Contract Review AI: Comparing LegalOn, Spellbook, and Luminance
Three platforms, three architectures, three very different sweet spots. Here is a technical comparison that goes deeper than the feature matrix.
The contract review AI market has matured past the phase where every vendor claims to do everything. Good. Because in my experience evaluating software platforms across industries, the products that try to do everything well end up doing nothing particularly well. Specialization is a feature, not a limitation.
78% of corporate legal departments and law firms are either actively using AI for contract review, evaluating solutions, or exploring capabilities. The average in-house attorney still spends 4.5 hours daily on contract review. Those two numbers together explain why this market is growing fast and why getting the selection right matters.
Let me take these three platforms apart and show you what is actually happening under the hood.
LegalOn: The Playbook Architecture
Best for: In-house teams and firms handling standard contract types at volume.
LegalOn's core design decision is to ship with pre-built, attorney-crafted playbooks that work immediately. Over 600 pre-built rules across standard contract types. Users report reviewing contracts within one hour of installation — no training period, no custom configuration.
From an engineering perspective, this is a rules-forward approach with AI augmentation. The playbooks define what to look for, what language to prefer, what to flag, and when to escalate. The AI applies these rules at speed. It is closer to a sophisticated decision tree than to open-ended generative AI.
The strength: consistency and speed. When your NDA playbook says "flag perpetual confidentiality obligations," every NDA gets checked the same way by every reviewer. Team-level consistency is hard to achieve through training alone. Playbooks encode it into the system.
The limitation: the system works best when your contracts fit the available playbook types. Novel or highly bespoke agreements may not benefit as much. And the customization ceiling, while adequate for most standard work, may frustrate teams with unique review requirements.
Pricing context: custom pricing model, but market reports suggest small teams pay $3,000-$8,000 annually.
When to choose it: you need immediate deployment, your contracts are standard types (NDAs, MSAs, procurement, SaaS terms), and you value consistency across reviewers.
Spellbook: The In-Document Assistant
Best for: Solo practitioners and small firms that live in Microsoft Word.
Spellbook integrates GPT-4 directly into Word. This is a fundamentally different architecture from LegalOn's playbook approach. Instead of a separate platform that you upload documents to, Spellbook augments your existing drafting environment.
Suggestions appear as you draft rather than in a separate review phase. The AI spots risky clauses, suggests alternative language, flags missing provisions, and generates clause text — all inline. Ten million contracts reviewed across 4,000 legal teams in 80 countries.
The strength: zero context-switching. You never leave Word. For solo attorneys and small firms where the workflow is "open the contract in Word and start reading," this is the lowest-friction approach. The real-time suggestions during drafting genuinely change how the work feels.
The limitation: this is not built for large-scale review. If you need to process 200 contracts for a due diligence project, Spellbook's document-by-document, in-Word approach is the wrong architecture. It is also individual-focused — team review workflows with coordinated commenting and approval chains are not its strength.
Pricing context: starting around $589 per user per month on annual payment. At that price point, it adds up quickly for teams.
When to choose it: you are a solo practitioner or small firm, you work primarily in Word, you want real-time assistance during drafting, and per-seat cost at $589 per month is acceptable.
Luminance: The Pattern Recognition Engine
Best for: Large firms and enterprises handling M&A due diligence and large document portfolios.
Luminance is built on a proprietary legal LLM trained on over 150 million verified legal documents. The platform uses unsupervised machine learning to detect patterns and anomalies — a fundamentally different approach from both playbook-based review and in-document assistance.
The Legal Inference Transformation Engine (LITE) combines pattern recognition with supervised and unsupervised ML across inference, deep learning, NLP, and pattern recognition. In plain terms: instead of applying predefined rules to documents, it learns what "normal" looks like across your portfolio and flags deviations.
The strength: scale and anomaly detection. Purpose-built for reviewing thousands of documents in due diligence scenarios. Multi-language support without extensive data training. And anomaly detection is genuinely powerful — when the system has seen your standard commercial contract 500 times, it immediately identifies the one with unusual governing law, non-standard limitation of liability, or atypical force majeure.
A&O Shearman's ContractMatrix, built on Luminance, reportedly saves around seven hours from average contract review — approximately 30% efficiency gain.
The limitation: learning curve. Luminance's capabilities reward sophisticated users who understand how to configure and interpret its output. Enterprise pricing (quote-based, estimated $10-$100 per user) puts it out of reach for small teams. And using Luminance for routine NDA review is like using a Formula 1 car for grocery shopping — technically possible but wasteful.
When to choose it: M&A due diligence with large data rooms, lease portfolio analysis, regulatory compliance across document sets, or enterprise analytics with multi-language requirements.
The Architectural Comparison
| Dimension | LegalOn | Spellbook | Luminance |
|---|---|---|---|
| Core approach | Rule application | In-document generation | Pattern detection |
| Interface | Web platform | Microsoft Word | Web platform |
| Setup time | Hours | Hours | Days to weeks |
| Sweet spot | Standard contracts at volume | Solo drafting | Large portfolio analysis |
| AI type | Rules + AI | GPT-4 generative | Proprietary legal LLM |
| Team support | Good | Individual-focused | Enterprise |
| Approximate cost | $3K-$8K/year | $7K+/user/year | Enterprise quote |
What None of Them Can Do
Here is where I put the vendor comparison aside and talk about shared limitations.
Human judgment remains essential. All three tools flag issues and suggest changes. None of them can determine the right business response. Whether to accept a 24-month limitation of liability depends on the relationship, the deal size, the risk profile, and a dozen factors that no AI system currently processes.
Training data determines accuracy. Every tool performs better on contract types well-represented in its training corpus. If your contracts use unusual structures or domain-specific language, expect lower accuracy until you invest in customization.
Integration is not trivial. Getting any of these tools to work smoothly with your document management system, CLM platform, and approval workflows requires planning and often custom development.
Hallucination risk persists. Stanford research found that general AI tools hallucinate legal advice 69% of the time. Purpose-built legal AI is substantially better, but "better than terrible" is not "reliable." Every AI suggestion — whether from LegalOn, Spellbook, or Luminance — requires human verification before acting on it.
The Selection Framework
Do not choose based on which demo impressed you most. Choose based on how you actually work.
If your primary need is reviewing standard contracts consistently and quickly across a team, with minimal setup investment: LegalOn.
If your primary need is real-time drafting assistance for an individual attorney working in Word: Spellbook.
If your primary need is large-scale document analysis, portfolio intelligence, or M&A due diligence: Luminance.
If your needs span multiple categories, you may need more than one tool. That is not a failure of the evaluation — it is a recognition that different architectures solve different problems.
The worst outcome is buying a tool designed for one use case and forcing it into another. I have seen firms spend six figures on a platform that was excellent at what it was built for and terrible at what the firm actually needed it to do. Start with the workflow. Match to the architecture. Evaluate vendors within the right category.
Key Takeaways
- LegalOn's playbook architecture excels at consistent, fast review of standard contract types — best for teams needing immediate deployment
- Spellbook's Word integration provides zero-friction drafting assistance — best for solo and small-firm practitioners who live in Word
- Luminance's pattern recognition scales to M&A due diligence and large portfolios — best for enterprises with multi-language, high-volume needs
- All three require human oversight: AI identifies issues, but professional judgment determines responses
- Select based on how you actually work, not on demo impressions — architectural match to your workflow is the primary success factor