A new study conducted by Anidjar & Levine finds that artificial intelligence is reshaping legal workflows faster than courts and regulators can respond. The data shows rapid adoption across research, review, and drafting, with measurable time savings. It also shows error rates and reliability gaps that directly impact client outcomes, malpractice exposure, and judicial trust. This is not a theory-of-change paper. It is an accounting of where efficiency gains exist, where accuracy still fails, and where firms must set guardrails to keep both sides of the ledger in balance.
Adoption is broad and concentrated in time‑heavy tasks
The study aggregates recent legal tech statistics to quantify where AI sits in daily practice.
- Document review: 74 percent of legal teams used AI in 2024, and 77 percent expect to in 2025.
- Legal research: 73 percent reported AI use in 2024, rising to 74 percent in 2025.
- Document summarization: 72 percent in 2024, 74 percent in 2025.
- Brief and memo drafting: 59 percent in both 2024 and 2025.
- Contract drafting: 51 percent in 2024, increasing to 58 percent in 2025.
- Correspondence drafting: 50 percent in 2024 and stable in 2025.
These figures align with the study’s core observation. AI’s footprint is largest where work is repetitive, text‑heavy, and time bound. The tasks most expensive in billable hours are now the tasks most likely to be machine‑assisted.
Efficiency gains are real and measurable
The study identifies time savings as the top reported advantage.
- Primary benefit: 54.4 percent of legal professionals list time savings as the leading gain.
- Workflow impact: Faster contract and discovery review, accelerated case law retrieval, and rapid first‑draft generation for internal memoranda and client correspondence.
- Resource reallocation: Firms report shifting more attorney hours toward case strategy, witness preparation, and negotiations when routine work is partially automated.
The upshot is straightforward. AI can compress throughput on standardized tasks, increasing attorney capacity without expanding headcount. For clients, this translates into lower costs for commoditized work and potentially faster matter timelines.
Accuracy is still the limiting factor
The study is equally clear about the constraints.
- Top concern: 74.7 percent rank accuracy as the primary risk.
- Reliability and privacy: 56.3 percent cite reliability issues, while 47.2 percent flag data privacy concerns.
- Hallucination rates: General models show 58 to 82 percent error without domain‑specific training.
- Specialized tools: Westlaw AI hallucinated in 34 percent of tests; Lexis+ AI showed error rates above 17 percent even with retrieval‑augmented generation.
- Real‑world exposure: Attorneys have been sanctioned for filing briefs with fabricated citations generated by AI.
The data tells a cautionary story. Efficiency without accuracy in law is not a gain. An hour saved on a draft with false citations is not an hour saved. It is an hour lost plus reputational risk.
Disclosure, supervision, and judicial guardrails
The study documents a rapid shift in courtroom expectations.
- Judicial disclosure: More than 40 federal judges now require disclosure when filings are assisted by AI, up from about 25 in mid‑2024.
- Bar guidance: State bar associations in California, New York, and Florida mandate attorney supervision of AI‑generated work.
- Statutory activity: At least eight states are drafting or have passed AI‑in‑law regulations focused on consumer protection, malpractice liability, and transparency.
These moves respond directly to the accuracy data. Courts and bars are signaling that AI can be used, but not unsupervised, and never without accountability for the final product.
Client demand outpaces lawyer trust
The study highlights a generational shift in expectations.
- Client expectations: 68 percent of clients under 45 expect lawyers to use AI tools.
- Perceived benefit: Only 39 percent of lawyers believe AI improves client outcomes.
- Market signal: 42 percent of clients say they would consider hiring a firm that advertises AI‑assisted representation.
This split matters. Firms face pressure to modernize, yet the data says use must be supervised. Meeting client expectations requires a clear narrative about where AI is used, where it is not, and how accuracy is guaranteed.
A practical framework for responsible efficiency
The study’s numbers point to a pragmatic operating model.
- Task selection: Use AI on standardized, low‑risk tasks with strong retrieval pipelines. Avoid unsupervised use in high‑stakes analysis.
- Verification: Require source‑anchored outputs and run automated citation checks before attorney review.
- Human‑in‑the‑loop: Set confidence thresholds. Below the threshold, route to manual processes.
- Transparency: Disclose AI assistance where required and document supervision in the file.
- Data hygiene: Protect client data through vetted systems, access controls, and on‑prem or compliant cloud configurations.
Efficiency is achievable without sacrificing accuracy when firms design workflows around the weakest link in the chain.
Bottom line
The study conducted by Anidjar & Levine concludes that legal AI delivers measurable time savings where tasks are repetitive and evidence is easily retrievable. It also concludes that accuracy remains the gating factor for trust. The firms that will gain the most are those that treat AI as a supervised accelerator, not an unsupervised analyst. Efficiency and accuracy are not mutually exclusive, but they require guardrails that align with how courts, bars, and clients read the same data.