’s written commitment, which must be delivered within 21 days of execution.” Check every calculated date involving a conditional trigger. Also: the holiday caveat matters. Texas and Florida both have specific business-day definitions in their standard forms; California’s CAR form defines “days” differently than “business days” within the same document.

Commercial adaptation: Add a row type for “Recurring obligations” — rent rolls, tenant estoppel certificates, and SNDAs often have rolling delivery windows that don’t fit a single-date format. Prompt the model to flag these separately rather than force them into a single-date row.

4. The earnest money and escrow audit prompt

Earnest money terms are the ones clients argue about most when deals fall apart. This prompt isolates every provision touching the deposit: where it goes, who holds it, when it’s released, and what triggers a forfeiture or refund.

Focus exclusively on the earnest money deposit, escrow, and related financial provisions in this purchase agreement. Extract and analyze the following:

1. Exact earnest money amount and any additional deposits required (second deposit, option fee, etc.)
2. Escrow holder identity (title company, attorney, broker, other) and their state of licensure if stated
3. Deadline for deposit delivery and the mechanics of delivery (wire, check, etc.)
4. Conditions under which Buyer receives a full refund of earnest money — list each one verbatim
5. Conditions under which Seller retains earnest money as liquidated damages — list each one verbatim
6. Whether the contract specifies liquidated damages as the sole remedy or whether Seller retains the right to sue for specific performance in addition to retaining the deposit
7. Any provisions addressing dispute resolution between Buyer and Seller over earnest money release (interpleader, mediation, etc.)
8. Any interest earned on the deposit — who receives it?
9. Any provisions for additional deposits triggered by contingency removals or extensions

Flag any provision where the refund and forfeiture conditions appear to conflict or leave a gap — quote the conflicting language directly.

What to expect: This prompt reliably pulls the right clauses. The conflict-flagging instruction at the end is doing real work — I’ve seen models surface genuine ambiguities in hand-drafted contracts that a fast first read would miss. In New York, the contract of sale typically holds the down payment (often 10%) in the seller’s attorney’s escrow account — the model may not know that’s unusual relative to other states unless you tell it. In Texas, earnest money held by a title company and option fees paid directly to the seller are separate instruments; the model handles this correctly on TREC forms but occasionally conflates them on hand-drafted deals.

5. The closing-cost allocation review prompt

Who pays what at closing varies by state, by deal, and by negotiation — and the contract often allocates costs in scattered provisions rather than one tidy section. This prompt consolidates the full cost picture.

Review the purchase agreement for all provisions addressing closing costs, fees, and expenses. For each cost item below, identify which party pays (Buyer, Seller, Split, or Not Addressed) and cite the specific section of the contract. If the contract is silent on an item, note [CONTRACT SILENT] — do not assume a default.

Cost items to check:
- Owner's title insurance premium
- Lender's title insurance premium
- Title search / title examination fee
- Survey (new survey vs. existing)
- Transfer tax / documentary stamp tax
- Recording fees (deed, mortgage)
- Escrow / closing fee
- Real estate broker commission
- HOA transfer fee and/or estoppel letter fee
- Prorations: property taxes, HOA dues, rent (if tenant-occupied)
- Home warranty (if any)
- Repair credits or seller concessions
- Loan origination fees or buyer closing-cost credits from seller
- Attorney fees

After the cost allocation table, add a short paragraph flagging any cost item where the allocation is unusual relative to standard practice in [INSERT STATE], or where the contract language is vague enough to create a dispute at the closing table.

What to expect: The [INSERT STATE] placeholder is doing significant work here. Transfer tax is split by custom in New York but paid entirely by the seller in many Florida counties; documentary stamp taxes in Florida follow a specific statutory formula the model knows but sometimes misapplies to commercial deals. California’s county transfer tax structure differs from its city transfer tax structure in cities like Los Angeles. Tell the model the state and county for best results. The “CONTRACT SILENT” flag instruction is critical — without it, the model will silently apply a default assumption.

Commercial adaptation: Add to the cost list: environmental inspection costs, zoning verification fees, tenant notification costs, and assumption fees if the buyer is assuming an existing loan. Commercial deals also commonly include a separate “prorations methodology” section — prompt the model to locate and summarize it.

6. The markup-and-redline rationale memo prompt

This is the deliverable prompt — the one that produces something you can actually send to a client or use to brief a junior attorney on what changes to request. Run it after all five prior prompts have run in the same thread. It synthesizes the prior outputs into a structured memo with recommended changes and the reasoning behind each.

Based on your full review of this purchase agreement — including the structure summary, red flag scan, contingency timeline, earnest money audit, and closing cost allocation — draft a redline rationale memo addressed to the reviewing attorney (not to the client).

The memo should:
1. Open with a one-paragraph deal summary: parties, property, purchase price, closing date, and transaction type (residential / commercial, cash / financed).
2. List every recommended contract change in priority order (High priority first), using this format for each item:
   - Issue: [plain-language description]
   - Location: [section or paragraph number]
   - Recommended change: [what the contract should say instead, in plain language — do not draft replacement contract language]
   - Rationale: [why this change protects the buyer / corrects a gap / reflects market standard]
3. Note any provisions that appear unusual but are acceptable without change — briefly explain why.
4. Close with a "Watch list" section: three to five items the reviewing attorney should monitor through closing (e.g., contingency deadlines, seller-side document deliveries, title commitment timing).

Format: memo style, not bullet-only. Write in plain professional English. Do not provide legal advice, legal strategy, or predictions about litigation outcomes. Flag issues and describe changes only.

Label the memo: DRAFT — FOR ATTORNEY REVIEW ONLY. NOT LEGAL ADVICE.

What to expect: This is the most variable prompt in the set. Claude 3.5 Sonnet produces tighter, better-organized memos; GPT-4o tends toward wordier output that needs trimming. Both will occasionally include a “recommended change” that is actually legal strategy — flag these and remove them before the memo leaves your desk. The “do not draft replacement contract language” instruction keeps the output from drifting into practicing-law territory, but review everything before you use it. The DRAFT label is there to remind you — and anyone who sees the file — that this output is a starting point, not a finished work product.

Commercial adaptation: Add to item 2’s priority list: representation and warranty survival periods, indemnification caps, and material adverse change definitions — these are largely absent from residential forms but central to commercial purchase agreements.

Notes on using these prompts

Model choice: Run this sequence on Claude 3.5 Sonnet if you have access. It handles long contract documents with less drift and catches more missing-provision issues than GPT-4o in my testing. If you’re using ChatGPT Plus with GPT-4o, the sequence still works — but budget time to tighten Prompt 6’s output manually.

Context window limits: A standard residential purchase agreement with addenda typically runs 15–30 pages. Both Claude and GPT-4o handle this comfortably in one thread. A heavily negotiated commercial deal with exhibits can push past 80 pages — at that length, consider splitting the document and running the red flag scan (Prompt 2) per exhibit rather than on the full document at once.

State-specific variables: Texas TREC forms are highly standardized and the models know them well; hand-drafted Texas commercial deals are not. Florida FAR/BAR forms perform well. New York contracts of sale — which are attorney-drafted by custom and vary widely — produce the most variable model output of the four; plan for more manual review on NY deals. California CAR forms are well-represented in training data, but California’s local transfer tax patchwork (Los Angeles, San Francisco, and Santa Monica all differ) requires you to supply county-level context explicitly.

Where the sequence breaks: These prompts work on contracts in English, in standard legal formatting, pasted as clean text. Scanned PDFs converted to text via OCR produce errors the model will reproduce without flagging. Run your OCR output through a quick visual check before pasting. The sequence also does not handle title commitment review, survey analysis, or lien search output — those are separate document types that need their own prompt sets.

Verification is not optional: Every extracted date, dollar figure, and party name from these prompts should be spot-checked against the source document before it goes into a client file or a docketing system. The models are accurate most of the time on structured data — but “most of the time” is not a standard for a closing deadline.

The six-prompt sequence won’t replace a careful attorney read on a complex deal. What it does is front-load the mechanical extraction work so your careful read focuses on judgment calls, not on hunting through boilerplate for the earnest money release conditions. That’s the trade worth making.

The 6-Prompt Sequence for Reviewing Real Estate Contracts