Why “best AI for lawyers” is the wrong question #
The phrase “best AI for lawyers” is itself a procurement mistake. Lawyers do six categorically different things with AI, and the best tool for each is different.
- Legal research. Best categories: legal-research AI with primary-source corpora (CoCounsel, Vincent, Lexis+ AI, Westlaw Edge AI). General-purpose AI is acceptable for surrounding analysis but not for citation lookup post-Mata.
- Drafting. Best categories: drafting AI (Spellbook, BlackBoiler) for transactional firms; general-purpose enterprise AI (Claude, ChatGPT, Gemini) for litigation drafting where flexibility outweighs polish.
- Document review and analysis. Best categories: contract-review AI (Kira, Luminance) for transactional and M&A; specialist medical-records AI (RecordsONE, Casepoint) for personal injury; e-discovery AI (Everlaw, Relativity) for litigation.
- Practice management embedded AI. Best: whatever ships with the firm's PMS. Clio Duo, MyCase IQ, Smokeball Archie. Lower marginal cost; integrated workflow.
- General-purpose work. Best: ChatGPT Enterprise, Claude for Work, Gemini for Workspace. The choice between them has narrowed; they are nearly substitutable in 2026.
- Practice-area specialists. Best: practice-specific tools where they exist. ProPlaintiff for PI; Docketwise / INSZoom for immigration; case-management AI in family-law PMS.
The first question is not “what's the best AI?” It is “what category of AI work am I trying to do?” The category decision determines the shortlist; the vendor decision is downstream.
The seven trade-offs #
Once the category is settled, vendors within the category are differentiated on seven trade-offs. Different firms weight them differently.
- Corpus quality. Tools trained on or retrieving from primary legal sources (Westlaw via CoCounsel, vLex via Vincent, Lexis via Lexis+ AI) outperform general-purpose AI on legal-domain tasks. The premium is real, but only on legal-corpus-dependent work.
- Integration depth. Native integrations with the firm's DMS (NetDocuments, iManage), PMS (Clio, MyCase), and email (Outlook, Gmail) save more time than features in many cases.
- Confidentiality posture. The data-protection addendum, the retention windows, the sub-processor list, the governmental-disclosure terms. The Rule 1.6 question.
- Pricing. $25-$1,500/seat/month spread; volume discounts above 5 seats; per-document or per-matter pricing as alternative; pilot terms.
- Training overhead. Time the firm has to invest in training attorneys and staff per Rule 1.1. Polished workflows reduce this; flexible general-purpose AI increases it.
- Vendor stability. The category is consolidating. Some vendors will be acquired; some will pivot; some will close. Procurement should weight financial health.
- Litigation defensibility. Will the AI's output survive Rule 707 scrutiny? Does the vendor produce documentation packages for litigation use? An emerging differentiator (see FRE 707 analysis).
What the listicles get wrong #
The bulk of “best AI for lawyers” content in 2026 is sponsored content, affiliate marketing, or vendor self-promotion. The patterns to recognise:
- Affiliate-driven listicles. Spellbook, Harvey, and other vendors with active affiliate programs appear in disproportionate numbers in the top-ranked “best AI” lists. The lists are revenue-driven, not buyer-driven.
- Single-vendor ‘comparisons.’ Spellbook publishes “Spellbook vs Iqidis” comparisons. Iqidis publishes “Iqidis vs Spellbook” comparisons. Each piece reaches the conclusion that the publishing vendor wins.
- Feature-list rankings without weighting. Most listicles count features. A vendor with 80 features beats a vendor with 60. The buyer rarely needs more than 20 of the features and probably wants 5 specific ones; the count is misleading.
- No mention of confidentiality. Many listicles do not mention Rule 1.6, do not mention the DPA, do not distinguish consumer-tier from enterprise-tier. These omissions are not neutral.
- Outdated information. The category moves fast. Listicles published more than 6 months ago likely list discontinued products, missed price changes, and missed major releases. Date the article you are reading.
The honest answer for common firm types #
Generic recommendations for the most common firm types in 2026, on the assumption that no single tool dominates and the firm needs a stack rather than a single product.
Solo or 2-attorney firm, mixed practice. Practice management (Clio or MyCase) with embedded AI, plus one general-purpose enterprise AI subscription (Claude for Work or ChatGPT Enterprise). Total: $130-$200/month per lawyer. Covers 70-80% of workflows.
Plaintiff's-side personal injury, 5-10 attorneys. Plaintiff-PMS (CASEpeer or SmartAdvocate or Clio) plus a specialist medical-records chronology tool (RecordsONE or Casepoint or the medical-records workflow in CoCounsel) plus a general-purpose enterprise AI. Total: $300-$700 per attorney per month. The medical-records tool pays back in the first quarter.
Defense-side civil litigation, 10-30 attorneys. Document-review AI (Everlaw or Relativity) plus legal-research AI (CoCounsel or Lexis+ AI) plus general-purpose enterprise AI. Total: $400-$900 per attorney per month. The document-review tool dominates the budget.
Transactional / corporate, 5-25 attorneys. Contract-review AI (Kira, Luminance, or Spellbook depending on draft-vs-review mix) plus PMS-with-embedded-AI plus general-purpose enterprise AI. Total: $300-$800 per attorney per month.
Immigration, solo or small. Immigration-specific PMS (Docketwise or INSZoom) with form-population AI, plus general-purpose enterprise AI for RFE responses and country-conditions work. Total: $200-$400 per attorney per month.
In-house legal department, 5-50 attorneys. Contract-lifecycle platform (Ironclad or Agiloft) plus matter-management plus general-purpose enterprise AI plus targeted research tooling. Total: $500-$1,500+ per user per month.
Each of these is a starting point. Specific firms have specific constraints (insurance carrier requirements, client-imposed restrictions, legacy system commitments) that modify the answer.
The minimum-viable AI stack #
For lawyers asking “what's the cheapest credible setup?”, the minimum-viable stack in 2026 is:
- One general-purpose enterprise AI ($25-$60/month). ChatGPT Enterprise, Claude for Work, or Gemini for Workspace. Verify the data-protection addendum is executed.
- Practice management with embedded AI ($80-$150/month). Most modern PMS platforms ship with at least a baseline AI feature set; if the firm already has PMS, this is a no-incremental-cost addition.
That's it. $130-$210 per lawyer per month. This stack covers research, drafting, document analysis, matter summarisation, and intake at a competence level that satisfies Rule 1.1 with appropriate training. It does not cover specialist workflows (medical-records chronology, e-discovery review, contract-lifecycle management); those are upgrades the firm makes when the work-mix justifies them.
The mistake firms make is over-buying in the first year. The right pattern: minimum-viable in months 1-3, add specialist tools in months 4-12 as specific work-mix gaps become visible.
The 2026 trend: practice-area specialists pulling ahead #
Through 2024 and most of 2025, the dominant trajectory was horizontal platforms. The bet was that one tool would win all the work types. By mid-2026, that bet looks wrong.
The fastest-growing tools in 2026 are practice-area specialists: ProPlaintiff and similar for personal injury; Docketwise and INSZoom in immigration; specialised IP tools for prior-art search; specialised criminal-defense tools for Brady review and constitutional-issue surfacing. They beat horizontal AI on workflow fit because they are built around a single practice's data shapes and user expectations.
The Reddit-validated phrase from the legal-tech community is “scalpel AI over Swiss Army knife.” A Category-2 contract review tool that does only contract review beats a horizontal platform that does contract review plus six other things, because the specialist gets the contract-review-specific UX right.
What this means for procurement: the right answer for many firms is a hybrid stack. Horizontal AI for general work (research, drafting, summarisation), plus specialist AI for the firm's core practice. Not one tool. A stack.
Vendor stability as a procurement criterion #
Through 2024 the legal-AI category looked like a stable startup-and-incumbent story. Through 2025 it began to consolidate (Casetext acquired by Thomson Reuters, Kira acquired by Litera). Through 2026 the consolidation is accelerating. Some vendors will not be here in 24 months.
The procurement implication: weight vendor financial health alongside features. The minimum diligence:
- Funding round timing and size (public information for VC-backed vendors)
- Customer-count trajectory (often disclosed in marketing materials, sometimes verifiable)
- Acquisition-by-incumbent risk (a flag for and against, depending on whether the buyer wants the incumbent integration)
- Open-source or proprietary stack (proprietary stacks fail harder; open-source stacks have migration paths)
- Data-export terms in the contract (can the firm leave with its data, intact and usable?)
This evaluation will not stop a firm from picking a vendor that turns out to be acquired or shut down; it will reduce the cost of the transition when it happens. Data-export terms are the single most important clause from this perspective.
Frequently asked.
What's the single best AI tool for lawyers in 2026?
There isn't one. The right tool depends on which of six task categories you're solving for: legal research (CoCounsel, Vincent, Lexis+ AI), drafting (Spellbook, general-purpose enterprise AI), document review (Kira, Luminance, Everlaw, RecordsONE depending on document type), practice management (Clio, MyCase with embedded AI), general-purpose (Claude for Work, ChatGPT Enterprise), or practice-area specialist (ProPlaintiff, Docketwise). The category determines the shortlist.
Is ChatGPT Enterprise good enough for legal work?
For general drafting, brainstorming, document analysis, and surrounding work: yes, with the data-protection addendum executed. For legal research with verifiable citations: no — use a primary-corpus tool (CoCounsel, Vincent, Lexis+ AI) for citation work to satisfy the Mata v. Avianca verification standard.
How much should a solo lawyer spend on AI tools?
$130-$210 per month covers the minimum-viable stack: one general-purpose enterprise AI plus practice management with embedded AI. That handles roughly 70-80% of typical solo workflows. Add specialist tools as specific gaps appear; the right pattern is incremental, not all-at-once.
Should I pick the most popular tool?
Popularity is a signal but not a procurement criterion. Harvey is popular at large firms because it markets to large firms; CoCounsel is popular because Westlaw integration matters to litigators; Spellbook is popular among transactional solos because the workflow fit is good. Match the popularity signal to your firm type. Generic 'most popular' rankings are usually marketing.
What if my preferred tool gets acquired or shuts down?
Read the data-export terms in the contract before signing. Vendors with strong export terms (your data leaves with you, in a usable format, on demand) survive consolidation gracefully from the customer's perspective. Vendors with weak export terms are the lock-in problem. The single most important contract clause from a vendor-stability perspective.
Citations and further reading.
Dan Hughes is the founder of IXSOR. Twenty years in the UK, twenty-five in North Carolina. Ex-BBC. Ex-Apple. Lifelong technologist. And, most importantly: not an attorney. He writes about legal AI from the operational and infrastructure side, where the rules meet the machines. Reach: [email protected]. More by Dan →