Yes, as a first pass: AI software can summarise a subcontract, flag unusual terms and compare it with your usual terms in minutes, but it can miss what is absent and misread amendments. Read payment terms, retention, liquidated damages and indemnity yourself, line by line, and have a solicitor sign off before you sign.
The danger is not that the software is useless. It is that it reads fluently, so a wrong answer sounds as sure as a right one. A tool that tells you "retention is 5%, which is standard" when a schedule at the back raises it to 10% has made your decision worse than silence would have.
The way to find out how far to trust a tool is to run it on a contract whose problems you already know, then on a copy with changes you planted, and to score both against an answer key you wrote first. Four clause types stay with you whatever the result.
What review software can and cannot do on a subcontract
On the plus side, software reads 80 pages faster than you will, applies the same checklist every time, pulls defined terms into one place, and can compare notice periods across documents. Those are real gains when a main contractor sends a 60-page subcontract with a 20-page schedule of amendments on a Friday afternoon.
The weaknesses are consistent across tools, and they are where a builder's money is lost:
- Absent clauses. A subcontract with no payment notice provision gives the tool nothing to flag. It summarises what is there, not what should be.
- Amendments at the back. The particular conditions, schedules and appendices often override the standard form. A tool that weights the main body most heavily reports the wrong rule.
- Numbers. Percentages, day counts and rates in tables or on scanned pages are easy to misread, and one wrong digit changes the deal. OpenAI's file-upload FAQ does not say whether ChatGPT reads image-only scans, so check before you rely on one.
- Cross-references. "Subject to clause 14.3 as amended by Schedule 4" is easy for a person to follow and easy for a model to flatten.
- False reassurance. "Standard" and "market" are judgements the tool cannot make for your trade and your size of firm.
- Limits. Claude reads PDFs of up to 1,000 pages, with visual layout analysed only up to 100 pages. A long subcontract with drawings may need splitting.
None of this makes the tools useless. It tells you what to test: absence, amendments, numbers and cross-references.
Three kinds of tool a builder can test
Prices and features below were read from vendor pages in October 2026. Everything a vendor says about itself is a vendor claim and unverified.
| Option | What it is | Price as published | Question to ask or test |
|---|---|---|---|
| General assistant (ChatGPT, Claude, Gemini) | Upload a PDF, give it your checklist, get clause-by-clause answers | About $20 a month on an individual plan; business plans about $20 a seat on annual billing, $25 monthly | Does it stay accurate on numbers and amendments when you supply the checklist? |
| Document Crunch | Describes itself as an AI risk intelligence platform built for construction; its page lists pursuit-stage redlining and version comparison | No prices on its page; a free trial is offered | Does it review subcontracts specifically? Its page does not say. Does it train on your documents? Not stated |
| Lexilio | Construction contract analysis: uploaded contract suites, obligations, notice tracking and drafting; aimed at contractors' commercial teams | Starter $29 a month (1 seat, 1 active project, 25 contracts); Professional $299 a month (10 seats); Enterprise $1,999 a month; 7-day free trial, card required, and the plan starts unless you cancel | Its pricing page lists cross-document conflict detection under Enterprise only. Which of the features you saw on the homepage are in your tier? |
| Mastt | Capital project management software for project owners, developers and their representatives; its pricing page lists AI contract review and AI payment review on the Professional plan, and the plan card describes them as autonomous AI review of invoices and payments | Free Starter plan; Professional $165 a month, shown as an average cost per project; Enterprise and the AI Agent priced on request | Built for the owner's side: does its contract review read a subcontract's clauses, or check invoices and payments against a contract? |
Lexilio's homepage also says it never trains AI on customer data. That is a vendor claim, and the useful move is to ask for it in the data-processing terms rather than rely on a footer. For a longer list of demo questions, see ten questions for contractors at an AI contract software demo.
The old-contract test: build the answer key first
Choose a finished subcontract where you know how the job went. The best candidates are contracts that cost you something: a late payment, a retention release that dragged, a deduction you disputed. Before you run any tool, write down what a good review would have told you at signing. That list is your answer key, and writing it first stops you grading the tool by what it happens to find.
Here is a key for an $84,000 groundworks subcontract:
| Topic | What the contract said (illustrative) | What it cost on the job |
|---|---|---|
| Payment | Pay when paid, 14 days after the main contractor receives payment | Paid 68 days after our valuation |
| Retention | 5% held; half at practical completion, half 12 months later | Second half arrived 16 months after we left |
| Retention, schedule | Particular conditions raise retention to 10% | Not noticed until the first valuation |
| Liquidated damages | $750 a day, no cap, tied to the main contract's completion date | Charged 9 days ($6,750) for another trade's delay |
| Indemnity | Against "all losses howsoever arising" | Never tested, but uninsured exposure |
| Time bar | Delay claims must be notified within 7 days | Missed by two days; claim refused |
| Set-off | Right to deduct sums due on any other contract | Deduction for an unrelated job |
| Termination | For convenience on 7 days' notice, payment for work done only | Not used, but a cash risk |
Eight issues is plenty. Run each tool with the same prompt (a good one is below), and mark every issue found correctly, found partly, missed, or stated wrongly. Wrong statements matter most: a miss is expected, which is why you read the money clauses yourself, but a wrong figure creates confidence you should not have.
The planted-changes test
An old contract can only test what you already know. Planting changes tests what the tool can find in a document it has never seen. Make a copy of the old subcontract, change five things with the changes tracked in a separate note, and run the copy through each tool. Plants and results might look like this.
| Planted change | Tool A | Tool B | Tool C |
|---|---|---|---|
| Retention raised from 5% to 10% in a schedule | Missed | Caught | Missed |
| Cap on liquidated damages deleted | Caught | Caught | Caught |
| Payment term lengthened from 30 to 60 days in the particular conditions | Caught | Caught | Missed |
| A new 7-day time bar for claims added | Missed | Caught | Missed |
| A cross-reference pointed at a clause that does not exist | Missed | Missed | Missed |
| Caught | 2 of 5 | 4 of 5 | 1 of 5 |
Combine both tests. In a sample run, Tool A found six of the eight known issues but stated retention wrongly, Tool B found seven with no wrong statements, and Tool C found four. Tool B is the only one worth testing further. Tool A's wrong statement is exactly the retention error described at the top, so it fails the money-clause test however many other issues it found. Tool C is summarising, not reviewing. Notice also that nobody caught the broken cross-reference, which is why you read cross-references and the four clauses below yourself.
Turn the answer key into a playbook any tool can run
Once you know what hurt you, write it as rules. Construction-specific tools call these playbooks; Lexilio's Starter plan, for example, lists five playbooks of ten rules each, and Document Crunch's page also mentions running playbooks. A general assistant can follow the same rules if you paste them in. The table shows a ten-rule playbook built from the answer key. The preferred positions are starting points to agree with your solicitor, not market norms.
| Rule | Flag if | Preferred position | Fallback |
|---|---|---|---|
| 1. Payment condition | Payment depends on the main contractor paying, with no long-stop date | Pay within 14 to 30 days of valuation | Long-stop of 60 days |
| 2. Payment notices | Notice steps are not stated | A written timetable for applications and notices | Agreed in the order of precedence |
| 3. Retention rate | Above 5%, including in schedules | 5% | 5% with a cap |
| 4. Retention release | Tied to the main contract's end | Half at your completion, half after the defects period | A long-stop date for release |
| 5. Retention security | Not held separately | Held apart, or replaced by a bond | A written statement of where it is held |
| 6. Damages trigger | Tied to the main contract's completion date | Tied to your own late completion | Pro-rata to your share of the works |
| 7. Damages cap | No cap | A cap, such as 5% of the subcontract sum | 10% |
| 8. Indemnity scope | "All losses" or the other party's own negligence | Losses caused by your breach or negligence | Capped at the contract value or your insurance limit |
| 9. Time bars | Notice periods under 14 days | 14 days or more | 10 days with a written reminder |
| 10. Set-off | Deductions allowed for other contracts | Only sums due under this subcontract | Written notice before any deduction |
Applied to the $84,000 subcontract, the playbook turns the answer key into asks. A 5% cap on damages would be $4,200 against the $6,750 actually charged for nine days of someone else's delay, a difference of $2,550. Holding retention at 5% instead of 10% keeps $4,200 of your money. A 60-day payment long-stop would have shortened the 68-day wait by eight days. None of these is guaranteed to be accepted, but each is a specific request rather than a vague unease.
A prompt that makes an assistant useful rather than confident
A general assistant does far better with a structured prompt that forces quotes and allows "not found".
You are helping a small builder review a subcontract before signing. I am the
subcontractor. Read ALL uploaded documents, including schedules and particular
conditions. Where documents conflict, say so; do not choose silently.
For each topic below give: the clause number; a quote of the key words (25 words
maximum); what it means in plain English; a risk rating (low, medium, high) with
one line on why. If a topic is not covered, write NOT FOUND. Do not assume.
Topics: payment (due dates, conditions, notices, set-off); retention (percentage,
cap, release dates, separate account); liquidated damages (rate, cap, trigger,
extension of time); indemnity (scope, cap, insurance); notices and time bars;
variations; termination; defects period; insurance; flow-down from the main contract.
End with: conflicts between documents, and questions I should ask the other party.
Do not give legal advice.Then verify the output the boring way. Take each quoted phrase and search for it in the PDF. A quote that cannot be found has been invented or reworded, and everything near it is suspect. Check every number against the page it came from. A good result lists retention as "5% in clause 11, rising to 10% in Schedule 4, conflict flagged", rates it high, and asks whether the second half can be released at practical completion. A bad one says "retention 5%, standard" and moves on. If the assistant writes "NOT FOUND" for payment notices, that is useful information about the contract and a prompt for your solicitor.
A realistic misread, and how it shows up
A typical slip: the summary says retention is 6%, but the first valuation deducts 5%. The cause is a scanned page in the middle of the PDF, where the tool read a 5 as a 6. It shows up as a mismatch between the summary and the money, usually weeks later. Two habits prevent it. Ask for the exact sentence and find it on the page, and where a clause matters, rerun the tool on the native Word file if the other party can send one. A scanned page is also the place to look first when a number in a summary seems odd.
The four clauses you read yourself
Use the software to point you at these. Do not let it summarise them for you.
Payment terms
Look for the due date, whether payment is conditional on the main contractor being paid, the notices and time limits that must be met, rights to set off, interest, and how the final account is agreed. Take monthly valuations of $14,000. If payment depends on the main contractor being paid, say 45 days after your valuation plus 7 to pass it on, that is about 52 days, and you are funding roughly $24,300 of work at any time ($14,000 x 52 / 30). Where AI goes wrong: it treats "pay when paid" as boilerplate and misses a notice that must be served before you are entitled to payment at all. The tutorial on payment notice templates and deadlines covers that side.
Retention
Look for the percentage, any cap, when each portion is released (practical completion, which some contracts call substantial completion, is the usual first trigger), whether it is held separately, and what replaces it (a bond or similar). On the $84,000 job, 5% is $4,200; at 10% it is $8,400, which would equal the whole profit on a job at a 10% margin. Check whether the second half waits for the main contract's end rather than yours, and whether the contract says the money is held apart from the contractor's own funds. Where AI goes wrong: schedules and amendments that change the percentage. Tracking retention and chasing release with reminders picks up after you sign.
Liquidated damages
Look for the daily or weekly rate, a cap, what triggers it (your delay, or any delay to the whole works), and the extension-of-time process, including the notice period. Nine days at $750 is $6,750, or 8% of an $84,000 sum. With no cap, and a trigger tied to the main contract's date rather than your own work, you can pay for another trade's delay. Where AI goes wrong: it reports the rate and skips the trigger and the notice period, which decide whether you can claim time.
Indemnity
Look for what you are indemnifying (third-party claims, or "all losses"), whether it covers the other party's own negligence, any cap, and the insurance you must hold. Many insurance policies limit cover for liability you take on only by contract, so ask your broker to read the clause before you sign. Where AI goes wrong: it describes the indemnity as "mutual" or "standard" when the wording is one-way. Quote the exact words to your solicitor.
Other clauses to read, in a shorter pass
- Notices and time bars. Any "within 7 days or lose the claim" wording. Put each date in your diary system.
- Variations. Who may instruct them, in what form, and how they are priced. Turning site messages into change orders shows the paperwork.
- Termination. Whether the other party can end it for convenience and what you are paid.
- Set-off and back-charges. Whether they can deduct for other contracts.
- Insurance. The limits and types required, and whether you have them.
- Flow-down. Whether you are bound by a main contract you have not seen. Ask for it.
- Order of precedence. Which document wins when two conflict.
- Defects period. How long, and what the retention release depends on.
A routine to run before you sign anything
- Confirm you have every document: the form, particular conditions, all schedules and, if it is referred to, the main contract.
- Check the confidentiality clause before uploading anything.
- Run the structured prompt and your playbook.
- Search the PDF for every quoted phrase, and check every number against its page.
- Read the four money clauses and the cross-references yourself, in full.
- Put every notice date and time bar in your diary system, with a reminder.
- Send the one-page review note to your solicitor, and ask the other party the questions the tool raised.
- Save the version you signed, the review note and the answers you received in the job file.
Confidentiality, privacy and where to upload
Before uploading, read the contract's own confidentiality clause. Many subcontracts forbid sharing the terms with third parties, and a cloud tool is a third party. If it does, ask the other side for consent, redact names, or use a tool under a data-processing agreement. Consumer plans of ChatGPT and Claude let you switch off the use of chats for model training, while business plans such as ChatGPT Business and Claude Team do not train on your content by default. Document Crunch's page says "your contracts stay yours" and cites SOC 2, but does not say whether customer data trains its models, while Lexilio's footer says it never trains AI on customer data. For a longer view, see whether chat-with-PDF tools are safe for contracts and client files.
Handing the flagged clauses to a solicitor
The point of the AI pass is to make the lawyer's time cheaper and better aimed, not to avoid it. Give them a one-page list rather than a 70-page PDF and a vague worry.
SUBCONTRACT REVIEW NOTE
Job: [description]. Other party: [name]. Value: $[x]. Programme: [dates].
Documents: subcontract form, particular conditions, Schedules 1-5, main contract (not seen / seen).
HIGH RISK, clause: payment, 9.2 (pay when paid). Question: notice needed to claim?
HIGH RISK, clause: retention, 11 and Schedule 4 (5% rising to 10%). Question: separate account?
HIGH RISK, clause: liquidated damages, 14.1 (no cap; tied to main completion). Question: cap?
HIGH RISK, clause: indemnity, 17 ("all losses howsoever arising"). Question: insurance fit?
NOT FOUND: payment notice provisions. Cross-reference problem: clause 20.3 refers to 20.9.
Conflicts between documents: [list]. Questions I want answered by the other party: [list].A solicitor or construction lawyer can then confirm, rewrite or advise you to negotiate. Ask for a fixed fee, and keep the note and the contract together in the job file so the next subcontract starts with a list of terms you have already pushed back on.
Contracts you issue: the same test in reverse
If you are the main contractor sending subcontracts to trades, the test runs the other way. Your standard terms should come from a solicitor, not from an AI draft. What AI does well is compare a trade's marked-up version against your template: paste both in and ask for every difference, grouped by clause. Plant five changes in a copy of your own template, such as a payment term, a retention rate, a cap on damages, a notice period and an insurance level, and see whether the tool finds them. A tool that misses two of five on a document it can compare line by line is not ready for the harder job. Then keep your template, the trade's mark-up and the differences list together, so the lawyer sees only what changed.
What it costs against what it saves
A builder who reviews three subcontracts a month, with an AI pass that trims 30 minutes from each because the money clauses still need a personal read, saves 90 minutes a month, about $90 at $60 an hour. A general assistant at about $20 a month, or Lexilio's Starter at $29, is covered by the time alone. Lexilio's Professional plan at $299 would need about five hours of saving a month, which three contracts cannot give, so it makes sense only for a firm reviewing many. The real return is catching one uncapped damages clause or one hidden 10% retention, either of which costs more than a year of subscriptions.
Whichever route you take, finish the same way: run the old-contract test, plant the changes, score the tools, read the four clauses yourself, send the note to a solicitor, and keep the results so you can rerun the test when the product changes. For firms that issue subcontracts as well as sign them, AI tools for contracting businesses covering subcontracts and change orders covers the other direction.
Questions builders ask about AI contract review
Can AI replace a solicitor for a subcontract?
No. AI can speed up your first read, point you to clauses and draft questions, but it can miss what is absent, misread amendments and sound certain when wrong. A solicitor or construction lawyer should review anything you sign, and your issues list from the AI can shorten their time.
Is it safe to upload a subcontract to an AI tool?
Check the contract's confidentiality clause first, because uploading to a third party may breach it. Then use a business plan or a tool whose terms say it does not train on your data, and redact names where you can. Consumer chat plans have a privacy switch, but business plans do not train by default.
How long does the old-contract test take?
Plan on about two to three hours for the first tool and an hour for each further one: an hour building the answer key from a contract you know, then 30 to 60 minutes running the same prompt and scoring the result. It is cheaper than finding out on a live subcontract.
Which tool should a small builder start with?
A general assistant on a business or paid plan with a proper prompt is the cheapest start, at about $20 a seat a month. Trial a construction-specific tool only if you review several subcontracts a month, and score it on the same old contract.
Can AI draft a better subcontract for me to issue?
It can tidy wording and flag gaps against your checklist, but your own standard terms should come from a solicitor. Use AI to compare incoming subcontractor terms with your template, and have the lawyer approve any change to the template itself.
Further reads
- AI Contract Review for Small Firms: What It Catches and Misses — What AI contract review catches and misses for small firms.
- AI in Contracting: Subcontractor Paperwork Chasing Checklist — A checklist for chasing subcontractor paperwork.
- AI Software Contracts: Auto-Renewals, Price Rises, Notice Periods — Read the review tool's own contract: renewals and notice.
- AI for Building Contractors: Chasing Subcontractors and Progress Updates — Chasing subcontractors for progress and paperwork.
- AI Solutions for Construction Companies: One Platform or Point Tools? — One construction platform or several point tools? Four tests (jobs, integrations, data ownership, upkeep) and two stacks priced from vendor pages.
- Best AI App for General Contractors: Site and Office Tools Compared — Site and office AI tools priced per month for a six-person firm, each with a worked example, plus four stacks from $189 a month.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Document Crunch home page, Lexilio home and pricing pages, Mastt home and pricing pages; Claude, ChatGPT and Google plan pages; Anthropic and OpenAI help articles on file limits and data use. Features and prices read in October 2026; all vendor claims are unverified.