Spellbook vs Copilot: Contract Drafting in Word for a Small Firm

For most small firms already on Microsoft 365, trial Copilot's legal skills first: they review a contract against your playbook inside Word and propose tracked changes, for about $21 per user a month. Choose Spellbook when contract work is your core practice and you want market benchmarks, a clause library and vendor-built playbooks, accepting a quote instead of a price.

Both vendors publish comparison pages, and both are partisan and ageing fast. Spellbook's own Copilot page, dated 20 October 2026, says Copilot returns prose rather than tracked changes, while Microsoft's support page for its legal skills describes redlines as tracked changes. Only your own non-disclosure agreement (NDA), your own playbook and a stopwatch settle it, so most of this tutorial is that test. Neither product gives legal advice; both draft for a lawyer to check.

Follow me on Instagram@sagnikteaches

What Copilot's legal skills do inside Word

Microsoft made its legal skills generally available in mid-October 2026, according to launch coverage, building on a "Legal Agent" it had previewed earlier. They live in Copilot in Word desktop and need an active Microsoft 365 Copilot commercial licence. They are on by default: Copilot notices when a legal document is open and uses the skills for the rest of that chat, and a user can switch the Legal plugin off under Manage plugins and skills.

Connect on LinkedInSagnik Bhattacharya

The skill that matters for this comparison is Review contract. According to Microsoft's getting-started page, you ask Copilot to review the contract using a playbook and attach the playbook as a .docx file. Copilot breaks it into topics, each with rules and example clauses, and shows them to you so you can correct any it misread. You press Run Review, the analysis streams in one topic at a time, and each proposed change arrives as a tracked change you can accept singly or all at once. Microsoft states one limit plainly: Copilot can currently review a contract only from a playbook, not from a template or another document.

Subscribe on YouTube@sagnikteaches

Drafting works from a precedent. Attach an earlier agreement, call the draft skill, and Copilot turns the precedent into a template with placeholders that you fill from a chat or an attached document. Microsoft's transparency documentation lists further skills that a small firm will use more than it expects: check defined terms, check cross-references, summarise redlines, compare key terms and check execution readiness. Copilot also works inside a document that already contains tracked changes, keeping earlier revisions separate from its own proposals, and its answers carry clickable citations that highlight the source language.

The same documentation lists the failure modes, and a firm should read them before any demo. In complex or lengthy documents Copilot may miss relevant clauses or misapply playbook guidance. It is not jurisdiction-aware by default. Fluent wording is not evidence of legal accuracy. The skills were tested mainly in English. Microsoft's own tip is to hand it one issue at a time ("Update the limitation of liability provision to exclude consequential damages for the buyer") rather than asking it to remediate every risk at once.

What Spellbook adds inside the same Word window

Spellbook is a Word add-in built only for contracts. Its feature pages list five parts: Review (redlines and comments, with playbooks), Draft, Ask, Market (shown as Compare in its menu: a clause set against thousands of similar agreements, filtered by industry, jurisdiction and deal type) and Associate, an agent for work across several documents. Spellbook says its edits appear under the reviewer's name, so a redline can go to a client without rework. Microsoft's getting-started page does not say whose name Copilot's tracked changes carry, which is why the author field is the first thing the test below checks.

Draft works from scratch or from a clause library of your own precedents. Spellbook says it detects contract type, jurisdiction, party details and writing style, and that it can review, draft and chat in more than 140 languages. Integrations listed on its site include OneDrive, SharePoint, Dropbox, Google Drive and iManage. On security it states SOC 2 Type II compliance and zero-data-retention agreements with its model providers, so client text is not stored or used for training. Ask for the contract wording, not the web page.

Treat the benchmarking claims as the vendor's. Spellbook's guidance says its review is grounded in more than 20 million contracts and 270-plus clause types. What that data set contains, and how "market" is defined for a small commercial lease or a services agreement, is exactly what a demo should be made to show on your own document.

Licence cost: a $21 price against a quote

Copilot's cost is public. Spellbook's is a conversation: its pricing page says pricing follows the number of team members on the licence, offers a 7-day free trial and asks you to book a demo. Third-party roundups print figures, but none appear on Spellbook's own pages, so ignore them and ask for a written quote. How a small firm should read a Spellbook quote covers the line items.

A quick sum shows how hard to push on price. Divide the monthly price per seat by the value of a lawyer's minute. At an internal value of $200 an hour, about $3.33 a minute, Copilot at $21 pays back after roughly six minutes saved per seat per month. For Spellbook, use the quote: a placeholder of $150 per seat needs 45 minutes a month and a placeholder of $300 needs 90. Those two are placeholders to show the sum, not Spellbook prices.

For a five-seat firm on Copilot at the annual rate, the bill is $105 a month, or $1,260 a year; with the first-year promotion, if the firm qualifies, it is $1,080. The sum is the same one legal AI pricing for small firms applies to per-seat and pay-as-you-go tools, and Microsoft 365 Copilot pricing by licence explains which base plan you need underneath. Keep the trial inside Word: Cowork is billed by usage and would muddy the comparison.

Write one playbook, then load it into both

A playbook is a written standard. For each clause it records the position you want, what you will accept instead and when to hand the document to a partner. Both tools are only as good as that document, so write it once, in plain language, before opening either. This one is filled in for an imagined four-lawyer firm acting for small business clients. It is illustrative: adapt the positions, and have a lawyer own it.

The wording of each rule matters more than the table suggests. A rule such as "check the confidentiality period is reasonable" gives either tool nothing to measure against. Compare it with the same rule rewritten so that it can be applied the same way on every review:

Rule 4 - Duration of confidentiality
Standard: 3 years from the date of disclosure.
Accept: up to 5 years. Trade secrets stay protected for as long as they remain trade secrets.
Flag: any perpetual obligation for other information, or a period that restarts on each disclosure.
Suggested wording: "The Receiving Party's obligations under this Agreement continue for three (3) years from the date of each disclosure, except that obligations in respect of any trade secret continue for as long as the information remains a trade secret."

To load it into Copilot, save the playbook as a Word document, open the NDA and ask Copilot to review the contract using a playbook. Read the topics it builds before pressing Run Review. A topic that merged two of your rules, or dropped the "accept instead" column, will review differently from what you meant. If you have no playbook yet, Microsoft's skill list includes one that creates a playbook from a set of precedents; treat its output as a first draft that a partner marks up.

To load it into Spellbook, remember where its own guidance puts playbooks: saved in the Review feature and reused on later contracts. Its comparison page says a legal team's playbook is configured by its customer success team, while the pricing page offers group training sessions to law firms. During the trial, ask who builds the playbook, how long that takes, how many playbooks a plan allows and whether changing a rule later costs anything.

Run one NDA through both tools

Start from a real but old NDA: anonymise it and seed it with eight problems taken from the playbook, so the right answer is known before either tool starts. A six-page mutual NDA is long enough to be realistic and short enough to run twice in an afternoon.

Run both tools under identical conditions:

1. Use the same laptop, the same day, the same playbook content and the same Word build for both.
2. Start the clock when you ask for the review and stop when every proposed change is accepted or rejected.
3. Score each seeded problem 0 (missed), 1 (flagged, but the redline is unusable) or 2 (usable redline).
4. Count false alarms: changes you rejected because they contradicted the playbook or the document.
5. Open the author field on the first tracked change and note whose name it carries.
6. Repeat on a long agreement, such as a 25-page services agreement, because Microsoft names length and complexity as where misses happen.

Here is how a filled-in sheet might look. These scores are invented for a made-up run to show the method; they are not a benchmark of either product.

Break ties in a fixed order so nobody argues afterwards: fewest problems missed or unusable (A has three, B has two), then fewest false alarms, then total minutes, then price. On these invented numbers B wins on quality and A on speed, and the five-minute gap is worth about $17 at $200 an hour. B wins only if tidying A's extra weak redlines and false alarms takes longer than five minutes, so time that clean-up as well.

Three edits to inspect before you accept anything

The cross-reference trap. Delete the residuals clause with Accept all and everything below it renumbers, while clause 11 still points to "Section 9.4". Copilot lists a check-cross-references skill in its transparency documentation; for Spellbook, ask in the demo what it does after a deletion. Either way, update fields in Word and read every reference once before the document leaves the building.

Defined-term drift. A redline that introduces "Recipient" into a document built on "Receiving Party" reads fine in isolation and fails on review. Copilot has a check-defined-terms skill. Whatever the tool, search the final text for each defined term and its near-synonyms.

A leftover the tool did not touch. Try a narrow instruction in Copilot, then read what it changed and what it left alone.

Prompt: In clause 6, change the confidentiality period to three years from disclosure, keep trade secrets protected while they remain trade secrets, and leave everything else unchanged.

Sample output (illustrative): a tracked change replacing "ten (10) years" with "three (3) years from the date of disclosure, save that obligations in respect of any trade secret continue for as long as the information remains a trade secret".

What you would fix in that output: "save that" against the firm's house style of "except that", and the survival clause in section 12.2, which still says "ten years" and was never mentioned. The narrow prompt did exactly what it was told, which is the point of Microsoft's one-issue-at-a-time advice, and the leftover is the cost of it. Search the document for the old figure before accepting. AI contract review for small firms goes through what this class of tool tends to catch and miss.

Data handling and client consent before a real matter

The two tools handle confidentiality differently, and neither answer is automatically better. Spellbook's position is contractual: zero-data-retention agreements with its model providers and SOC 2 Type II compliance, which you verify in the order form and its trust documentation. Microsoft's position is architectural: it says Copilot runs within the Microsoft 365 security, compliance and governance controls the organisation already uses, so a document Copilot can read is a document the user could already open. That second point is only as safe as your permissions. Cleaning up SharePoint permissions before Copilot is the job to do first if client folders are shared more widely than they should be.

Before the first live matter, check three things with someone who knows your professional conduct rules. Does the engagement letter cover AI-assisted drafting? Does a client's outside-counsel guideline restrict it? Who reviews every AI redline before it leaves the firm? Run the whole trial on anonymised documents first, and keep a note of which version of each tool and which model settings you tested, because both change often.

Questions to ask each vendor before you commit

For the Spellbook demo:

• Which Word versions and platforms are supported, and does the add-in need admin approval in our Microsoft 365 tenant?
• Is Associate in the quote, and what is the minimum number of seats and the term?
• Who writes our playbook, how long does it take and who pays for changes?
• What is in the market data set, and how is "market" defined for a deal like ours?
• Whose name appears on tracked changes, and can a reviewer change that?
• Show the zero-data-retention wording in the contract and the latest SOC 2 report.

For the Microsoft licensing partner or your admin:

• Is the first-year Copilot Business price open to us, and what is the renewal price?
• Does our base plan qualify, and are there seat minimums?
• Is the Legal plugin switched on for the right people only, and can it be limited by group?
• Which Word build do we need, and does the web version of Word support these skills?
• What does our current SharePoint and OneDrive permission model expose to Copilot?

If a head-to-head with a third tool is on the table, Spellbook against Gavel Exec covers a different pairing with the same test in mind.

Which small firm should pick which

For a three-lawyer firm on Microsoft 365 Business Standard that mostly reviews NDAs, leases and supplier terms for small clients, the price gap and the existing licence point to a Copilot trial first. The firm writes the eight-rule playbook above, runs the seeded NDA and the long agreement, and moves to a Spellbook trial only if Copilot misses more than two seeded problems or the partners want market benchmarks they cannot get otherwise.

A six-lawyer commercial firm that reviews vendor and software agreements every week is a different case. Contract work is the product here, so a Spellbook quote deserves a place in the comparison from day one. Start both 7-day windows on the same Monday, run the same documents through each and decide by Friday. The quote needs to clear the break-even sum above at the firm's real minutes saved, not the vendor's.

The third case is a solo lawyer who sees about four contracts a month. Saving 10 minutes on each is 40 minutes a month. That repays a $21 seat many times over, but it falls short of a hypothetical $150 seat, which needs 45 minutes. In practice the solo lawyer's answer is usually the cheaper licence plus a good playbook, revisited once volume grows.

Whichever way it falls, keep the playbook. It is the asset that moves with you between tools, and it is the part of this exercise that improves your drafting even if neither product is bought.

Questions small firms ask before choosing

Can Copilot review a contract without a playbook?

It can summarise a contract, explain clauses with clickable citations and propose edits when you describe one issue at a time. Microsoft's Review contract skill, though, currently works only from a playbook; it cannot yet review from a template or another document. Without a playbook you are prompting clause by clause, which is slower and less repeatable.

Does Spellbook train on my clients' documents?

Spellbook states that it holds zero-data-retention agreements with its model providers, which prevent client text being used for training, and it lists SOC 2 Type II compliance. Treat both as claims to verify: ask for the clause in your contract and the security documentation, and run the first trial matters on anonymised documents.

Does a small firm need Spellbook's Associate agent?

Only for work that spans several documents, such as a data room or a master agreement with amendments. A firm reviewing single NDAs and services agreements will rarely use it. Spellbook's pricing page lists Associate in the same suite as the Word add-in, so ask whether it is inside your quote or priced separately.

Does free Copilot Chat include the legal skills?

Microsoft's support page describes the legal plugins as available to Microsoft 365 Copilot commercial subscribers, so the free chat does not appear to qualify. Licences differ by tenant and plan, so open Word on a licensed test account and check that the Legal plugin appears under Manage plugins and skills before you plan around it.

Further reads

• ChatGPT or a Legal AI Tool: Which Should a Small Firm Use? — Adds the general-assistant option to the Spellbook and Copilot choice.
• AI App for Lawyers: Score Accuracy, Privacy and Cost in a Short Trial — A scoring sheet for any legal AI trial, beyond contract tools.
• How to Stop AI Inventing Case Law: A Small-Firm Checking Routine — The checking routine that applies to every AI draft a lawyer signs.
• Which Legal Tasks Should a Small Firm Never Hand to AI? — Draws the line between drafting help and tasks to keep human.
• Copilot in Word: Draft Proposals From Your Own Files — How Copilot in Word behaves with your own files, outside legal skills.
• Does Microsoft 365 Copilot Keep Your Business Data Private? — What Microsoft says happens to your data when Copilot reads it.
• Copilot for Legal: What It Adds and Whether You Need a Paid Licence — Copilot for Legal adds contract review, drafting and chronology skills to Word and Cowork. Which licence it needs and what a small firm gets for the money.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Spellbook pricing, security, review, draft, market and playbook pages; Microsoft Support pages for Copilot for Legal and its transparency documentation; Microsoft Copilot business pricing page; Copilot for Legal launch coverage from October 2026.