Spellbook vs Harvey 2026: Compare Legal AI Workflows

Choose the scope before choosing the vendor

Choose Spellbook when your buying priority is transactional drafting, redlining and consistent contract positions. Shortlist Harvey when that requirement sits alongside legal research, document analysis and workflows across practice groups. The overlap is meaningful: Harvey also works in Word, and Spellbook includes a multi-document agent. A comparison that reduces this choice to “Word add-in versus research chatbot” misses both products’ current scope.

Legal AI is software that uses artificial intelligence to assist lawyers with tasks such as drafting, research and document analysis while leaving professional judgment with the lawyer.

TL;DR: Start with Spellbook for a contract-centered requirement and Harvey for a broader legal platform requirement. Compare both on your actual contract work if that is the deciding use case. Neither has a published commercial rate card that establishes a price winner, so obtain comparable written quotes before making an economic recommendation.

Both products retain their names. For procurement identity, Spellbook’s terms updated September 17, 2026 identify Spellbook US Inc. for US customers and Dialog Enterprises Inc. for customers elsewhere, according to Spellbook. Harvey’s provider is Harvey AI Corporation; its platform agreement updated January 9, 2026 places fees and payment terms in customer order forms, according to Harvey. Contract with the identified legal entity rather than treating an investor or underlying model provider as the product owner.

Key Takeaways

• Spellbook is the more focused starting point for a transactional team buying contract assistance; that is a fit judgment, not a measured accuracy ranking.

• Harvey warrants consideration when research and broader document workflows belong in the same procurement, including firms outside the largest enterprise tier.

• Both support drafting in Word, so evaluate redline quality and precedent handling rather than awarding a winner simply for having an add-in.

• Spellbook’s Associate expands its scope beyond individual documents; its advertised contract-management expansion requires separate availability and entitlement confirmation.

• Both require commercial quotes. Normalize included capabilities, users, implementation and renewal terms before comparing total cost.

• The decisive pilot measures usable work after lawyer review, with client-data restrictions and a documented handoff back into the firm’s existing systems.

Who this is for

This guide is for an innovation lead who can name the legal work the firm wants to improve and involve practicing lawyers, knowledge management, information security and procurement in the decision. It also suits a transactional practice lead deciding whether a focused tool is sufficient or a broader platform belongs on the shortlist.

Red flags: Your firm cannot authorize the proposed client-data processing; nobody owns substantive review of generated work; the requirement is only document storage or deterministic template assembly.

Those conditions can disqualify either purchase. A convincing demonstration does not resolve an unsuitable data arrangement, and a broader feature catalog does not create adoption ownership. If the underlying problem is an unclear playbook, fix that before expecting software to apply the firm’s positions consistently.

Avoid screening by invented firm-size thresholds. Scope, approved data access, workflow ownership and commercial terms matter more than assuming that a boutique must choose Spellbook or that an international firm must choose Harvey.

How we evaluated the buying decision

This is a public-information buyer’s guide, not a product test. Vendor statements establish advertised capabilities and commercial structure; independent reporting supplies context. The weights below are our proposed procurement framework, not survey findings or vendor performance scores.

Use the same approved documents and instructions for each candidate. Apply the weights only after resolving hard requirements such as permitted processing, necessary jurisdictions and acceptable export. A product that fails a mandatory requirement should not win through weighted averaging.

The weights total 100% by design. A contract-only evaluation can redistribute the research allocation before the pilot, but apply the revised framework equally. Do not alter weights afterward to justify a preferred vendor.

Surveyed legal professionals using AI: 79% according to Clio (2024). That broad adoption measure includes AI use in some capacity; it does not establish either product’s suitability, legal accuracy or return on investment.

Keep technology enthusiasm separate from purchasing evidence. The broader editorial approaches in our SaaS procurement comparison and agency workflow comparison offer adjacent reading on evaluating workflow fit; neither substitutes for legal-specific validation.

Where the capabilities overlap—and where they diverge

The matrix normalizes advertised scope around buying questions. An entry describes a public capability or an item to confirm, not feature parity, entitlement in every contract or independently demonstrated performance.

Read the research row carefully. Spellbook can assist with questions and multi-document analysis; that does not establish equivalence to a jurisdiction-specific research service. Harvey’s research positioning likewise does not establish coverage of every authority you require. Ask for the precise corpus, source access and validation behavior relevant to your practice.

The integration row is deliberately conditional. An advertised connection, a Word add-in and an externally callable API are different things. Procurement should distinguish retrieving documents, invoking analysis, exporting results and recording approvals. Confirm each required action rather than assuming a logo means the whole matter workflow is connected.

A proposed workflow from US Tech Automations could begin when an approved intake record receives a contract attachment: validate the matter identifier and document version, create a review task, and package the approved playbook for the lawyer’s chosen tool. Prerequisites would include authorized intake and document-system APIs or exports, access permissions and a supported handoff into the vendor. If vendor automation access is unavailable, that handoff remains a manual task. A lawyer approves scope before analysis; the output is a traceable review packet, not an automatically accepted contract.

Spellbook: the contract-centered shortlist

Spellbook’s published suite combines its Word add-in with Associate, plus training and support, and lists review, drafting, questions, benchmarks and playbooks. Its pricing depends on team members rather than a public commercial amount. Published trial duration: 7 days according to Spellbook. Confirm what the trial and eventual quote include before planning the evaluation.

Best fit: A team whose procurement begins with negotiating contracts, applying approved positions and reducing repetitive drafting work. The practical attraction is concentrating the evaluation on work lawyers already recognize: a document, a clause, a proposed change and the reason for that change.

Meaningful limitation: A transactional emphasis may leave your broader research or practice-wide requirements unresolved. Associate makes “single-document only” an inaccurate description, but multi-document assistance does not automatically satisfy every diligence, litigation or knowledge-management requirement. Demand evidence for the actual workflow instead of extrapolating from the product category.

The scope is expanding. Its current official materials advertise early access. Treat availability, contract inclusion and operational responsibility as procurement questions; do not assume that purchasing the existing suite includes every advertised expansion.

Implementation should start with approved precedents and a clear negotiating playbook. Have practice leads distinguish preferred positions, acceptable fallbacks and matters requiring escalation. Then evaluate whether suggestions preserve defined terms, cross-references and the client’s commercial intent.

Ask lawyers to review outputs without vendor coaching. Record which suggestions they accept, reject or rewrite and why. A polished first pass is insufficient if senior lawyers must reconstruct the analysis before trusting it.

Choose Spellbook when that focused evaluation meets the firm’s needs and its commercial scope is acceptable. Disqualify it if required research coverage, controls or broader workflows cannot be demonstrated under the proposed contract.

Harvey: the broader platform shortlist

Harvey’s public product navigation covers agents, Vault, Knowledge and collaboration alongside contract-focused capabilities. Its Word functionality is also established: the drafting announcement dated August 26, 2025 describes Ask and Edit modes, redlines, playbooks and pulling precedents from Vault, according to Harvey. Word integration therefore cannot be the sole reason to prefer Spellbook.

Best fit: A firm seeking to evaluate contract work alongside research, document analysis and reusable workflows across practice groups. Harvey deserves consideration when the purchasing question is how those activities share context and governance, rather than how to improve an isolated drafting task.

Meaningful limitation: Breadth can increase evaluation and implementation work. That is a buyer-side consequence, not evidence of a mandatory long deployment. A firm buying only clause suggestions should establish why the additional scope is valuable before accepting a broader commitment.

Implementation should identify an initial practice use case, the permitted knowledge sources and the people responsible for validating outputs. Decide which documents can enter shared collections, how ethical walls must behave and who can revise reusable instructions. Avoid treating general access as permission to upload every client file.

Ask for evidence of source handling using authorities and documents from the intended practice. Review whether an answer accurately represents its cited material, distinguishes supplied facts from inference and identifies missing information. Inspect exported work as well as the on-screen response.

A second proposed US Tech Automations workflow could trigger when a lawyer marks a diligence analysis as approved: collect the permitted output and source references, attach the document-version record, and create follow-up tasks for unresolved issues in the matter system. This requires supported vendor exports or authorized API access, a writable matter-system interface and an agreed evidence format. The supervising lawyer approves issue classification and each external deliverable; failed exports enter a staffed exception queue. The intended output is an internal matter packet with outstanding actions, not autonomous legal advice.

Choose Harvey when the broader scope survives that evaluation and the quote matches the needed capabilities. Disqualify it if the firm cannot govern shared context, required sources are unavailable or the proposed commercial scope includes substantial capabilities without an adoption owner.

For related model-level context, read our Harvey Tenet explainer. Keep model announcements separate from the features and terms actually included in your purchase.

Pricing and total cost: compare the written scope

Pricing checked October 10, 2026.

Spellbook publishes custom pricing and a user-based structure. Harvey’s official sales route does not provide a commercial rate card, and its agreement assigns fees to order forms. The appropriate public-price entry for each is Quote-based. There is no supported basis here for declaring either product cheaper.

Ask each vendor to quote the same approved user group and required workflows. Separate committed license cost from internal configuration, training, quality review and maintenance. State which costs are included, excluded or unknown instead of treating unquoted work as free.

Read renewal, cancellation, reassignment and export provisions before signing. A nominally attractive quote can be a poor fit if the firm cannot reduce unused commitments or retrieve its work in a usable form.

Time savings are capacity, not automatically cash. Under hourly arrangements, less recorded work can change revenue rather than simply improve margin. Under fixed fees, usable capacity still depends on demand, staffing and actual review effort. Build the business case around your billing model and observed pilot outcomes.

Confidentiality and evidence are buying gates

Firms using cloud-based legal tools: 73% according to American Bar Association (2024 survey). That context supports evaluating legal AI alongside existing document and practice systems; it does not prove that any particular upload is appropriate.

Before either pilot, map the document path through storage, retrieval, processing, output and deletion. Identify which data enters which service, the governing contractual terms and the administrative controls your firm can actually exercise. Model-provider retention and retention of files by the application are separate questions.

Formal Opinion 512, issued July 29, 2024, addresses competence, confidentiality, communication and fees in generative-AI use, according to American Bar Association. Your professional-responsibility team should apply relevant jurisdictional and client requirements to the proposed use rather than treating a vendor assurance as a legal conclusion.

The useful procurement artifact is a written data-flow and responsibility map. Include document access, source validation, approval ownership, exception handling and the evidence retained after review. If a required control remains unconfirmed, mark it unresolved rather than awarding a passing score.

An illustrative pilot that measures the whole task

Illustrative scenario, not a vendor result: A transactional team evaluates 20 approved agreements, assumes a baseline of 30 minutes per agreement and sets an illustrative pilot outcome of 18 minutes including lawyer review. Baseline effort is 20 × 30 = 600 minutes; pilot effort is 20 × 18 = 360 minutes; the difference is 240 minutes, or 4 hours. An optional n8n handoff can record $execution.id, the documented identifier for the current workflow execution, according to n8n. Those workload figures are explicit scenario assumptions, not figures reported by n8n or either legal vendor; the execution identifier traces the handoff and does not prove the legal analysis is correct.

Replace those assumptions with observed results before recommending a purchase. Count correction, source checking and completion of the matter-system handoff. Track significant missed issues separately, because a faster review can still be an unacceptable review.

Use comparable tasks and reviewers, and preserve the instructions, document versions and final approved work. Include unusual clauses and conflicting provisions, not just clean templates. The innovation lead needs to see whether the tool helps on the work that actually creates review pressure.

Build the handoff yourself or configure ownership

Zapier, Make, n8n and an in-house integration are fair alternatives for moving intake records, approved files and follow-up tasks. These approaches can support run histories, retries, error branches and audit evidence when configured. Their suitability depends on the connections, deployment options and controls your firm approves.

The buyer must still design and own observability, idempotency, escalation, access controls and maintenance. Idempotency means a repeated trigger does not create duplicate matter tasks or repeat an action unintentionally. Run history alone does not establish which document version a lawyer approved.

A proposed managed design could specify a duplicate-detection rule, approval checkpoints, exception ownership and the evidence package before building the connection. Those are configurable design choices, not exclusive capabilities. The comparison is whether your team can own and maintain the same requirements economically.

When NOT to use US Tech Automations

Use the vendor’s existing workflow when it already provides the required review and handoff. Choose a straightforward internal procedure when the only gap is recording an approval or placing an exported file in the correct matter folder. Keep an existing no-code or in-house integration when it already meets your access, exception-handling and maintenance requirements; adding another orchestration layer would create unnecessary ownership.

Questions to settle before the order form

Is Spellbook or Harvey better for contract drafting?

Spellbook is the more focused starting point for contract-centered procurement, while Harvey should also be evaluated when broader legal work matters. Compare accepted redlines, correction effort and precedent handling on the same approved documents before choosing.

Does Harvey work inside Microsoft Word?

Yes, Harvey offers a Word add-in with drafting and editing capabilities. Evaluate its workflow alongside Spellbook rather than relying on outdated comparisons that portray Harvey as web-only.

Is Spellbook limited to individual contracts?

No, Spellbook’s advertised suite includes Associate for multi-document work. Confirm the intended document workflows and feature entitlements, and assess ACM availability separately from the existing suite.

Which vendor has the lower price?

Public pricing does not establish a cheaper vendor because both require commercial quotes. Compare matching users, capabilities, implementation assumptions and contract terms before calculating total cost.

Can either tool replace lawyer review?

Neither should replace the accountable lawyer’s substantive review. Your evaluation should include accuracy, source checking, confidentiality and approval responsibilities rather than assuming fluent output is acceptable work.

Can we connect either product to our matter system?

Possibly, but the proposed connection must be confirmed against supported APIs, exports and contractual permissions. Use a manual handoff where necessary instead of assuming that an add-in provides unattended integration access.

Make the recommendation the firm can defend

Recommend Spellbook when contract-centered work passes the firm’s practical and governance checks. Recommend Harvey when broader research and document workflows justify the additional scope. Keep both on the shortlist when the decisive task is contract work and the pilot has not yet separated them.

Before the order form, confirm required capabilities, approved data handling, reviewed output quality, handoff ownership and comparable quotes. If either candidate fails a mandatory condition, state the disqualifier directly. If evidence is missing, leave that point unresolved.

Where the remaining problem is routing approved work between existing systems, see how US Tech Automations configures a review handoff around supported access, explicit approval points and exception ownership. Choose the legal tool for the legal task, then specify who owns the surrounding workflow.