For a small firm that wants contract review inside Word, trial Spellbook and Gavel Exec side by side and use Copilot in Word as the baseline. Spellbook is priced by quote and offers a 7-day trial; Gavel Exec lists $160 per user per month. Run all three on one contract with planted errors and choose on what each catches.
One caveat shapes the whole test: only Gavel Exec publishes a price, so work out the three-lawyer cost before any sales call, then judge on narrower questions than a feature list. Which tool catches the problems in the contracts your firm actually reviews, and what happens to a client's draft once it leaves Word? Both specialist tools are add-ins that mark up a contract with tracked changes; the difference is in the details below.
What each tool is, in plain terms
Spellbook
Spellbook describes itself as a Word add-in suite with Review, Draft, Ask, Benchmarks and Playbooks, plus Associate, which it presents as a multi-document agent that works through a matter with your oversight. Its pricing page says pricing is determined by the number of team members on a licence and that you book a demo for a custom quote. It advertises a 7-day free trial, with longer trials possible for larger organisations. On privacy, it states that it has Zero Data Retention agreements, meaning data is not used for training, and SOC 2 Type II compliance. There is no price list to compare, so that part of the decision happens in a conversation. If you have a quote in hand, how a small firm should read a Spellbook quote shows which lines to question.
Gavel Exec
Gavel Exec runs as a Microsoft Word add-in for Windows, Mac and Word on the web, and as a web app. It reads a contract against your playbook or built-in rules and returns redlines as tracked changes with the reasoning behind each, plus comments, missing-term lists and summaries. Its product page lists the price: 25 free queries per user with no card, then $160 per user per month or $1,740 per user per year, with separate team pricing from 10 seats. It states SOC 2 Type II, Zero Data Retention agreements with its AI providers, and no training on customer data. It is a different product from Gavel's older document-automation plans, which are priced separately, so make sure you are reading the Exec page and not the general pricing page.
Copilot in Word
Microsoft's own documentation describes access to Copilot in Word in tiers: a paid add-on licence, standard access, and Copilot Chat. Microsoft 365 Copilot Business is listed at $21 per user per month on an annual commitment for organisations up to 300 users ($25.20 if that commitment is billed monthly; a promotional $18 on annual billing runs to 31 December 2026), and the enterprise version at $30 per user per month on top of a base licence. The app has also been renamed Microsoft Copilot in places, so expect both names on the vendor's pages. Copilot is a general assistant inside Word. It will not arrive configured with your negotiation positions, so what it can do is limited by how well you prompt it and what you give it.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowSide by side for a three-lawyer firm
| Question | Spellbook | Gavel Exec | Copilot in Word |
|---|---|---|---|
| Published price | No; quote based on seats | $160 per user a month, or $1,740 a year | $21 per user a month (Business, annual) for the paid add-on |
| Free way in | 7-day trial | 25 free queries per user | Copilot Chat with a work account; check what your plan includes |
| Where it works | Word add-in | Word add-in and web app | Word, Outlook and other Microsoft apps |
| Playbooks and rules | Playbooks, benchmarks | Playbooks you write or generate from your own checklists and signed agreements | None built in; you paste your rules into a prompt |
| Redline output | Tracked changes in Word | Priority-ranked redlines as tracked changes, with reasoning | Suggestions and rewrites; check tracked-change behaviour on your version |
| Data statement | Zero Data Retention agreements, SOC 2 Type II | Zero Data Retention with AI providers, SOC 2 Type II, no training on customer data | Governed by your Microsoft 365 agreement and settings |
Do the sums for the firm before the demo, so the demo cannot move the number. For three lawyers, Gavel Exec at the annual rate is 3 x $1,740 = $5,220 a year; at the monthly rate it is 3 x $160 x 12 = $5,760, so paying monthly costs $540 more. Copilot Business for three is 3 x $21 x 12 = $756 a year, if the firm has a qualifying Microsoft 365 plan. Spellbook sits wherever its quote lands. If the quote for three seats is higher than Gavel's annual figure, ask what the extra pays for, such as training sessions or the multi-document agent.
Whether any of these beats a general chatbot is a separate question that the ChatGPT versus legal AI tool comparison covers.
Build the test contract once, use it three times
A demo on the vendor's own sample contract proves little. Build one contract you control, then run it through every tool on the same day. Take a supplier or services agreement template your firm already owns and edit it to plant eight problems. Keep an answer key in a separate file.
- Liability cap with a carve-out that swallows it. The cap is "fees paid in the last 12 months", then a later clause removes the cap for "any breach of confidentiality or data obligations".
- Auto-renewal with a 90-day notice window buried in the term clause, while the summary page says "12 months".
- One-way indemnity where only the customer indemnifies the supplier.
- Payment terms conflict: 30 days in the body, 60 days in the schedule.
- IP assignment that sweeps in pre-existing tools ("all materials used in connection with the services").
- Defined term used but not defined ("Deliverables" appears in four clauses and nowhere in the definitions).
- Termination for convenience for the supplier only, with no refund of prepaid fees.
- Non-solicitation of staff lasting 24 months, far beyond the rest of the agreement.
Add two clean clauses that look suspicious but are fine, such as a standard confidentiality carve-out for legally required disclosure. They test false positives, which waste your time as surely as a miss does. If you want a ready structure for scoring any legal AI trial, the two-week trial scorecard covers accuracy, privacy and cost together.
The instruction to give each tool
Use the same words for all three so you compare the tools and not your prompts. For the specialist tools, load a playbook with your positions where the product allows it; for Copilot, put the positions in the prompt.
You are reviewing a services agreement for the supplier's customer.
Our positions:
- Liability cap must apply to all claims except fraud and death or personal injury.
- Renewal needs at least 30 days' notice and a reminder from the supplier.
- Indemnities must be mutual.
- Payment terms: 30 days from invoice, consistent throughout.
- IP: the supplier keeps pre-existing materials; we own what is created for us.
- Every defined term must be defined.
- Restrictive covenants no longer than 12 months.
For each position, say: met, partly met or not met; quote the clause number;
propose a change as a tracked change; and give a one-sentence reason.
List any other risks you find, separately.Specialist tools can run a rule set in one pass; Copilot will probably need the contract in chunks or a clear instruction to read the whole file. Record any steps you had to add, because an extra step repeated forty times a month is a cost.
Score the results on a sheet, not by feel
For each tool record the same six measurements. Below is an illustrative filled-in sheet showing how to record them. The numbers are invented to show the format; they are not results from any vendor, and your own run will differ.
| Measure | Tool A (illustrative) | Tool B (illustrative) | Tool C (illustrative) |
|---|---|---|---|
| Planted issues caught (of 8) | 7 | 6 | 4 |
| False alarms on clean clauses (of 2) | 1 | 0 | 2 |
| Redlines you would send as written | 4 of 7 | 5 of 6 | 1 of 4 |
| Citations to the right clause | 7 of 7 | 6 of 6 | 3 of 4 |
| Minutes from open to review done | 9 | 11 | 22 |
| Extra steps needed | 0 | 1 | 4 |
Weight the rows for your practice. A firm that negotiates for clients every day will weight "redlines you would send as written" most, because every edit you make by hand cancels the time saving. A firm that mainly reviews other side's paper and writes advice notes may care more about summaries and clause citations. Whatever the weights, a miss on the liability carve-out or the one-way indemnity should count heavily: those are the issues a client would be unhappy to learn were missed.
Where review tools slip, and how to see it
Four failure patterns are worth looking for in your test, because they pass a casual read.
- Right clause, wrong conclusion. The tool quotes the cap clause and says it is fine, because it did not connect it with the carve-out thirty pages later. Look specifically at issue 1.
- A tidy redline that changes meaning. A suggested rewrite that quietly drops "in writing" or swaps "shall" for "may". Compare each proposed change with the original word by word for the first ten redlines.
- Mismatch between summary and body. Tools reading a front-page summary may accept "12 months" and miss the 90-day notice buried below. This is issue 2.
- Silence on missing terms. Finding what is wrong is easier than noticing what is absent. Issue 6, the undefined term, shows whether the tool checks consistency across the document.
Treat every output as a first pass that a lawyer reads, and keep the verification step in the matter file. The wider limits of this kind of tool are covered in what AI contract review catches and misses for small firms.
Walking one planted issue through the tools
Take issue 1, the liability cap whose carve-out swallows it, because it shows the difference between a tool that reads a clause and one that reads a contract. In the dummy agreement, clause 14.1 caps liability at fees paid in the preceding 12 months. Clause 14.3, two pages later, says the cap "does not apply to breach of clauses 9 (Confidentiality) or 11 (Data)". For a supplier handling client data, that carve-out is wide: almost any serious incident is arguably a data breach.
What you want from a review tool is an answer that connects the two clauses. A weak answer reads:
Clause 14.1 limits liability to fees paid in the previous 12 months. This is a common and reasonable position.
A useful answer reads more like this (illustrative wording, not output from a named tool):
Clause 14.1 sets a cap equal to 12 months' fees, but clause 14.3 removes the cap for breaches of clauses 9 and 11, which makes liability for the most likely serious claims uncapped. Not met. Suggested change: replace 14.3 with a separate super-cap of [x] times annual fees for data and confidentiality breaches, and keep fraud and personal injury outside any cap.
Score the first as a miss even though it quotes the right clause. Then check whether the suggested tracked change actually produces valid wording when you accept it: some tools suggest a replacement that leaves a cross-reference to a clause that no longer exists. Open the document's navigation pane after accepting each change and look for broken references.
Fitting the trial windows to the calendar
Trial limits shape how you test. Spellbook's trial is 7 days, so do not start it until the dummy contract, answer key and prompt are ready, and block two hours on day one and one on day three. Gavel Exec's allowance is 25 queries per user, not a time limit, so plan how to spend them: a full-contract review might consume several queries (one for the playbook run, a few for follow-up questions), and you want some left to test a second contract. Write a query budget before you start, for example 8 for the main run, 6 for follow-up questions on missed issues, 6 for drafting a replacement clause and 5 held back.
A suggested schedule for one reviewing lawyer:
- Day 0 (90 minutes). Build the dummy contract, the answer key and the prompt. Have a colleague check that the planted issues are really present and that the clean clauses really are clean.
- Day 1 (2 hours). Run the first specialist tool and fill the sheet. Note every extra click.
- Day 2 (2 hours). Run the second tool and Copilot on the same file.
- Day 3 (1 hour). Redo the weakest result with a revised playbook or prompt, to see whether the gap is the tool or the instruction.
- Day 4 (1 hour). Compare sheets, ask each vendor your data questions, and write a one-page recommendation with the three-lawyer cost.
Redoing the weakest result on day 3 matters. If a tool misses issue 3 (the one-way indemnity) because your playbook said nothing about mutuality, that is a configuration gap you can fix in a minute. If it misses it with the rule in place, that is a tool gap, and you should weigh it.
Before and after: a redline you would actually send
Take issue 5, the IP assignment. The original reads: "All materials used in connection with the Services shall be owned by the Customer." A reviewer who acts for the supplier's customer might like that wording; one who acts for the supplier would not, and the point of the playbook is to say which side you are on. Acting for the customer, the problem is different: the clause is so wide it may be unenforceable or provoke a fight, and it ignores the supplier's genuine pre-existing tools.
Before: "All materials used in connection with the Services shall be owned by the Customer."
After (illustrative): "The Supplier retains ownership of its pre-existing materials and tools. The Customer owns all deliverables created specifically for the Customer under this agreement, and the Supplier grants the Customer a perpetual licence to any pre-existing materials incorporated into those deliverables."
The test is whether the tool proposes something close to the second version without being told how. If it only flags the clause as "broad", you still have to write the fix yourself, which is a time cost to count on the sheet.
Check the data terms before the real contract goes in
Do the demo on the dummy, then put these questions to each vendor in writing before any client document goes in:
- Is customer content used to train any model, ever, by you or your AI providers?
- Where is content processed and stored, and for how long are prompts and documents retained?
- What happens to our documents and playbooks when the contract ends, and how fast?
- Which sub-processors handle content, and will you tell us before adding one?
- Can an administrator turn features off for particular users or matters?
The vendors' statements above (Zero Data Retention agreements, SOC 2 Type II) are claims published by the vendors; ask for the reports or the trust portal link and have whoever handles your firm's risk read them. For Copilot, the answers sit in your Microsoft agreement and tenant settings, so ask whoever administers your Microsoft 365 account to confirm them.
Which firm should pick what
Illustrative profiles, not rules.
- A three-lawyer commercial firm that reviews supplier and customer agreements weekly. Trial both specialist tools; the playbook feature will earn its cost if the same ten positions come up every time.
- A solo estate planner who sees a contract once a month. Even one Gavel Exec seat at $1,740 a year is hard to justify. Start with Copilot Chat and a good checklist, and revisit if volume grows.
- A firm already on Microsoft 365 Copilot for other reasons. Use the Copilot test result as your baseline: if it catches six of eight, the case for a specialist tool rests on the last two and on redline quality.
- An in-house team of two. Spellbook's in-house tier mentions a playbook build service; ask what that costs and how long it takes.
A cautious rollout once you choose
Pick one tool, one contract type, and one lawyer for the first month. Have a second lawyer read ten of the first fifteen reviews against the original contract without looking at the AI's notes, and note any gaps. Only add a second contract type once the first has produced a month with no misses that mattered. For the longer view, the first 90 days for a small law firm sets out what to do next.
Questions lawyers ask before trialling contract review AI
Is Spellbook's price published?
No. Spellbook's pricing page describes custom pricing based on the number of team members on a licence and sends you to a demo. It does advertise a 7-day free trial. Ask for a written quote that states seats, term, renewal price and what happens to your data on cancellation.
Does Gavel Exec publish a price?
Yes, on its product page: 25 free queries per user with no credit card, then $160 per user per month or $1,740 per user per year, with team pricing for 10 or more seats. Confirm the figures on the vendor page before you budget, as they can change.
Is Copilot in Word enough for contract review?
It can summarise, compare and suggest wording, but it does not ship with a contract playbook, clause benchmarks or a redline workflow built for negotiation. Whether that matters depends on how much of your work is repeat contract review. Test it on the same contract as the specialist tools.
Can I put a client's contract into a free trial?
Only with the client's agreement and your firm's rules on confidentiality. Safer: build a dummy contract with planted errors from a template you own. You learn the same thing without exposing any client document to a tool you have not yet approved.
Further reads
- Spellbook vs Copilot: Contract Drafting in Word for a Small Firm — Compare Spellbook and Copilot for drafting in Word.
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Weigh per-seat pricing against pay-as-you-go.
- Can Solicitors Use ChatGPT Without Breaching Confidentiality? — Check confidentiality before using any AI tool.
- Best AI for Solo Lawyers: A Stack That Fits One Person — See a lighter stack for a one-lawyer practice.
- Best AI Agents for Law Firms: What Clio and Spellbook Mean by Agent — Understand what vendors mean by an AI agent.
- Best AI Chatbot for Lawyers: General Models vs CoCounsel — Compare general chatbots with legal-specific ones.
- AI Paralegal Software: What a Small Firm Should Test Before Paying — Three kinds of AI paralegal software, a six-test trial run on your closed matters, and the questions to ask when pricing is quote-only.
- AI Software for Estate Planning Attorneys: Automation or Generative? — Rules-based document assembly against generative drafting for wills and trusts: how each fails, what each costs and the review step that suits each.
- Claude or ChatGPT for a Law Firm: Which One to Pay For — Team seat costs, long-document limits, privacy defaults and Claude for Legal plugins compared, with a weighted three-task trial a small firm can run.
- Family Law Document Automation: Gavel Workflows or Generative AI? — Rules-based assembly against prompting for family law documents: costs for a four-fee-earner firm, how each fails, and a hybrid that uses both.
- Gavel Pricing: Which Workflows Plan a Small Firm Needs — Gavel Workflows from $83 to Scale at $417 a month, matched to a small firm's seats, templates and integrations, with break-even sums and a 7-day trial plan.
- Best AI for Lawyers in Canada: Privacy, Tools and a Safe First Task — ChatGPT, Claude, Copilot and Clio Manage AI compared by task for a small Canadian firm, with confidentiality checks and a low-risk first job.
- Which AI Is Best for Lawyers in Australia? Choose by Task — A decision rule for choosing AI by task for an Australian law practice, comparing general assistants with the AI built into practice software.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Spellbook pricing page and FAQ, Gavel Exec product page, Microsoft Learn Copilot in Word documentation, checked 1 November 2026.