Every file starts the same way: an executed contract lands, and someone has to read all of it, pull the terms into a checklist, and calendar every deadline before anything else can move. That first-pass extraction is slow, repetitive, and exactly what AI is good at. The trap is that the same tool that builds you a clean term sheet in a minute will also tell you the title looks marketable or quietly miscalculate an inspection deadline, and the legal judgment and the signature are the attorney's, not the model's. This is a plain look at where AI genuinely helps with contract review, where it does not, and a set of prompts you can paste in and test on your own real contracts.
The honest picture
AI is genuinely useful for the reading and extraction, the part a paralegal does the same way on every file:
- What AI does well today: read an executed purchase contract and addenda, pull the key terms into a structured checklist (parties, legal description, price, deposit and escrow, contingency and financing deadlines, closing date, fee allocation), flag blank fields and internal inconsistencies, and lay the dates out so a human can build the calendar. It compresses the intake, not the judgment.
- What AI does not do: practice law. It cannot opine on whether title is marketable, cannot decide how to resolve a defect, cannot advise the client, and cannot render the attorney's opinion or sign the closing documents. In Florida the legal judgment and the signature belong to the licensed attorney. AI builds the first-pass checklist; the attorney owns every legal call and the file.
The right way to think about it: AI is a fast reader that organizes what the contract says, not a lawyer. The term sheet is a draft to verify. The opinion and the signature are yours.
The two things it will get wrong if you trust it
Two failures show up on almost every contract, and both are caught only by a human who reads against the document:
- It will reason from memory instead of the page. Ask it about a standard contingency period and it may answer with the form's usual number rather than what your executed contract actually says, which may be struck, amended, or blank. The fix is to make it read only this document and cite the clause for every term, and to make it say "not specified in this contract" rather than fill in a default.
- It will get the date math wrong. Computing "the inspection deadline is 15 days after the effective date, counting business days, excluding the closing date" is exactly the kind of step-by-step counting language models are unreliable at. Never take a computed deadline on faith. Make it show the base date, the day count, and the counting rule it used so you can check the arithmetic yourself, and treat every date as proposed until a human confirms it.
The setup that keeps the judgment yours
Two habits make AI safe for intake, and the prompts below build them in:
- One document, cite every term. Feed it the executed contract and addenda and tell it to use only what is on the pages and to cite the clause behind every extracted value. A model told to cite invents far less than one asked to summarize from memory.
- Extract and flag, never opine. Tell it to pull terms, flag blanks and inconsistencies, and surface deadlines, but never to give a legal conclusion, judge marketability, or advise. You want a checklist and a flag list, not an opinion.
How to test it on your own work
Do not trust a polished demo, including this one. Pull two or three of your own recent executed contracts, a timer, and the prompts below. Rate each output 1 to 5 on usefulness and accuracy, and compare the time against how your team does intake today. Keep what wins. Use your firm's enterprise or no-train settings before you paste any client document, and confirm you are comfortable with that. This is a workflow guide, not legal advice.
Paste-ready prompts
This guide is about workflow, not law. It does not give legal advice and does not replace a licensed attorney's review. Copy these as written. Bracketed text is what you swap per file.
Test 1: Build the term sheet (text model)
I am attaching an executed residential purchase contract and its addenda.
Extract a term-sheet checklist and read only what is on these documents.
Include: buyer and seller names, property legal description, purchase price,
deposit/escrow amount and holder, financing terms, each contingency with its
stated period, the closing date, and how fees are allocated.
Rules:
- Use only what the documents state. For each item, cite the page and clause.
- If a field is blank, struck, or not specified, write "not specified in this
contract." Do not fill in a standard or default value.
- Do not give any legal conclusion or opinion. Extract and cite only.
Documents: [attach or paste the contract and addenda]
Watch for: did it cite a clause for each term, and did it correctly mark the blanks instead of filling them with form defaults?
Test 2: Lay out the deadlines (text model)
From the same contract, list every date-driven deadline (effective date,
contingency deadlines, financing deadline, inspection period, closing date).
For each one, show your work so I can check it:
- The base date you started from and where the contract states it.
- The number of days and whether the contract says calendar or business days.
- The counting rule the contract specifies (for example whether the start date
is counted, how weekends or holidays are handled).
- The resulting date.
If the contract does not clearly state a base date, day count, or counting
rule, say so and do not compute the date. Do not assume a standard rule.
End with this exact note: "These dates are computed by an AI and are not
reliable. Verify every date by hand against the contract before calendaring."
Watch for: this is the high-risk test. Check every computed date by hand. Did it use the contract's own counting rule, and did it refuse to compute when the contract was ambiguous rather than guessing?
Test 3: Internal consistency check (text model)
Review this executed contract for internal problems only. For each issue, quote
the exact text and the page. Flag:
1. Conflicts: the same term stated two different ways (for example a price or
date that does not match between the main form and an addendum).
2. Blanks and missing signatures or initials the document itself calls for.
3. Impossible dates: any deadline that falls after the closing date or before
the effective date.
Do not give legal advice and do not resolve the issue. Only flag it for the
attorney.
Contract: [attach or paste it]
Watch for: does it catch the addendum that contradicts the main form, and does it stay on flagging rather than telling you how to fix it?
What success looks like, and where it could go
If your own testing shows real time savings, the next step is a small pilot: run a week of new files through the prompts and measure the intake hours. If that holds up, the natural next step is a simple agent, running on your firm's own cloud, that you use in plain language. The most useful version takes the executed contract PDF and returns the term sheet plus a deadline calendar with every value linked back to the exact page and clause, and the date math shown for each deadline, so the attorney verifies and the file opens in minutes instead of an afternoon. It extracts and cites; the attorney opines and signs.
The principle holds the whole way through: AI gives you a faster first pass and a second set of eyes on consistency. It does not render the opinion of title, and it does not sign the file. Keep that line clear and the rest is upside.
Want a straight answer for your firm?
I build practical AI and custom software for businesses, on Google Cloud. If you want a second set of eyes on how AI could fit your intake and closing workflow, or on a tool you are considering buying, tell me what you are working with. No pitch, just a straight answer.