A law firm looking to automate its work today faces a paradox. AI tools are everywhere, colleagues are using them, and yet usage remains scattered: everyone works in their own corner, with no shared method or framework. In 2026, the question is no longer whether AI has a place in a law firm. It's which tasks to automate first, in what order, and with what safeguards.
This article answers those three questions with recent data, a list of concrete tasks, and one simple rule: what repeats gets automated, what carries the firm's liability stays human.
Key Takeaways
- 74% of French lawyers and in-house counsel regularly use AI, up from 53% in 2024.
- The leading uses are document summarizing (74%), contract review (69%), and case-law research (55%).
- A Goldman Sachs study estimates that 44% of legal-profession tasks are automatable.
- 63% of lawyers don't tell their clients they use AI: confidentiality remains the weak point of rollout.
- AI handles research, analysis, and simple drafting. Legal qualification and strategy stay human.
Adoption has tipped over in two years
France's 2025 figures leave no room for doubt. The annual study by Lamy Liaisons, conducted with OpinionWay among 181 lawyers and in-house counsel, found that 74% of lawyers and legal professionals regularly use AI, up from 53% in 2024. The same study reports that 90% of respondents believe AI will have a major impact on their profession, with 45% calling that impact "very major."
Wolters Kluwer's 2026 reference report on small French law firms confirms the trend at a different scale: 79.8% of surveyed French firms actively use AI tools, the strongest engagement among all countries studied. Legal research platforms lead the uses (81.6%), ahead of generative AI tools like ChatGPT (71.3%).
Two years were enough to move from individual experimentation to everyday use. What's left to do is no longer to convince, but to organize.
What firms are already automating
The Lamy Liaisons 2025 study ranks use cases by frequency:
- Writing summaries of legal documents (74% of users);
- Automated review and adjustment of contracts (69%);
- Identifying relevant case law (55%).
These three uses match the categories set out in the French Senate's information report on generative AI and the legal professions (December 2024), which distinguishes three broad categories of tasks that can be entrusted to AI: research, analysis of a body of documents, and drafting simple content. The details of these tasks, and what they change in practice, are worth a closer look.
Legal research
This is the most widespread use, and the simplest to set up. AI turns case-law research into a conversation: you describe the question, the tool queries the databases, and it suggests rulings and texts. The Senate report speaks of "personalized legal research, through conversational agents rather than simple search engines."
The limit is known to every practitioner: a model that invents a ruling produces an error that's hard to spot. That's why 98% of lawyers and legal professionals surveyed by Lamy Liaisons want AI that cites its sources. Good practice treats the tool as an efficient research assistant: it finds, it flags, it doesn't conclude.
Contract analysis
Contract review is the second most cited use (69%). In practice, AI reviews a batch of contracts and flags discrepancies: missing clauses, inconsistent dates, amounts that don't match between two documents, special conditions that deviate from the general terms. It prepares a reading grid that the lawyer then validates line by line.
This use case illustrates a good division of labor well. AI does the scanning work, the kind that's tiring and erodes attention. The human keeps interpretation and decision-making. The Senate report also mentions detecting anomalies in case files, in particular procedural irregularities that could lead to nullity, an area where AI provides immediate value to lawyers handling first-instance cases.
Summaries and memos
Producing a summary memo from a bulky case file is the most cited use (74%). AI reads through all the documents, extracts the facts, positions, and dates, and delivers an outline that the professional then takes over and completes. The gain is twofold: reading time is cut, and human review works from an already-clean structure rather than a stack of documents.
Notaries, according to an estimate cited in the Senate report, could save up to an hour and a half of processing time per file on this kind of preparation work.
Administrative tasks
Alongside the core of the job, a firm accumulates repetitive tasks that eat up time without adding value: preparing letters, answering appointment requests, call notes, reminders for missing documents. An AI voice agent can handle the phone reception, sort requests, and forward urgent ones. A simple automation can send standard letters and follow up with clients.
These building blocks are often more profitable than legal-specific tools: they run all day, don't require legal expertise to set up, and free up the assistant for tasks that demand judgment.
What stays human
The line isn't a matter of technology, but of liability. AI produces a plausible text, never a certain one. It gets things wrong with confidence, which makes checking mandatory.
Three activities can't be delegated:
- Legal qualification: classifying the facts, choosing the applicable rule - that's reasoning, not extraction;
- Strategy: weighing the options, deciding whether to litigate or settle, setting the amount sought;
- Signature and commitment: the firm's liability doesn't transfer to the software.
The Senate report puts the limit another way: AI offers help for research, for analyzing a body of documents "and not a situation," and for drafting simple content. Whatever requires understanding a situation, a person, or a stake stays human. Human review isn't a precaution: it's the professional act itself.
Confidentiality, the constraint that changes everything
This is the point the adoption figures obscure. 63% of lawyers surveyed by Lamy Liaisons prefer not to tell their clients they use AI. The reflex is understandable: professional secrecy covers a case's documents, and most consumer-grade tools send submitted content to third-party servers.
The practical guide from the Conseil national des barreaux (France's national bar council), published on France Num, is explicit: "by default, all query data is reused by the owner of the generative AI solution to train its model." For a regulated profession, entrusting a filing or a legal document to a tool whose terms allow data reuse means taking a risk that the GDPR and professional ethics rules don't cover.
The answer that preserves both adoption and confidentiality is local, no-code AI: the model runs on the firm's own server, documents never leave the premises, and nothing is retained by a third party. This architecture, detailed in our article on no-code AI in law firms, makes it possible to automate without breaching professional secrecy. It's the only setup that holds up for handling covered documents and filings.
Where to start
A firm doesn't need a complete overhaul to get going. Here's the order of priority that produces first results within a few weeks:
- Pick a repetitive task with a stable form: preparing letters, summarizing a recurring type of case file, checking a batch of contracts. A good candidate comes up often and can be validated with a quick review.
- Put mandatory human validation in place: every AI output goes through a review screen before it's used. Without this step, no one trusts the chain, and the tool sits unused in a corner.
- Frame confidentiality before scaling up: identify what can go to a cloud service and what must stay local. Documents covered by professional secrecy stay on-site.
- Train a point person within the firm, who tests use cases and documents what works. Adoption is won through use, not installation.
An assessment maps out the firm's tasks, identifies which ones are suited to automation, and helps avoid projects that would add nothing. It's also the moment to check that the tools chosen comply with the GDPR and the AI Act, whose obligations increase with the risk level of the use.
Training matters as much as the tool. A team that knows how to question, verify, and correct gets better results from an ordinary model than an untrained team with the best software on the market. Our AI training for teams is designed for firms that want to move from individual use to organized practice.
FAQ
Will AI replace lawyers?
No, and the adoption figures show it: the firms that use AI most heavily are also the ones hiring the most staff. AI removes tasks, not professions. It shifts work toward analysis, strategy, and client relationships - where the firm's value is created.
Can you use ChatGPT to handle client files?
Not without controls. Content submitted to consumer-grade tools is, by default, reusable by the provider. For a document covered by professional secrecy, you need a solution that sends nothing outside: a model hosted on the firm's own server, or a professional offering with a written commitment not to reuse data.
What's the first use to automate in a small firm?
Document summarizing or contract review, the two most cited uses among practitioners. They come up often, follow a stable form, and a human review is enough to validate the result. Start with one specific type of case, not the whole firm.
How long does it take to set up an automation?
A few weeks for a targeted use: identifying the task, choosing the tool, setting up validation, training users. What takes time is rolling it out across the whole firm, because that touches work habits and how tasks are shared.
Conclusion
French firms have adopted AI faster than any of their European neighbors: 74% of lawyers and legal professionals use it regularly, and usage is already concentrated on summaries, contracts, and case law. The gap between firms that get a real gain from it and the rest doesn't come down to the tool or the budget, but to three decisions: which task to automate, which review to keep, which data to hand to a third party.
The rule fits in one sentence: what repeats gets automated, what carries liability stays human, and what's covered by professional secrecy stays inside the firm. If you want to review which of your firm's tasks are suited to automation, contact NexeAI: we work with no-code AI agents, installed locally, built for regulated professions.
