Spellbook Review: Contract Drafting, Word and Multi-Document Work

JSH LAW / LEGAL AI PRODUCT RESEARCH

England and Wales perspective · Sources reviewed 23 October 2026 · Documentary assessment

What product is being reviewed?

Spellbook is the supplier and product brand assessed here. Its company material describes AI software for legal professionals, with a strong emphasis on contract work. The relevant purchase is a specialist workflow around documents, not an authoritative domestic-law database.

The pricing page distinguishes a suite containing the Word add-in and Associate. The current installation guide also covers Google Docs. Older descriptions calling Spellbook exclusively a Word tool are therefore incomplete as a description of the present product family.

That does not mean every interface has identical functionality or that one entitlement covers every task. Ask which products and features are included in the proposed licence, which editors are supported and how account permissions work across them.

Word, Associate and Google Docs do different jobs

The Associate overview describes a distinction between single-document drafting or review in Word and tasks spanning several documents in Associate. It says Associate can reference Word, PDF and text files while editing Word documents, and recommends starting with a small document set for best results.

Confirm how an output moves between interfaces. A correct comparison can become unusable if the exported draft loses a material comment, source reference or change history. Ask the supplier to demonstrate the complete handoff into the team’s normal review process.

What a UK contract reviewer should test

Spellbook’s international availability is not proof that every proposed clause is suitable under English law. The buyer needs to test the law governing the agreement, the commercial role, the practice area and the client’s actual risk appetite.

Watch for language imported from US precedents without explanation. The issue is not whether a clause sounds American; it is whether the proposed legal effect matches the transaction. A term that seems protective may be impractical, inconsistent with another clause or unacceptable in the relevant market.

A playbook should identify approved language, acceptable alternatives, matters needing approval and rules that depend on context. A list of preferred clauses alone may be inadequate where a concession is allowed only for a particular transaction type or contract value.

For specialised arrangements, use an appropriately qualified reviewer. Contract automation does not remove the need to understand employment, consumer, property or other relevant law. Do not use the presence of legal terminology as evidence that the output has dealt with those issues.

Three hypothetical contracting exercises

An NDA with a hidden asymmetry

Create a fictional mutual NDA in which one schedule defines confidential information more narrowly for one party. Ask for review against an approved English-law playbook and require a reason for each proposed change.

The test is whether the tool recognises the interaction between the main agreement and the schedule. A conspicuous edit to the confidentiality clause is not enough if the asymmetry remains elsewhere. Review omissions as well as suggested changes.

A services agreement with an agreed commercial compromise

Use a synthetic agreement, a term sheet and an email recording a limited concession. Ask Associate to prepare a revised draft reflecting the agreed position. Include a separate proposal that was expressly rejected.

The human answer ledger should distinguish the agreed instruction from the negotiating history. A system that treats every suggestion as an instruction can create a contract no one agreed to. Check the final definitions, cross-references and schedules, not only the clause mentioned in the prompt.

A template update across several examples

Provide an invented standard template, a new internal policy and three earlier agreements. One earlier agreement contains an exceptional approval that should not become the default. Ask for an updated template and an explanation of changes.

The aim is to test whether the system follows the current policy rather than averaging historic concessions. Keep the original template, proposed revision and human-approved version. These are proposed evaluation scenarios; JSH Law has not run them in Spellbook.

Comparison, version control and meaningful oversight

A useful review tool should reduce the effort of finding and checking changes, not simply produce a large volume of commentary. Decide how reviewers will accept, reject or escalate proposals and how that decision will be recorded.

When comparing agreements, distinguish a textual difference from a legal difference. Moving a clause may not change its effect; changing a definition may affect many provisions without producing an obvious standalone red flag. Include both types in the pilot.

Ask for an actual export. Does it preserve tracked changes and comments in the required format? Can an external recipient review it without the same subscription? Can the team identify the last approved version? We have not verified a universal export or audit configuration, so these remain procurement checks rather than asserted features.

Document limits also need a practical test. Uploading a file does not establish that all its contents were read. Scanned schedules, embedded images, large tables and converted PDFs can introduce omissions. Start with a manageable sample and expand only after checking the processed text and output.

Confidentiality and the limits of a no-training statement

Spellbook’s security page and pricing disclosures describe security controls and zero-data-retention arrangements with model providers. We have not inspected the underlying reports or contractual exceptions.

The distinction to clarify is between a model provider’s retention and the platform’s storage of projects, playbooks, prompts or history. Ask what persists, who can see it, where it is processed and what happens to backups and exported copies. A service that saves reusable work needs a clear platform-level retention account.

For shared resources, confirm whether the organisation’s permission model is fine-grained enough. A clause library containing a confidential client concession should not become a general precedent available to another team. Review source permissions before connecting a repository or sharing saved instructions.

Legal professional privilege is a legal question, not a software feature that can be promised universally. Nor does a no-training commitment resolve UK GDPR lawfulness, transparency or international-transfer questions. The research hub explains the common framework; a live deployment requires the customer’s own assessment.

Pricing and the real value of a specialist tool

The official pricing page describes customised pricing based on the team and licence. It also advertises a trial for eligible legal users. We did not obtain a UK quotation or verify a universal published monthly tariff.

Ask whether the quotation covers Word, Associate, Google Docs, playbook setup, training and ongoing support. Confirm minimum commitments, renewal, cancellation and the ability to export reusable resources. Do not assume that an installation guide establishes entitlement to every product it describes.

The business case is strongest where contract work is frequent enough to justify maintaining a playbook and checking outputs. Measure the time to a reviewed agreement, including unnecessary redlines and correction work. If the team’s main problem is an unclear instruction or an unmaintained precedent bank, buying AI may not address the cause.

Where Spellbook would be unnecessary or unsuitable

An unrepresented parent seeking help with a position statement does not normally need enterprise contract software. A family-court allegation schedule is not a contract playbook, and reviewing a document for drafting risks is different from preserving evidential status.

Similarly, a contract tool should not be treated as a complete legal research service merely because it can answer a legal-sounding question. Check current authorities through appropriate sources. The reviewer's responsibility extends beyond the wording of the proposed clause to the client’s objective, facts and applicable law.

There may also be no purchase case where a team handles only occasional simple agreements and already has suitable precedents and review support. Proportionality matters: the narrowest tool that reliably solves the actual problem may be the better choice.

A useful pilot and JSH Law’s assessment

Use a small set of synthetic agreements with known defects, agreed commercial instructions and at least one issue that should be escalated. Have a competent reviewer prepare the answer ledger first. Record missed issues, wrong-law assumptions, harmful changes and the time spent correcting them.

Then repeat with a changed playbook and a later amendment. Test the output in the team’s actual editor and review process. A successful demonstration should show the quality of the final approved document, not just the speed of the first markup.

Compare Microsoft Copilot for Legal for a Microsoft-native workflow, Luminance for broader enterprise contract intelligence and Wordsmith for in-house intake and workflow needs. Those are different purchase cases, not a league table.

Our assessment: a plausible specialist option for repeated professional contracting work, particularly where document-native review and reusable instructions matter. It needs a disciplined governing-law and version-control trial. Its relevance to ordinary family-court preparation is limited and should not be inflated to make the review sound commercially exciting.

Can Associate read a PDF and edit it as if it were Word?

The reviewed documentation distinguishes reference reading from editing Word documents. Confirm the workflow and format before relying on a PDF conversion.

Does a suggested redline mean the contract should be changed?

No. It is a proposal. Legal effect, commercial authority and consistency with the rest of the agreement still need a reviewer’s decision.

Sources and methodology

Reviewed on 23 October 2026: company material, pricing, security disclosures, Associate overview and installation guidance. The trust portal did not expose substantive audit material to this review. We have not tested performance, obtained a contract or certified compliance. Recheck after material interface, retention, pricing or playbook changes. Return to the research hub for the comparison directory.

If the problem behind your search is a family-court document rather than a commercial contract, JSH Law can help with defined document-preparation tasks. Book a Consultation to identify the appropriate scope before sharing confidential papers.