Three business models behind three legal AI assistants
Firms that shortlist these three products expect to compare chatbots. They're really comparing business models. Harvey is a standalone generative AI platform that works on your firm's own documents and connects to other publishers' content. CoCounsel is Thomson Reuters' assistant, and its research value comes from Westlaw and Practical Law. Lexis+ AI is LexisNexis' assistant, built into the Lexis+ research subscription alongside its Protégé agent.
The underlying language models matter less than demos suggest. Each vendor draws on frontier models from major providers and updates them as better ones ship, so any model advantage tends to be temporary.
The useful version of the Harvey AI vs CoCounsel question, with Lexis+ AI added, comes down to four things:
- Content: whose case law, legislation and practice guidance the tool can cite in the jurisdictions where you practice.
- Task fit: whether your lawyers mostly need research answers or drafting and document review.
- Security terms: what the contract says about data retention, model training, hosting region and separation between clients.
- Commercial model: whether you're buying a new platform or adding a module to a research subscription you already pay for.
One development shifts the comparison. Harvey and LexisNexis have a strategic alliance that brings LexisNexis primary law and citation checking into Harvey for customers who license both. Harvey is no longer purely a drafting tool with no authority behind it, but that access depends on what you license and where you practice. Confirm which content is enabled for your jurisdictions before you count on it.
We haven't tested these products ourselves. This analysis draws on vendor documentation, published product scope and how procurement for these tools usually runs. Use it to build a shortlist and a pilot plan, not to replace either.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Research content coverage by jurisdiction
An AI research answer is only as good as the authority under it. A fluent answer that cites nothing verifiable is a liability, which is why jurisdictional coverage should be your first filter, ahead of interface or speed.
| Region | CoCounsel (Thomson Reuters) | Lexis+ AI (LexisNexis) | Harvey |
|---|---|---|---|
| United States | Deepest: Westlaw primary law, KeyCite, Practical Law | Deep: Lexis primary law, Shepard's, Practical Guidance | Via the LexisNexis alliance where licensed; otherwise your own documents and public sources |
| UK and Ireland | Westlaw UK and Practical Law UK content where offered | Lexis+ UK content and practice guidance | Used by a number of large UK and international firms; research authority depends on content partners |
| Canada, Australia, New Zealand | Regional Westlaw and Practical Law products in several markets | Regional Lexis content in several markets | Used by firms in these markets; authority coverage varies |
| Continental Europe | Patchier; depends on local Thomson Reuters publishing | Selected markets, such as France, through local LexisNexis products | Some EU sources plus your own materials; multilingual drafting is a strength |
| India, Southeast Asia, Middle East, Africa, Latin America | Limited and market-specific; confirm before buying | Local publishing presence in several of these markets; confirm whether AI features are switched on | Works on the documents you upload; local primary law coverage is generally thin |
Two caveats. First, a publisher holding content in a country doesn't mean its AI assistant can search it. AI features typically launch in the US first and reach other markets later, sometimes much later. Second, in common-law markets with smaller publishing ecosystems, such as Singapore, Malaysia, Kenya, Nigeria or the Gulf financial free zones, you may rely on a local research provider for authority and use one of these platforms only for drafting and review.
Ask each vendor for a written list of the databases its AI can cite in every jurisdiction you need, and attach it to the contract as an exhibit. A slide isn't enough.
Where each tool is strongest: drafting versus research
The three products overlap more with each release, but each still leans one way.
Harvey: drafting, review and firm-specific workflows
Harvey does its best work on your own material: reviewing a data room, checking a hundred leases against a playbook, producing a first-draft memo from transaction documents, or turning a partner's preferred clause positions into a repeatable workflow. Its Vault feature handles large document sets, and firms can build custom workflows around their own precedents. Multilingual work is another strength, which matters for cross-border teams. The weak spot is authority. Unless LexisNexis content is licensed and enabled for your jurisdiction, research answers need independent verification.
CoCounsel: research tied to Westlaw, plus practical tasks
CoCounsel's answers link back to Westlaw authority and KeyCite treatment, and it can draw on Practical Law know-how. Its deep research mode is built for multi-step questions, and its skills cover document review, timelines, deposition preparation and contract policy checks. For firms already living in Westlaw and Practical Law, the main gain is less switching between tools.
Lexis+ AI: research with citation checking built in
Lexis+ AI is the most research-first of the three. It offers conversational search over Lexis content, Shepard's-linked citations, summaries of uploaded documents and drafting grounded in Practical Guidance. Protégé adds personalization and agent-style tasks. Drafting is good enough for memos, letters and clauses, but the product isn't built around bulk transactional review the way Harvey is.
Hypothetical worked example: say a 60-lawyer firm splits 70% litigation and 30% corporate. Litigators spend most of their AI time on research and motion drafting; corporate uses it for diligence a few times a quarter. Here, CoCounsel or Lexis+ AI covers most of the demand, and an existing contract review tool handles diligence. Reverse the split and Harvey moves to the top of the list.
Enterprise security terms to negotiate
All three vendors market the same baseline: no training on customer data, encryption in transit and at rest, and independent attestations such as SOC 2 Type II or ISO 27001. Treat that as the floor. What protects you is in the contract, and the details vary by country.
Data protection law shapes the negotiation. Firms under the EU GDPR or UK GDPR need a data processing agreement and a lawful transfer mechanism if data leaves the region. India's Digital Personal Data Protection Act, Brazil's LGPD, Singapore's PDPA, South Africa's POPIA and data laws in the UAE and Saudi Arabia each add their own transfer and notification rules. Government, financial-services and defense clients often impose contractual data residency on top of any statute. Ask each vendor which hosting regions are live today, not which are on the roadmap.
Security review checklist:
- Model provider retention: written zero-data-retention terms with every underlying model provider, plus advance notice before a new provider is added.
- Training: an explicit ban on using prompts, outputs or uploaded documents for training or fine-tuning, binding subcontractors too.
- Hosting region: named regions for storage and processing, and the locations from which support staff can access data.
- Ethical walls: permissions that mirror your matter-level security, so one client's workspace can't surface another client's documents.
- Retention and deletion: configurable retention periods and deletion certificates when a matter closes or the contract ends.
- Audit logs: exportable records of who queried what, for client audits and disputes.
- Breach notice: vendor notification deadlines short enough for you to meet your own obligations, which can be as tight as 72 hours under GDPR-style rules.
Red flags: a vendor that won't name its model providers, security commitments that live only on a web page the vendor can edit, hosting described as "global" with no named regions, or a key client's outside counsel guidelines that ban the tool outright. Have your general counsel or outside data protection counsel review the final terms, because requirements differ by jurisdiction and by client.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
How Harvey, CoCounsel and Lexis+ AI are sold and priced
None of the three publishes enterprise prices, and real deals swing widely with seat count, term length, bundling and region. The figures below are typical 2026 market ranges in USD and vary by contract. Local pricing, currency and tax treatment differ by country, so confirm with each vendor.
| Factor | Harvey | CoCounsel | Lexis+ AI |
|---|---|---|---|
| How it's sold | Direct enterprise sale; firm-wide or practice-group deployments | Add-on or bundle with Westlaw and Practical Law | Add-on or tier within a Lexis+ subscription |
| Indicative cost per user per month | Roughly $150–$500+, usually with seat minimums | Roughly $100–$450, on top of or bundled with research | Roughly $100–$400, on top of or bundled with research |
| Typical term | Annual or multi-year | Multi-year, often aligned to the research renewal | Multi-year, often aligned to the research renewal |
| Main leverage point | Seat count, pilot results, practice-group scope | Timing of the Westlaw renewal | Timing of the Lexis renewal |
What drives the cost:
- Seat model: per-user versus firm-wide licensing. Firm-wide looks expensive until you count how many lawyers would otherwise need a seat.
- Bundling: both publishers price their AI more attractively when it's tied to a multi-year research renewal. That's good value if you'd renew anyway and a lock-in if you wouldn't.
- Content add-ons: using publisher content inside Harvey generally means paying for both licenses.
- Usage caps: some plans limit heavy features such as bulk review or deep research. Ask what happens when you exceed them.
Hypothetical worked example: say a 40-lawyer firm already pays for Westlaw. Adding CoCounsel for all 40 lawyers at an assumed $200 per seat per month costs about $96,000 a year. Harvey for 15 transactional lawyers at an assumed $350 per seat per month costs about $63,000 a year, and the firm still pays for Westlaw. The smaller line item isn't automatically the better deal. Compare cost per matter-hour saved during a pilot.
Verdicts for global firms, mid-size firms and in-house teams
Global and large multi-office firms
Harvey is the strongest candidate for the firm-wide platform, provided you pair it with authoritative research content. Large firms get the most from generative AI in workflows on their own documents, multilingual drafting and firm-specific tools shared across offices, which is exactly Harvey's ground. Keep your main research subscription. If it's Lexis, push for the alliance content to be enabled in every jurisdiction you need.
At this size, Harvey AI vs CoCounsel is rarely either-or. Expect to run Harvey for transactional and review work and a publisher assistant for research. Budget for both, and make Harvey demonstrate value beyond the publisher assistant you'd be offered at renewal anyway.
Mid-size firms (roughly 20–200 lawyers)
For most mid-size firms, the better first purchase is the AI module from your existing research publisher: CoCounsel on Westlaw, Lexis+ AI on Lexis. You get cited answers, a single vendor relationship, renewal leverage and less change for lawyers to absorb. Switching publishers to get a different assistant rarely pays off once retraining and migration are counted. Consider Harvey only if transactional review is a large share of billable work and a pilot shows clear time savings.
In-house legal teams
In-house teams do less case-law research and more contract review, policy questions and multi-country advice. That tilts toward Harvey or a contract-focused tool, with Practical Law or Practical Guidance content worth more than case law. If you already license Practical Law, CoCounsel's integration is a strong option. Teams in markets with little publisher AI coverage should weight drafting and review above research features.
Decision rule: run a 60- to 90-day paid pilot with two finalists on 20–30 real, low-risk tasks from your own matters. Score each on citation accuracy, time saved and how often lawyers go back to it unprompted. Sign with the winner on those numbers, and only once the security terms above are in writing.