Legal AI Tools for Small and Mid-Sized Firms: A Grounded Survey

A small firm does not need to predict where legal artificial intelligence is going. It needs to understand the few categories of tools that exist today, what each one is actually for, and the professional duties that attach the moment any of them touches a client matter.

The market can feel crowded, but most products sort cleanly into four buckets: AI-assisted legal research, contract and transactional assistants, practice-management AI, and general business-tier chatbots. A 3-to-30-attorney firm rarely needs one tool from every bucket at once. It needs the right one or two, adopted deliberately, with verification built into the workflow. What follows is a grounded survey, not a ranking, and the cited authorities are the part that will still be true after the product names change.

The four categories, briefly

The useful first cut is by function, because function determines both the value and the risk.

• AI-assisted legal research answers legal questions against a curated database of cases, statutes, and secondary sources.
• Contract and transactional assistants draft, review, and redline agreements, often inside a word processor.
• Practice-management AI sits on top of your matter, billing, and document data to handle intake, deadlines, time entry, and routine client communication.
• General business-tier chatbots are horizontal tools (the same ones other industries use) that you point at legal work.

These categories overlap at the edges, and the larger vendors are deliberately blurring them, but the distinctions still tell you what a tool is built to do well and where it is likely to fail.

AI-assisted legal research

This is the category most firms think of first, and it is dominated by the two incumbents. Thomson Reuters acquired the legal AI startup Casetext for 650 million dollars in a deal that closed in August 2023, and folded its CoCounsel assistant into the Westlaw franchise. The research feature most relevant to firms is Westlaw Precision AI-Assisted Research, which takes a natural-language question and returns a synthesized answer grounded in Westlaw's content. LexisNexis offers the competing Lexis+ AI, now packaged with an assistant it markets as Protégé, covering conversational research, document analysis, and drafting.

The selling point of both is retrieval-augmented generation: the model is supposed to answer only from a trusted, licensed database rather than from its own training. That design reduces fabrication, but it does not eliminate it. In the most cited empirical work on this question, Stanford's RegLab and Human-Centered AI institute published "Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools" (Magesh et al.), released as a preprint in May 2024 and later peer-reviewed in the Journal of Empirical Legal Studies in 2025. The study found that the leading AI legal research tools from LexisNexis and Thomson Reuters hallucinated between roughly 17 and 33 percent of the time, depending on the product and the measure. That is a meaningful improvement over a general chatbot and nowhere near the "hallucination-free" framing some marketing implied.

The practical lesson for a small firm: these are legitimately among the best AI tools for law firms doing research, and they are built on authoritative content you already trust. They are not a substitute for reading the cases. Treat the synthesized answer as a fast, sometimes-wrong first pass, and pull and read every authority before it goes in a brief.

Contract and transactional assistants

For transactional and small-business firms, the higher-value category is often drafting and review. Spellbook is the clearest example aimed squarely at smaller practices: it runs as a Microsoft Word add-in and uses large language models to draft clauses, suggest redlines, and flag missing or aggressive terms inside the document you are already editing. The appeal is workflow. The lawyer never leaves Word, and the tool works against the contract in front of it rather than a separate web app.

CoCounsel, now a Thomson Reuters product, also performs contract analysis, document review, and deposition preparation alongside its research features, which is part of why the research-versus-transactional line is blurring. At the top of the market sits Harvey, an enterprise platform built for large law firms and corporate legal departments, which raised at a reported valuation of about 11 billion dollars in a 2026 funding round. Harvey is worth knowing about for context, but its enterprise pricing and deployment model put it outside what most 3-to-30-attorney firms will buy. The point of naming it is to show the shape of the market: enterprise platforms for the largest firms, and Word-native or browser-native assistants for everyone else.

For a small firm, the realistic move in this category is a focused drafting assistant for the contract types you handle repeatedly, evaluated on whether it actually catches what an associate would catch, not on demo polish.

Practice-management AI

The quietest category may be the most consequential for firm economics, because it touches the work that is hard to bill and easy to drop. Clio, the dominant cloud practice-management platform for small firms, released Clio Duo, generative AI built directly into its platform, in October 2024, and has continued to expand it into operational tasks such as surfacing matter insights, drafting routine communications, and assisting with time and billing entries. Because the AI lives inside the system that already holds your matters, calendar, and documents, it can do things a standalone chatbot cannot, such as extracting deadlines from a filing or drafting an invoice narrative from recorded activity.

The same proximity is the risk. Practice-management AI operates directly on client data and on the records that determine your deadlines and your bills. An extracted date that is wrong is a malpractice exposure, not a typo. The right posture is to use these features for drafting and suggestion while keeping a human approval step on anything that creates a deadline, sends a client a substantive message, or finalizes a bill.

General business-tier chatbots

The last category is the one your associates are probably already using, with or without a policy: ChatGPT (OpenAI), Claude (Anthropic), Microsoft Copilot, and Google Gemini. These are horizontal tools, not legal products, and the distinction that matters is which tier you are on.

Consumer free tiers and the contractual terms of business and enterprise tiers are not the same, and the difference is central to your confidentiality duty. The paid business and enterprise offerings (for example, ChatGPT Enterprise, Claude's team and enterprise plans, and Microsoft 365 Copilot under its commercial data-protection terms) generally commit not to train their models on your inputs and offer stronger data-handling and administrative controls. Read the actual data-processing terms of the specific tier you buy rather than relying on the brand. These tools are genuinely useful for non-privileged work: summarizing a long document you paste in, restructuring a memo, drafting a client-update email for your review, or brainstorming arguments. They are the worst possible choice for legal research, because a general chatbot with no connection to a legal database will invent citations confidently. That failure mode is exactly what produced the sanctions discussed below.

The ethics floor every tool must clear

No tool, in any category, changes a lawyer's professional obligations. The governing reference is ABA Formal Opinion 512, issued July 29, 2024, the ABA's first formal ethics guidance on generative AI. It maps the relevant Model Rules onto AI use, and three points matter most for adoption decisions:

• Competence (Rule 1.1). A lawyer must understand, at a reasonable level, the benefits and limitations of the specific AI tool used, and must keep that understanding current.
• Confidentiality (Rule 1.6). Opinion 512 cautions that lawyers should evaluate the risk that client information could be disclosed or accessed, and indicates that self-learning AI tools may require a client's informed consent before client confidences are entered. The opinion also states that generic, boilerplate consent in an engagement letter is generally not sufficient.
• Fees (Rule 1.5). A lawyer generally may not bill a client for the time spent learning to use an AI tool as a matter of general competence, and where AI produces real efficiencies, the lawyer should consider how billing reflects the actual time spent.

State guidance points the same direction. The State Bar of California published its Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law on November 16, 2023, and the Florida Bar approved Ethics Opinion 24-1 on January 19, 2024. Both permit AI use while emphasizing confidentiality, competence, candor, supervision, and honest billing. If you practice in another state, check your own jurisdiction, because more bars have issued guidance since.

The supervision duty (Rules 5.1 and 5.3) deserves its own line. AI output is work product produced under your name, and it must be reviewed exactly as you would review a junior lawyer's draft.

What happens when the floor is ignored

The cautionary cases are now concrete, and they all share one fact pattern: a lawyer relied on AI output and filed it without verifying the authorities. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), Judge P. Kevin Castel sanctioned two attorneys and their firm in June 2023 after they submitted a brief citing cases that ChatGPT had fabricated, imposing a 5,000 dollar penalty and finding the lawyers had acted in bad faith in defending the fake citations. In Park v. Kim, 91 F.4th 610 (2d Cir. 2024), the Second Circuit referred an attorney to its grievance panel after she cited a nonexistent case generated by ChatGPT, holding that counsel must at a minimum read the authorities they cite. These are not exotic outcomes. They are the predictable result of skipping verification, and courts have been consistent about it.

A practical adoption path

A small firm can move deliberately without falling behind. A workable sequence:

• Start with one category and one real workflow. Pick the task that costs you the most time, then choose a tool built for it rather than a general one.
• Buy the business or enterprise tier and read its data terms. Confirm in writing that your inputs are not used for training and that the vendor's security posture fits your confidentiality duty.
• Write a one-page AI policy. Say which tools are approved, what data may never be entered, and that all output is verified before use.
• Keep verification non-optional for legal authority. Every citation gets pulled and read. No exceptions, regardless of which tool produced it.
• Train the people who will use it, and address client communication and consent where the matter and your jurisdiction call for it.
• Reassess quarterly. This market moves fast enough that last quarter's choice deserves a fresh look.

Adopted this way, the current generation of legal AI tools can absorb real drudgery from a small firm. The firms that get into trouble are not the ones that adopt slowly. They are the ones that let a tool stand in for the lawyer's own judgment and review.

This is general information for lawyers and law-firm leaders, not legal advice, and it does not create an attorney-client relationship. The authorities are cited so you can read them yourself.