What AI tools do law firms actually use?

Law firms in 2026 use three categories of AI tools: legal-specific products like Harvey, Luminance, and CoCounsel for specialised tasks; general-purpose large language models like ChatGPT, Claude, and Microsoft Copilot for research and drafting; and custom-built systems for firm-specific workflows like client intake, document generation, and matter management. Most firms use a combination. The right mix depends on firm size, practice areas, budget, and whether you need broad capability or deep automation of specific workflows.

Short answer: Law firms use legal-specific AI (Harvey, Luminance, CoCounsel), general LLMs (ChatGPT, Claude, Copilot), and custom-built systems. The choice depends on whether you need off-the-shelf capability or systems built for your processes.

Why this question matters now

The number of AI tools available to law firms has grown from a handful in 2023 to over 150 products in 2026. The market is noisy, and most managing partners cannot tell which tools deliver genuine value and which are demos wrapped in marketing copy.

The adoption data tells a clear story about what firms actually use versus what gets the most press coverage. A 2025 LawTech UK survey found that 62% of solicitors used general-purpose LLMs (ChatGPT, Claude) for work-related tasks at least weekly. Only 18% used legal-specific AI products. And only 11% had access to custom-built AI systems. The pattern is similar in the US, where a 2025 ABA Tech Report found 58% individual AI adoption and 24% firm-level deployment.

The gap between individual adoption and firm-level deployment is the critical issue. Solicitors and associates using ChatGPT on their personal devices is not the same as the firm deploying AI at the workflow level. Individual use creates inconsistency, compliance risk, and no measurable ROI. Firm-level deployment creates efficiency, compliance, and measurable returns.

Understanding what tools exist, what they actually do, and how they compare is the first step toward moving from individual experimentation to firm-level deployment.

What are the main categories of AI tools for law firms?

Legal-specific AI products

These are purpose-built for legal work. They have legal training data, understand legal concepts, and are designed for specific legal tasks.

Harvey AI. The most prominent legal AI product. Uses OpenAI’s models fine-tuned on legal data. Primary use cases: legal research, contract analysis, memo drafting, and due diligence. Strengths: deep legal knowledge, firm-level deployment, strong security infrastructure. Limitations: enterprise pricing (typically £100+ / $130+ per user per month), best suited to large firms, requires significant onboarding.

Luminance. Specialises in contract analysis and due diligence. Uses proprietary AI models trained on legal documents. Strengths: pattern recognition across large document sets, anomaly detection, multi-language support. Limitations: focused use case (not a general legal AI), pricing favours high-volume users.

CoCounsel (by Thomson Reuters). Integrated into Westlaw. Handles legal research, document review, summarisation, and drafting. Strengths: direct access to verified legal databases, reducing hallucination risk. Limitations: tied to the Westlaw ecosystem, additional cost on top of existing Westlaw subscriptions.

Clio Duo. Integrated into the Clio practice management platform. Handles client communication drafting, time entry suggestions, and basic document generation. Strengths: native integration with a popular practice management system. Limitations: limited to Clio users, narrower capability than standalone legal AI products.

General-purpose LLMs

These are not built for legal work specifically, but lawyers use them extensively.

ChatGPT (OpenAI). The most widely used LLM. Enterprise tier offers data processing agreements and no training on your data. Used for first-draft research memos, email drafting, argument structuring, and summarisation. Strengths: broad knowledge, fast iteration. Limitations: hallucination risk on case citations, no verified legal database.

Claude (Anthropic). Growing adoption in legal. Known for longer context windows (useful for analysing lengthy documents) and more cautious responses. Enterprise tier available. Strengths: document analysis, nuanced reasoning. Limitations: same hallucination risks as any general LLM for case citations.

Microsoft Copilot. Integrated into Microsoft 365. Used for drafting in Word, email management in Outlook, and data analysis in Excel. Strengths: works within the tools lawyers already use. Limitations: general-purpose, not legal-specific; data governance concerns if not on the enterprise tier.

Custom-built AI systems

These are bespoke systems built for a specific firm’s workflows. They connect to the firm’s practice management platform, use the firm’s templates, follow the firm’s processes, and handle the firm’s specific data formats.

Common custom builds:

• Client intake automation (form processing, conflict checks, routing)
• Document drafting (template-based generation from structured inputs)
• Matter management automation (deadline tracking, task assignment, progress updates)
• Bundle preparation (document collation, indexing, formatting for court)
• Compliance checking (SRA or state bar rule verification, regulatory filing)

Strengths: tailored to your exact workflow, integrated with your systems, you own the code. Limitations: higher upfront cost than subscriptions, requires a consultancy or internal team to build.

How should a law firm choose between these options?

The decision framework is straightforward:

Use general LLMs when you need flexible, ad-hoc AI assistance across many tasks. Every fee earner should have access to an enterprise-tier LLM for research, drafting, and summarisation. Cost: £15-50 / $20-65 per user per month.

Use legal-specific products when you have a high-volume specialised task (due diligence, contract review, legal research) that justifies the subscription cost. These products add value when the volume of the specific task is high enough to warrant a dedicated tool.

Build custom systems when you have a specific workflow that you process at high volume and that is not well-served by off-the-shelf products. Client intake, document drafting from your templates, and practice management automation are the most common custom builds. The ROI is highest when the workflow is unique to your firm or your practice area.

The practical approach for most mid-market firms:

1. Deploy enterprise ChatGPT or Claude for all fee earners (immediate, low cost)
2. Evaluate one legal-specific product for your highest-volume specialised task
3. Build one custom system for your highest-volume operational workflow (intake is usually the best starting point)
4. Measure ROI on each, expand based on evidence

What we’ve seen at Formulaic

The firms we work with typically arrive having deployed general LLMs at the individual level but having struggled to move beyond that. They have tried one or two legal-specific products, found them valuable for specific tasks but insufficient for workflow-level automation, and are now looking for custom systems to close the gap.

The pattern that delivers the best results: general LLMs as the base layer (everyone gets access), one or two legal-specific products for specialised tasks (document review, research), and custom-built systems for the firm’s highest-volume operational workflows (intake, drafting, matter management).

When we built an intake system for a family law practice, it replaced a process that involved three different tools (a web form, a shared inbox, and manual CRM entry) with a single system that handles everything from initial enquiry to practice management record creation. The firm had tried using ChatGPT for parts of this workflow and found it helpful but fragmented. The custom system eliminated the fragmentation. Processing time dropped by 70%, and the conversion rate from enquiry to client increased by 22% because no enquiries were lost in the gaps between tools.

The lesson is consistent: individual AI tools are useful but limited. The real value comes from systems that connect tools to workflows. That connection is what turns scattered AI experiments into production capability.