Spellbook AI Contract Drafting Tool: Evaluation for Legal Teams

What Spellbook Is and What It Does

Spellbook is a contract drafting and review assistant built by Rally Legal (formerly Spellbook Legal Inc.), a Toronto-based legal technology company. It operates as a sidebar add-in inside Microsoft Word, which is how most of its users encounter it — there is no separate web interface for the core drafting workflow.

The tool uses large language models — primarily OpenAI's GPT-4 family — to generate contract language, suggest clause alternatives, flag potentially missing provisions, and explain what existing language means in plain terms. It is positioned squarely at transactional work: NDAs, SaaS agreements, employment contracts, commercial leases, vendor agreements, and similar document types.

Spellbook is not a contract lifecycle management (CLM) platform. It does not store contracts, manage signature workflows, or track obligation deadlines. Its scope is limited to the drafting and initial review stage — the moment when a lawyer is working in a Word document and needs language generated, checked, or explained.

Core Feature Set

• Clause drafting: Generate new contract language from a prompt or from context in the surrounding document. Supports both starting from scratch and filling gaps in an existing draft.
• Clause suggestions: Spellbook scans an open contract and flags provisions that appear to be missing or that are commonly negotiated in that document type. This is one of its more practically useful features for associates doing initial review.
• Redlining assistance: Suggest alternative language for a selected clause, with the option to specify the party whose interests you're representing. The output appears in the Word sidebar for manual insertion.
• Plain-language explanations: Ask what a selected clause means, what risks it creates, or how it compares to market-standard language. Output is conversational and non-citational.
• Playbook mode: Enterprise-tier customers can upload their own clause libraries and negotiation playbooks, which Spellbook uses as context when generating suggestions. This significantly improves output alignment with firm or client standards.

Pricing Structure

Spellbook publishes tiered pricing. As of the date of this evaluation, the structure is roughly as follows — exact figures should be confirmed directly with the vendor, as pricing has changed multiple times since 2023.

Data Privacy and Confidentiality

This is the question legal teams ask first, and it deserves a direct answer rather than a reassuring summary.

Spellbook routes requests through OpenAI's API under a data processing agreement that, per OpenAI's enterprise API terms, does not use customer data for model training. Spellbook's own terms state that document content is not retained after the session ends. The vendor does not claim on-premises deployment or data isolation at the infrastructure level — the model calls go to OpenAI's cloud.

For most transactional work involving commercial counterparties, this architecture is workable if the firm has reviewed and accepted the data processing terms. For matters involving government clients, regulated financial institutions, or any context where client confidentiality agreements restrict cloud processing, the architecture requires closer scrutiny before deployment.

Output Quality: What Works and Where It Falls Short

Where Spellbook Performs Well

Spellbook is genuinely useful for drafting first-pass language on high-volume, lower-complexity agreements. NDAs, standard vendor contracts, and SaaS subscription agreements are the sweet spot. For these document types, the clause suggestions are frequently on-point and the generated language is grammatically clean and structurally coherent.

The missing-clause detection is one of the more practically valuable features. Pointing Spellbook at a short-form vendor agreement and asking it what's missing often surfaces provisions — indemnification caps, data breach notification obligations, IP ownership — that a junior associate might not flag on a first pass. This does not replace attorney review, but it functions as a reasonable checklist supplement.

The playbook feature, available at the enterprise tier, materially improves output quality. When Spellbook has access to the firm's preferred fallback positions and approved clause language, the suggestions align much more closely with actual negotiating practice rather than generic market-standard language.

Documented Limitations

• No citation to authority: Spellbook does not cite case law, statutes, or regulations to support its suggestions. When it describes a clause as "market standard" or flags a risk, there is no underlying legal authority provided. Attorneys must independently verify any legal characterization.
• Jurisdiction sensitivity is limited: The tool can be prompted to consider a specific jurisdiction, but it does not natively adjust output based on detected governing law. A California-law employment agreement and a New York-law employment agreement may receive nearly identical suggestions unless the attorney explicitly specifies the jurisdiction in every prompt.
• Complex or bespoke transactions: For M&A agreements, credit facilities, or highly negotiated bespoke contracts, Spellbook's output is less reliable. The model's training on general contract language means it does not handle deal-specific economics, unusual structures, or highly negotiated terms as well as purpose-built tools or experienced attorneys.
• Hallucination risk remains: Like all LLM-based tools, Spellbook can generate plausible-sounding but legally incorrect language. This risk is highest when the tool is asked about specific legal requirements, regulatory thresholds, or jurisdiction-specific rules. Every output requires attorney review before use.
• No version control or audit trail: Spellbook does not maintain a record of what was generated, when, or by whom. For firms that need to document AI use in their workflow for ethics or malpractice purposes, this is a gap that requires supplemental process design.

Target Audience: Who Benefits Most

Integration and Deployment

Spellbook's Word add-in is its primary delivery mechanism. Installation is straightforward — it appears in the Microsoft AppSource marketplace and can be deployed centrally by IT administrators through the Microsoft 365 admin center. This makes enterprise rollout relatively low-friction compared to tools that require browser extensions or standalone applications.

There is no native integration with document management systems (DMS) like iManage or NetDocuments as of this evaluation. Documents must be open in Word for Spellbook to access them. For firms where the DMS is the primary working environment, this creates an extra step in the workflow.

Single sign-on (SSO) is available at the enterprise tier. Starter and Professional tiers use standard email-based authentication.

Comparison to Adjacent Tools

Spellbook occupies a specific position in the contract AI landscape. It is not a full CLM platform like Ironclad or Conga. It is not a document review tool focused on due diligence extraction like Luminance or Kira. It is a drafting assistant — closer in function to what Harvey offers for contract work, but with a narrower interface (Word add-in only) and a lower entry price point.

Harvey, by comparison, operates as a standalone web application and is positioned more broadly across research, drafting, and correspondence. Spellbook's Word-native approach is an advantage for attorneys who live in Word and want minimal context switching, but a limitation for teams that need multi-modal workflows or research capabilities alongside drafting.

Professional Responsibility Considerations

Several bar associations have issued guidance addressing attorney use of AI tools in client work. The consistent themes across these opinions are competence (understanding what the tool does and does not do), confidentiality (ensuring client data is protected), and supervision (ensuring attorney review of AI-generated output before use).

Spellbook use implicates all three. Attorneys need to understand the tool's limitations — particularly its lack of legal citations and its jurisdiction-sensitivity gaps — to use it competently. Client document content sent to the tool must be consistent with confidentiality obligations. And every output must be reviewed by an attorney before it enters a client deliverable; Spellbook's own terms disclaim any warranty on the legal accuracy of generated content.

Evaluation Summary

Spellbook is a competent drafting assistant for transactional attorneys handling high-volume, lower-complexity commercial agreements. Its Word integration is genuinely useful — it meets attorneys where they already work. The playbook feature makes it meaningfully more valuable at the enterprise tier for teams with standardized clause libraries.

Its limitations are real and should not be minimized. No citation to authority, limited jurisdiction sensitivity, and the absence of an audit trail are gaps that matter for professional responsibility compliance. The hallucination risk inherent in LLM-based tools means attorney review is not optional — it is a hard requirement.

For solo practitioners and small transactional teams, Spellbook's price point and ease of deployment make it worth evaluating seriously. For large firms handling complex, bespoke transactions, the value proposition is less clear without a rigorous internal pilot.