AI in the legal industry: what is real in production, by use case

AI in the legal industry is now a daily working tool: adoption nearly doubled in a year, even as courts kept sanctioning attorneys for unverified AI output. This guide maps the real AI use cases in legal by business function, shows the adoption data, and separates routine production work from the marketing claims, with the compliance landmines handled honestly.

By Kanika Mathur, Head of Service Delivery

Reviewed by Resourcifi engineeringPublished Jun 27, 2026Updated Jun 27, 202611 min read

Legal

Key takeaways

The short version

AI use cases in legal cluster around document-centric work: document review (74%), legal research (73%), and summarization (72%) are the highest-adoption uses, ahead of drafting and contract review.
Adoption is steep and recent. Thomson Reuters found organizational GenAI use in legal nearly doubled, from 14% in 2024 to 26% in 2025, with 72% of users running it weekly or more.
The honest framing: AI in legal is a drafting, research, and review accelerator under mandatory attorney supervision. It compresses associate and paralegal hours, but a licensed lawyer owns every output that leaves the building.
The accuracy risk is real even in purpose-built tools. A Stanford study found leading legal-research products still hallucinate, Lexis+ AI more than 17% and Westlaw AI-Assisted Research more than 34% of the time.
Courts are sanctioning attorneys for unverified citations, and ABA Formal Opinion 512 makes verification a non-delegable duty. Build for citation grounding, confidentiality, and human review from day one.

What AI in the legal industry actually looks like in 2026

AI in the legal industry is mostly document-centric work that a licensed attorney still signs off on, not an autonomous lawyer. Generative AI has moved from experiment to habit: Thomson Reuters found organizational adoption nearly doubled in a year, from 14% in 2024 to 26% in 2025, and 72% of current users now run it weekly or more, with over 40% using it multiple times a day.¹ The work genuinely happening is concentrated in review, research, summarization, and drafting; the autonomous AI lawyer that marketing promises remains out of reach.

The split is clean. The real, in-production uses are document review, legal research, summarization, and first-draft generation, plus e-discovery, which courts have accepted since 2012.⁷ The hype is the rest: autonomous filing without a human, and hallucination-free research. A Stanford study found purpose-built legal-research tools still hallucinate at meaningful rates and concluded that vendors' hallucination-free claims are overstated.² The honest 2026 framing is that AI in legal is a drafting, research, and review accelerator under mandatory attorney supervision. It compresses hours of associate and paralegal work, and a lawyer owns every output.

There is a gap worth naming up front: adoption is outrunning governance. Only 20% of legal organizations measure GenAI return on investment, 52% have no usage policy, and 64% of professionals have had no training.¹ The risk is not just bad tools. It is ungoverned good tools, which is the exact problem retrieval-grounded, verifiable legal software is built to remove.

AI use cases in legal, by business function

The strongest way to scope AI in a legal team is by function. The table below maps twelve recurring use cases across the legal value chain, what the AI actually does, how mature each is, and the honest caveat that comes with it. Adoption figures are from a named survey; productivity and time-savings figures are vendor-reported and labeled as such. Read every row as a task that accelerates a lawyer, never one that decides for them.

The pattern underneath all twelve is the same. Each is a retrieval-grounded, human-in-the-loop application rather than a raw chatbot: ground the answer in authoritative sources, attach a verifiable citation, filter privilege at the data boundary, log everything, and gate output behind attorney review. Knowledge and precedent retrieval over a firm's own document management system is the cleanest starting point, because the corpus is the firm's own vetted material.

AI use cases in legal, by function

Twelve use cases across research, drafting, transactional, litigation, and operations work. Maturity is verified where a named survey reports adoption; time-savings figures are vendor-reported and flagged.

AI use cases in legal, grouped by business function
Function	What AI does	Maturity	Honest caveat
Legal research	Find case law and statutes, answer doctrinal questions with citations	73% adoption	Leading tools hallucinate 17 to 34%; verify every cite
Document review	Summarize, classify, and extract facts across document sets	74%, the top use	Summaries can omit or distort; spot-check the source
Document summarization	Condense depositions, contracts, and case files into briefings	72% adoption	Privileged material must not enter an uncontrolled model
Drafting (briefs, memos, letters)	First drafts in firm templates and voice	59% memos, 50% letters	Rule 3.3 candor: never file an unverified citation
Contract drafting and review (CLM)	Generate clauses, redline against a playbook, flag risk	51%; 50 to 90% faster review (vendor)	Edge-case clauses still need transactional judgment
Due diligence (M&A)	Surface liabilities and clauses across data rooms	Widely deployed (vendor)	Treat as triage, never the final read
E-discovery / litigation review	Prioritize relevant, responsive, and privileged docs (TAR)	Court-accepted since 2012	The process must be defensible, not just the result
Litigation analytics	Judge, court, and outcome analysis to inform strategy	Established category (vendor)	Correlation informs, it does not decide
Compliance monitoring	Interpret regulatory text, map and track obligations	Growing in-house (vendor)	A qualified lawyer must validate any flagged obligation
Intake, triage, and legal Q&A	Classify matters, route, answer routine questions	Common in legal ops (vendor)	Unauthorized-practice and accuracy risk on public answers
Legal ops and billing	Draft time narratives, automate matter admin, analyze spend	Rising investment (vendor)	Rule 1.5: do not bill saved time as if performed manually
Knowledge and precedent retrieval	RAG over the firm's own document system for prior work product	Widely pursued, low risk	Enforce access controls and ethical walls at retrieval

Sources: adoption figures from Thomson Reuters (2025); e-discovery acceptance from Da Silva Moore v. Publicis Groupe (S.D.N.Y. 2012), per Proskauer; contract-review time savings are vendor-reported (Thomson Reuters buyer's guide, 2025). Rows marked "vendor" are directional and have not been independently audited.

How widely AI is used in legal

AI in legal is no longer fringe. Across the legal professionals Thomson Reuters surveyed in 2025, document review leads at 74%, with legal research at 73% and summarization at 72%, then drafting briefs and memos at 59%, contracts at 51%, and correspondence at 50%. These six are where AI use cases in legal are most concentrated today, and all six are document-centric.

The trajectory matters more than any single number. Organizational adoption nearly doubled in one year, from 14% in 2024 to 26% in 2025, law firms ran slightly ahead of corporate legal departments, and 95% of professionals expect GenAI to be central to their workflow within five years.¹ A separate 2026 survey by 8am reported that 69% of individual legal professionals now use generative AI for work.³ The two surveys measure different things, one organizational active use and the other any individual use, so treat them as two reputable readings of the same upward curve instead of a single line.

GenAI adoption in legal, by use case (2025)

Share of legal professionals using generative AI for each task, from one survey so the bars are directly comparable. Document-centric work dominates the top of the list.

Data behind this chart
Use case	Using GenAI
Document review	74%
Legal research	73%
Document summarization	72%
Drafting briefs and memos	59%
Drafting contracts	51%
Drafting correspondence	50%

Source: Thomson Reuters, 2025 Generative AI in Professional Services Report, executive summary for legal professionals (2025), 1,700+ respondents across the US, UK, and Canada.

Limitations, risk, and compliance

This is the section that decides whether legal AI is safe to deploy. Purpose-built tools still hallucinate, courts are actively sanctioning attorneys for unverified output, and ABA Formal Opinion 512 places non-delegable verification, candor, confidentiality, supervision, and fee duties on the lawyer. So AI in legal augments lawyers and never replaces attorney review. An application can draft, research, and cite. Only a licensed attorney can take responsibility for what gets filed or advised.

Hallucinated citations are documented and sanctionable. In Mata v. Avianca, the Southern District of New York sanctioned plaintiff's attorneys 5,000 dollars in 2023 after they filed a brief citing six non-existent cases generated by ChatGPT, and the court stressed that the deeper failure was not catching the error and then defending the fake cases.⁴ The problem scaled into 2025: a Westlaw review identified 22 US court cases with non-existent AI-generated citations in roughly one summer month, with fines and mandatory AI continuing-education orders.⁵

Even leading legal-AI research tools hallucinate. "It is a legal product, so it is safe" is false. The peer-reviewed Stanford RegLab study tested more than 200 queries and found Lexis+ AI hallucinated more than 17% of the time and Westlaw AI-Assisted Research more than 34%, and that providers' hallucination-free claims are overstated.² The table makes the accuracy reality concrete.

The accuracy reality, even in purpose-built tools

Stanford RegLab measured how often two leading legal-research products produced hallucinated output across 200+ queries. Both rates are far from the zero that "hallucination-free" marketing implies.

Hallucination rates, leading legal-research tools
Tool	Accurate (approx.)	Hallucination rate
Lexis+ AI	~65%	>17%
Westlaw AI-Assisted Research	~42%	>34%

Source: Stanford RegLab and HAI, Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (2024, peer-reviewed in the Journal of Empirical Legal Studies, 2025).

ABA Formal Opinion 512 sets the duties. On July 29, 2024 the ABA issued its first ethics guidance on generative AI.⁶ Under Model Rule 1.1 (competence), a lawyer must have a reasonable understanding of a tool's capabilities and limits, need not become an AI expert, and cannot blindly rely on output: every substantive AI element gets reviewed by a responsible attorney. Under Rule 1.6 (confidentiality), the opinion warns that self-learning tools by their nature raise the risk that one client's information is disclosed improperly, so lawyers must understand how a tool retains data and may need the client's informed consent before inputting confidential material. The opinion also covers candor to the tribunal (Rule 3.3, no unverified citations), client communication (Rule 1.4), supervision of AI as a non-lawyer tool (Rules 5.1 and 5.3), and reasonable fees (Rule 1.5, do not bill saved time as if performed manually).⁸

Pulling it together: governance lags adoption, the tools are imperfect even when purpose-built, and the duties land squarely on the lawyer. The engineering response is to design for verification, confidentiality, and supervision from the first commit, which is a build thesis instead of a disclaimer.

How to build legal AI that holds up

Legal AI that survives ethics scrutiny shares one shape: it grounds every answer in authoritative sources, attaches a verifiable citation, filters privilege and personal data at the boundary, logs every step, and routes output through a licensed attorney before anything leaves. The compliance duties above are not a layer added at the end. They are the architecture.

In practice that means five engineering commitments. Ground answers in primary law and the firm's own document system through retrieval, so assertions trace to a real source instead of model memory. Verify citations programmatically against an authority that confirms the case exists and stands for what it is cited for. Enforce confidentiality at the data boundary with privilege and personal-data filtering, tenant isolation, and no training on client data. Log inputs, retrievals, and outputs so the workflow is auditable and defensible, which matters directly for e-discovery. And gate every output behind attorney review, because verification is the lawyer's non-delegable duty.

That is the difference between a chatbot bolted onto sensitive matters and a system a firm can defend. It is the pattern our AI application development team builds around for legal clients, and it pairs with the architecture-level companion guide on AI agents for legal when a use case crosses into multi-step execution.

Frequently asked

AI use cases in legal questions

What are the main use cases for AI in legal?

The highest-adoption uses are document review (74%), legal research (73%), and document summarization (72%), followed by drafting briefs and memos (59%), contracts (51%), and correspondence (50%), per Thomson Reuters (2025). Beyond those, AI supports due diligence, e-discovery, litigation analytics, compliance monitoring, intake and triage, legal operations and billing, and knowledge and precedent retrieval, most of them document-centric tasks performed under attorney supervision.

How widely is AI actually used in the legal industry?

Organizational adoption nearly doubled in a year, from 14% of legal organizations actively using generative AI in 2024 to 26% in 2025, with 72% of users running it weekly or more, per Thomson Reuters (2025). A separate 2026 survey by 8am reported 69% of individual legal professionals now use generative AI for work. The two surveys measure different things, but both show steep growth.

Is it safe to use AI for legal work?

It can be, with guardrails, but the risks are documented. Even purpose-built legal-research tools hallucinate, 17% to 34% of the time in a Stanford study, and courts have sanctioned attorneys for filing fabricated AI-generated citations, such as the 5,000 dollar sanction in Mata v. Avianca (2023). Safe use requires retrieval grounding, citation verification, confidentiality-aware data handling, and mandatory attorney review under ABA Model Rules 1.1, 1.6, and 3.3.

Does AI replace lawyers?

No. ABA Formal Opinion 512 (2024) makes verification a non-delegable duty of the lawyer, and only a licensed attorney can take professional responsibility for a filing or advice. AI handles repetitive drafting, research, and first-pass review, while lawyers supervise, verify, and decide. The realistic frame is a tool a lawyer still signs off on, not a replacement for one.

What does the ABA say about lawyers using AI?

ABA Formal Opinion 512 (July 29, 2024) is the governing US ethics guidance. It requires competence in a tool’s capabilities and limits (Rule 1.1), protection of confidential information including from self-learning tools (Rule 1.6), candor to the tribunal so no unverified AI citation is filed (Rule 3.3), supervision of AI output (Rules 5.1 and 5.3), and reasonable fees that do not bill saved time as if performed manually (Rule 1.5).

Kanika Mathur

Head of Service Delivery, Resourcifi

Kanika Mathur is Head of Service Delivery at Resourcifi, where her engineering pods build retrieval-grounded legal applications that cite primary law, filter privileged material at the data boundary, and gate every output behind attorney review. She spends most of her scoping calls talking firms out of the chatbot they asked for and into the verifiable, logged system their ethics duties actually require.

Resourcifi on LinkedIn →