Free resource · Word Kit

AI Kit for Legal Teams: 30 Use Cases and Their Prompts, in 3 Levels

From the first prompt to the supervised assistant: the most useful daily uses of generative AI in a legal department, with their ready-to-copy prompts, to develop skills autonomously, step by step.

📄 Word, 48 pages🧭 30 cards in 3 levels🔒 Red zone and check🆓 Free

Download the kitSee all three levels

Generative AI is already present in legal departments: 92% of lawyers use at least one AI tool, according to Wolters Kluwer, and 74% of legal and compliance professionals use them several times a week, according to Thomson Reuters. Yet, 91% believe their organization isn't getting the most out of it. The problem is no longer access to the tool; it's knowing what to ask of it, with what data, and how to verify its output.

This page presents what the figures show, the rules to set before the first prompt, the three-level progression we recommend and an example of a complete sheet, then gives you the kit in Word format to carry it out independently, alone or in a team.

1. What the figures show: massive usage, still a slim value

The Future of Professionals report published by Thomson Reuters in June 2026, based on a survey of 1,816 legal, tax, audit, and compliance professionals in 62 countries, reveals a paradox. While usage is widespread—74% use AI several times a week, and 44% several times a day—value is lagging behind: 91% believe their organization is still falling short of what the technology could offer. The obstacles cited are not technical. Lack of training is the second most common complaint (43%), and a third of respondents use tools that their organization has not approved. Where an AI strategy is clearly defined and visible, 66% believe AI meets expectations, compared to only 22% elsewhere.

For legal departments, contract review is the focus of attention. LegalOn's 2026 survey of 452 in-house lawyers shows that 87% see this as the primary benefit of AI, for a task that takes an average of 3,1 hours per contract. Four out of five teams are already exploring AI-powered agents, but almost all require human oversight. Wolters Kluwer, for its part, reports that 62% of lawyers save between 6% and 20% of their weekly time thanks to AI.

The risk, however, is now documented in France. Between December 2025 and February 2026, the administrative courts of Orléans, Périgueux, Grenoble, and Bordeaux, respectively, noted inaccurate or nonexistent case law references produced by AI in their written submissions, reminding the courts of their duty to verify them. The database maintained by Damien Charlotin lists over a thousand such decisions worldwide. None of them implicate a faulty tool; all involve an unverified output.

Hence the bias of the kit: to progress by level, from the safest to the most delegated, with confidentiality and verification integrated into each sheet rather than relegated to the preamble.

2. Before the first prompt: the red zone, the template, the verification

The primary skill is not about prompting well, but about knowing what not to disclose. The guide "Ethics and AI," published by the National Bar Council in March 2026, reminds lawyers of this: no information covered by professional secrecy should be transmitted to a generative AI tool, and it is essential to verify where the data is hosted, who owns the tool, and what its confidentiality settings are. In-house counsel must adopt the same precautions regarding trade secrets, contractual confidentiality, and personal data.

Never seize	Examples	What to do instead
Personal data not needed	Names, contact details, contract numbers, health or banking data, individual situation of an employee	Pseudonymizing: using roles instead of names ("the Supplier", "the employee"), rounded amounts
Information covered by a secret	Legal consultations, litigation strategy, documents in a criminal case	Only in a tool validated by the IT department and the firm, in a partitioned space
Confidential third-party information	Contracts received under confidentiality agreements, partner data, due diligence documents	Review the confidentiality clause: it often prohibits any unauthorized third party from accessing it.
Trade secrets	Acquisition projects, unpublished results, cost prices, roadmaps	Validated tool only, based on the need-to-know principle
Security elements	Passwords, access keys, architecture diagrams	Never, in any tool

All the prompts in the kit then follow the same five-line template, so that everyone can learn to write their own:

• Role : the expected register and level of requirement (“experienced corporate lawyer”, “meticulous proofreader”).
• Context : your organization, your position as a customer or supplier, the issue at stake, the end reader.
• Task : specific and numbered questions, a closed perimeter ("only from the text below").
• Format : table, maximum length, named columns, required citations.
• Verification : what AI should do with its uncertainties ("writes unspecified", "quotes the exact passage").

And three rules apply to every release, without exception. Every cited reference exists and says what it's supposed to say: it must be opened in the official source before being used. Every quotation from a document must be found in the document, word for word. And the "I would sign" test: if you wouldn't sign the release exactly as it is, under your own name, it's not over.

3. Three levels to progress step by step

Level 1. Take control: save time today

Ten use cases where you provide the text and the output can be reviewed in five minutes: summarizing a long text into key points, explaining a clause in plain language, rewriting a sensitive email, preparing a meeting from a file, translating with a glossary, transforming a memo into a message for management, creating an internal FAQ from a policy, getting an explanation of an unfamiliar system, proofreading a draft letter, and extracting actions and deadlines from an email thread. The goal: to be self-sufficient in everyday use cases and to master the template.

Level 2. Analyze and write methodically

Twelve uses where you bring your documents and positions: extracting key contract clauses into a table, analyzing a contract against the playbook, comparing two versions, comparing several texts or jurisdictions, drafting a clause in two variants, preparing for negotiations, building a project compliance checklist, monitoring marketing communications, reviewing a supplier's terms and conditions, drafting a memo based on verified sources, structuring monitoring, and mapping the risks of a case. This is the core value for a legal department, and the level at which verification becomes systematic.

Level 3. Delegate to assistants and supervised agents

Eight fact sheets where the prompt becomes a permanent instruction: the team's playbook assistant, the weekly regulatory monitoring, the deadline monitoring, the first-level assistant for operational staff, the template generator, the end-to-end contract review, the bulk analysis of a contract portfolio, and the preparation of a litigation file. Each fact sheet provides the instructions to paste into the tool, the launch prompt, and the monitoring rules: things the assistant never does alone.

Family	Level 1	Level 2	Level 3
Contractual analysis	Actions and timelines of an email thread	Key clauses in table, contract against playbook, supplier's conditions	Assistant playbook, monitoring deadlines, end-to-end review
Comparison	·	Two versions of a contract, several texts or jurisdictions	Mass analysis of a portfolio
Copywriting	Translation with glossary, rereading	Clause in two variants, negotiation, memo	Frame generator
Risks and compliance	Internal FAQ based on a policy	Compliance checklist, marketing communications, risk mapping	Litigation case based on a corpus
Legal research	Summary of a long text, unknown concept	Structuring a monitoring system	Regulatory monitoring
Communication	clear language, delicate email, meeting, Note for management	·	First-level assistant for operational staff

The details of each entry are as follows: the situation, what the AI ​​does well and what it doesn't, then the first check before using the result; at level 3, what the assistant changes and its first supervision rule. The complete prompt, the assistant's instructions, or the mission order are included in the kit.

The 10 Level 1 worksheets: getting started

Worksheet 1: Summarizing a long text into key points

Legal research. Typical time saved: 30 to 60 minutes per text. A thirty-page decision, a decree, an official report, or an internal memo lands on your desk. You need to know within ten minutes whether it concerns you and what changes.

What AI does well: Identify the structure and key points of a text you provide; produce a hierarchical summary in the required format. What she doesn't do: It doesn't know your organization if you don't describe it; it doesn't guarantee completeness: a point may be missing. To check : Review the five points in relation to the text: does each quote exist word for word?

Worksheet 2: Explaining a clause or contract in plain language

Communication. Typical time saved: 20 to 30 minutes per request. A project manager asks you "what does" the liability clause in a supplier contract mean? You want a clear, jargon-free explanation that he can reuse.

What AI does well: Translates a clause into everyday language; provides concrete examples drawn from the reader's profession. What she doesn't do: Does not know the intention of the parties or the history of the negotiation; may gloss over a legally important nuance (an exclusion, a ceiling, a condition). To check : Reread the explanation from a legal perspective: has it removed an exclusion, a ceiling, a condition?

Fact Sheet 3: Rewriting a Tricky Email

Communication. Typical time saved: 10 to 15 minutes per email. You must refuse a request from an operational staff member, follow up with a supplier about a missing part, or correct a service provider, while remaining firm and courteous.

What AI does well: Suggest several tones for the same background; structure the message around a request and a deadline; identify aggressive or ambiguous wording in your draft. What she doesn't do: Does not know your relationship with the recipient; tends to lengthen and over-polish; does not engage your responsibility: you are the one sending. To check : Verify that dates, amounts and references are unchanged.

Fact Sheet 4: Preparing a meeting based on a file

Communication. Typical time saved: 30 minutes per meeting. Meeting tomorrow with a professional or partner regarding a matter you're not very familiar with. You have the latest memo, two emails, and a report.

What AI does well: Extracts the facts, past decisions and open points; proposes an agenda and classified questions; identifies inconsistencies between documents. What she doesn't do: Does not know the dynamics of the players or what is sensitive; can turn a hypothesis into a decision if the documents are ambiguous. To check : Cross-reference each "decision made" with the cited document.

Fact Sheet 5: Translating a Legal Document with Glossary

Writing. Typical time saved: 1 to 2 hours per document. A confidentiality agreement, internal policy, or correspondence must be sent in English or received in Dutch. The translation must be accurate and the terminology consistent with your other documents.

What AI does well: Translated quickly using legal terminology; adheres to a prescribed glossary. What she doesn't do: Does not master the concepts specific to each law: a resolutory clause is not exactly a termination clause; may smooth over an ambiguity intended by the drafter. To check : Reread the three highlighted passages and the definitions.

Fact Sheet 6: Transforming a Legal Note into a Message for Management

Communication. Typical time saved: 30 minutes per note. Your six-page memo is ready. The management committee wants three points and a decision to be made, on half a page.

What AI does well: Prioritizes and cuts; translates a legal risk into an operational consequence; adopts the "expected decision" format. What she doesn't do: Does not know what management already knows; may oversimplify an important reservation; does not carry your recommendation for you. To check : Check that the recommendation is indeed yours and that a reformulation has not displaced it.

Fact Sheet 7: Creating an Internal FAQ from a Policy

Risks and compliance. Typical time saved: 1 hour per policy. The new gift and invitation policy, or the signing procedure, has been published. You will receive the same ten questions by email.

What AI does well: Anticipates readers' questions; answers based on the policy text; adapts the tone to the intranet. What she doesn't do: Should not invent the organization's position on an unforeseen case; does not know the actual cases already reported; does not replace the validation of the policy owner. To check : Check each item reference.

Fact sheet 8: Getting an explanation of an unfamiliar diet or concept

Legal research. Typical time saved: 20 minutes to map a topic. A case study introduces you to a subject you have little experience with: competition law, export controls, or the labor law of another country. You want an overview of the topic before consulting the sources.

What AI does well: Provides a pedagogical overview; asks questions to understand your needs. What she doesn't do: Not a source: texts, dates and references cited must be verified one by one; may mix national laws. To check : Open the official texts indicated (Legifrance, EUR-Lex, websites of the authorities) and verify each concept.

Fact Sheet 9: Reviewing and correcting a draft letter or contract

Writing. Typical time saved: 20 to 40 minutes per document. Your draft amendment or formal notice is written. Before sending it, you would like a second look at the consistency, cross-references, definitions and typos.

What AI does well: Identifies inconsistent definite terms, broken cross-references, duplicates, impossible dates, typos; flags sentences with double meanings; classifies problems by severity. What she doesn't do: Does not judge the appropriateness of your positions; may slip in substantive changes under the guise of style. To check : Process the table line by line: you decide.

Fact Sheet 10: Extracting actions and deadlines from an email thread

Contractual analysis. Typical time saved: 20 minutes per line. A thread of forty emails about a supplier dispute or negotiation. You need to know who promised what, by when, and what is still owed.

What AI does well: Reconstruct the chronology; extract commitments, deadlines and those responsible; identify contradictions between messages. What she doesn't do: Does not know what was said orally; may attribute a commitment to the wrong person if the thread quotes incorrectly; does not legally qualify the commitment. To check : Verify each commitment in the source email.

The 12 Level 2 worksheets: analyzing and writing methodically

Worksheet 11: Extracting the key clauses of a contract into a table

Contractual analysis. Typical time saved: 1 hour per contract. A sixty-page supplier contract to be processed by tomorrow. You want a reliable identity card before reading the details.

What AI does well: Locate and quote the requested clauses; present them in a fixed-column table; indicate missing clauses and inconsistencies. What she doesn't do: Do not judge the balance of the contract without your positions; may miss a clause moved to an appendix or under an unusual heading. To check : Open each cited article and confirm the citation.

Sheet 12: Analyzing a contract against the playbook

Contractual analysis. Typical time saved: 1 to 2 hours per contract. The supplier's draft contract arrives. Your team has standard clauses: liability capped at 100% of the annual price, termination for convenience with three months' notice, data hosted within the European Union. This is your playbook.

What AI does well: Compare clause by clause to your positions; classify the differences according to your grid. What she doesn't do: Does not know your actual negotiating power; does not know the history with this supplier. To check : Review all clauses 3 and 4 in the contract text. The full details are reproduced below.

Worksheet 13: Comparing two versions of a contract

Comparison. Typical time saved: 45 minutes per round of negotiation. The supplier sends back "their" version. Change tracking is absent or incomplete. You want to know exactly what has changed and what that implies.

What AI does well: Identify additions, deletions and relocations; explain the consequence of each change. What she doesn't do: Does not see a change in an unsupplied appendix; may miss a changed word in a long clause (an "and" becoming "or"). To check : Also launch Word's automatic comparison (Review, Compare): AI explains, Word misses nothing.

Fact Sheet 14: Comparing several texts or jurisdictions on a point

Comparison. Typical time saved: 2 hours for a first grid. The company wants to launch the same service in four countries. You must compare the obligations (withdrawal period, pre-contractual information, retention period) based on the texts you have gathered.

What AI does well: Structure a comparison based on the provided texts; identify the differences and set up the correct columns. What she doesn't do: She does not reliably know the current positive law of each country: without the texts, she invents or mixes things up. To check : Each box must link to a text that you have provided; delete any box that does not have one.

Fact Sheet 15: Drafting a clause in two versions

Drafting. Typical time saved: 30 to 45 minutes per clause. You're missing a reversibility, audit, force majeure, or data protection clause tailored to your specific context. You want to start with a solid document, not a blank page.

What AI does well: Offers a structured writing service based on your instructions; available in two levels of requirements. What she doesn't do: Does not know the details of your applicable law without you specifying it; may produce a clause of Anglo-Saxon inspiration unsuitable for French law. To check : Check compatibility with the rest of the contract: definitions, cross-references, document hierarchy.

Fact Sheet 16: Preparing for a Negotiation

Writing. Typical time saved: 1 hour per session prepared. Negotiation meeting in two days regarding three unresolved clauses. You want to anticipate the other party's arguments and prepare your fallback positions.

What AI does well: Generates arguments and counter-arguments; proposes fallback positions and exchanges of concessions; drafts the message that opens the session. What she doesn't do: Does not know the balance of power or the real commercial stakes; may propose concessions outside your mandate; does not know what you do not want to reveal. To check : Confront the retreats with your actual mandate.

Fact Sheet 17: Building a Compliance Checklist for a New Project

Risks and compliance. Typical time saved: 1 hour per project. The marketing department is launching a service with a new tool that processes customer data, possibly using AI. We're asking for your feedback by next week.

What AI does well: List the questions to ask, corpus by corpus; structure a checklist with requirement, proof and responsible party. What she doesn't do: Doesn't know what the project really does (nor sometimes the job itself); doesn't apply sector regulations if you don't name them. To check : Check the requirements in the texts themselves and in your policies.

Fact Sheet 18: Controlling Marketing Communications

Risks and compliance. Typical time saved: 30 minutes per campaign. An email campaign, landing page, or advertisement needs to go live tonight. The marketing team wants a quick "legal OK."

What AI does well: Identifies missing information, claims to be substantiated, misleading wording; highlights risky practices: consent, opt-out, crossed-out prices, comparisons; proposes a corrected text. What she doesn't do: Does not know your regulated sector without you saying so; does not know your contractual commitments with publishers or partners. To check : Check the information in the texts and your charters.

Fact Sheet 19: Review a supplier's Terms of Service or Privacy Policy

Contractual analysis. Typical time saved: 45 minutes per tool. A business wants to subscribe to a SaaS tool "in three clicks." The publisher's terms and conditions and privacy policy are the only contractual documentation.

What AI does well: Extracts the points of friction with your requirements; identifies data location, subcontracting, unilateral modification, limitation of liability. What she doesn't do: Does not know your requirements if you do not provide them; does not see the additional documents (treatment agreement, service commitments) not provided. To check : Check the citations and the online version of the document.

Worksheet 20: Writing a structured memo using verified sources

Writing. Typical time saved: 1 to 2 hours per memo. A specific legal question arises: can this contract be terminated for this reason? Is this clause enforceable? You have gathered and verified the relevant texts and rulings. You want a structured first draft.

What AI does well: Organize the reasoning: facts, question, rule, analysis, conclusion. What she doesn't do: Does not provide sources: on those you do not paste, she invents references (the number 1 risk noted by the courts). To check : Check the final list of references: any reference you have not provided is removed, and the passage that relies on it is rewritten.

Worksheet 21: Structuring a monitoring system based on pasted texts

Legal research. Typical time saved: 1 hour per week. Every week, you receive newsletters, alerts, and articles. You want actionable information for the team: what's changing, for whom, when, and what to do.

What AI does well: Sort and summarize from the texts provided; qualify the impact according to your criteria. What she doesn't do: Don't look for sources at this level: that's the role of the level 3 lookout (sheet 24); may overestimate the importance of a text. To check : Open the sources of the "high impact" lines.

Fact Sheet 22: Mapping the Risks of a Case

Risks and compliance. Typical time saved: 1 hour per case. An emerging dispute, a proposed acquisition, an announced inspection. Management wants a prioritized view of the risks and measures.

What AI does well: Structure a probability and severity matrix; propose measures with responsible party and time frame; identify risks overlooked in an initial list. What she doesn't do: It does not assess the actual probability: it makes a suggestion, you note it; it does not know your risk appetite; it does not replace the firm's opinion on a dispute. To check : Re-note each risk yourself.

The 8 Level 3 fact sheets: delegating to assistants and supervised agents

Fact Sheet 23: His Contractual Playbook Assistant

Contract analysis, assistant configured. Typical time saved: 1 hour per contract, for the entire team. Each incoming contract is compared to the playbook (sheet 12). Rather than pasting the playbook back in each time, you fix it in a permanent assistant that the whole team uses in the same way.

What this changes: The playbook becomes a permanent instruction; every lawyer gets the same initial analysis; positions evolve in a single place, under the responsibility of an owner. Supervision rule: A playbook owner, a monthly review of positions, a version history.

Fact Sheet 24: Weekly Regulatory Monitoring

Legal research, monitoring with web search. Typical time saved: 1 hour per week. Manually monitoring Fact Sheet 21 takes an hour per week, and that's assuming the right alerts have arrived. You want an assistant to gather and pre-qualify what's been published within your scope each week.

What this changes: The assistant will search a closed list of official sources, apply your impact assessment grid, and you review it. Monitoring becomes regular and traceable. Supervision rule: A lawyer opens each link classified as "high impact" before publication.

Fact Sheet 25: Monitoring Contractual Deadlines and Disputes

Contractual analysis, configured assistant linked to a table. Typical time saved: avoiding an oversight is worth the deployment. Automatic renewals, notice periods, warranty expiry dates, hearing dates: deadlines are scattered throughout contracts and emails. An oversight is costly.

What this changes: The assistant extracts the deadlines from each document, populates a single table, and lists the sixty, thirty, and seven-day deadlines. The lawyer validates each entry. Supervision rule: A lawyer validates each new line, in particular the starting point and the method of calculation.

Fact Sheet 26: First-Level Assistant for Operational Staff

Communication, assistant configured, published to non-lawyers. Typical time saved: a quarter of the time spent answering recurring questions. The same questions come up every day: Can I sign this confidentiality agreement? Who approves a contract for less than 10,000 euros? Can I send this file to this service provider?

What this changes: An assistant answers questions covered by your policies and procedures, cites the internal source, and refers anything outside the scope to the legal counsel. The referred questions are then added to the FAQ (sheet 7). Supervision rule: Testing with twenty real questions before opening, then every quarter.

Fact Sheet 27: Frame Generator from the Model Database

Drafting assistant configured. Typical time saved: 30 minutes per frame. Confidentiality agreements, extension amendments, formal notices, mandates: requests for templates arrive daily. Your templates exist, but everyone adapts them to their own needs.

What this changes: The assistant uses only your templates, asks the necessary questions, generates the framework with the fields to be completed highlighted, and a list of the adaptations made. The clauses remain unchanged. Supervision rule: Versioned and dated models, one owner per model.

Fact Sheet 28: Orchestrating an End-to-End Contract Review

Contract analysis, mission agent in your files. Typical time saved: half a day per complex contract. An eighty-page contract with six annexes. The complete review (identity sheet, comparison to the playbook, redline, summary note) takes one day.

What this changes: You delegate the entire chain to an agent who works in your files, step by step, and you review each deliverable before launching the next one. Supervision rule: Human validation between each step: this is the condition set by almost all teams that deploy agents.

Fact Sheet 29: Mass Analysis of a Contract Portfolio

Comparison, mission agent in your files. Typical time saved: from several days to a few hours. Change of control, increase in indexation, new regulatory requirement: you need to know, in fifty or five hundred contracts, which ones contain such a clause and what it says.

What this changes: The agent reviews the portfolio, extracts the relevant clause for each contract along with its citation, produces a tracking table and a list of contracts to be processed as a priority. The lawyer reviews a sample and processes it. Supervision rule: Control sample: reread by hand 10% of the lines, including all the "missing" ones on critical contracts.

Fact Sheet 30: Preparing a litigation file from a corpus

Risk and compliance, mission agent in a compartmentalized space. Typical time saved: several days of data analysis. A legal dispute has arisen: three hundred documents (emails, letters, minutes, invoices). The lawyer needs a chronology, an inventory, and the key facts within two weeks.

What this changes: The agent builds the inventory and chronology from the evidence, links each fact to its corresponding piece of evidence, and identifies contradictions. You and the lawyer develop the strategy, which remains outside the tool. Supervision rule: Tool and space validated by the IT department and by the lawyer: confidentiality of the strategy and professional secrecy.

4. Example sheet: analyzing a contract against the playbook

The supplier's draft contract arrives. Your team has standard positions: liability cap, termination notice period, data location. This is your playbook. The AI ​​compares the contract clause by clause, categorizes the discrepancies according to your framework, and proposes reformulations based on your fallback positions. It doesn't know your actual negotiating leverage or the history with this supplier, and the legal counsel decides the order of negotiations. Because your playbook is a sensitive document, the form is only used in the tool approved by the IT department.

Before using the exit strategy, three checks are necessary: ​​review all the gaps classified as 3 and 4 in the contract, verify that the reformulations do not exceed what you can request, and decide on the order of negotiation yourself. At level 3, this same playbook becomes a permanent assistant that the entire team consults in the same way.

5. Deploy the kit within the legal team

The kit can be used independently, with twenty minutes per day and two weeks per level. However, it reveals its full value when a legal director or head of legal operations uses it as a team program, with a rhythm, a ritual, a place for sharing, and a measurement system.

Weekdays	The goal	What each person does
0	Frame	Reading the framework, self-assessment, confirmation of the validated tool with the IT department, selection of a kit owner and an AI ambassador within the team
1 and 2	Level 1	One worksheet per working day based on a real document, progress log, weekly 45-minute workshop where each person presents a case study
3 and 4	Level 2	The family cards that weigh most heavily on each person's week; in the workshop, a real contract drawn up using the playbook together
5 and 6	Level 3	Two assistants built in pairs, tested on twenty real-world cases, then made available to the team
7	Anchor	Review of time saved and established practices, team prompt library, cruising speed

Three indicators are sufficient: regular use, declared time saved (including proofreading), and the verification rate, which must remain at 100%. The dated progress log and the list of workshops also constitute a simple record of the measure taken under Article 4 of the European regulation on AI, which requires organizations to promote the development of AI skills among their staff (an obligation of means since the omnibus regulation (EU) 2026/1744).

6. What the kit contains

The kit is a 48-page Word document, with the fields to be replaced highlighted in yellow. It opens with a section to read before the first prompt: why this kit and how to use it, the red data zone and pseudonymization in three steps, the Role, Context, Task, Format, and Verification template, the three verification rules illustrated by recent French decisions, and what Article 4 of the European AI regulation expects from you. This is followed by 30 fact sheets, each with the same structure: the situation, what AI does well and what it doesn't, what not to paste into the tool, the prompt, the verifications, a variant, and a rating for each tool. Part 4 describes the six-week team deployment, the weekly workshop, the shared library, the indicators, and the five errors that lead to failure. Seven annexes complete the kit: self-assessment in ten questions, sheet template for your library, verification checklist and red zone to print, correspondence by tool, adaptation by sector, progress journal and sources.

7. Frequently Asked Questions

Is the kit suitable for lawyers who have never used AI?

Yes. Level 1 starts from scratch, with scenarios where you provide the text and proofread the output in five minutes. The ten-question self-assessment in Appendix A shows each participant their starting point, and lawyers who are already comfortable with the level can move directly to level 2 or 3.

Does it work with Copilot, ChatGPT, Claude, or our internal assistant?

Yes. The prompts are written for Microsoft 365 Copilot, ChatGPT, Claude, Gemini, Mistral, or an internal assistant. Appendix D indicates where to find, with each vendor, the functions useful at level 3: agents, projects, web search, and mission agents.

Can a contract be pasted into the tool?

Only in the tool approved by your IT department, never in a personal account, and preferably after anonymizing the parties and amounts. A contract received under a confidentiality agreement requires verification of the confidentiality clause. Each fact sheet specifies what should not be pasted in this particular case.

How to avoid references invented by AI?

By never asking him to serve as a source. As soon as a task touches on positive law, it is the jurist who provides the texts and decisions; the prompts require exact quotations and a mention "not specified" rather than a deduction; and each reference is opened in the official source before being taken up.

How long does it take to complete the three levels?

Six weeks, at twenty minutes per day and two weeks per level. In teams, a weekly forty-five-minute workshop is sufficient to share cases, correct prompts and build the common library.

Is the kit also suitable for law firms?

Most of the fact sheets are applicable in law firms, subject to the profession's ethical rules, particularly professional secrecy and client disclosure, as outlined in the National Bar Council's guide. However, the toolkit is primarily designed for the legal teams of companies and public organizations.

8. Going further

The kit is designed for self-directed skills development. A team that wants to progress faster, or reach level 3 with support, can follow the AI training for lawyers and legal teams From NextStart.AI: a half-day or 90-minute session, based on your specific cases and using your own tool, with 70% hands-on practice. AI usage charter template sets the framework that the kit puts into practice, and the AI Ambassadors Kit Helps set up relays throughout the organization. Prompts, agent, and mission sheets Cowork those mentioned in the kit can be found on the library app.nextstart.ai.

Read also : AI Charter in Business: The 2026 Model (Downloadable), AI Ambassador Network: The 2026 toolkit for building your collective, Manager's Kit for Facing AI, AI systems governance: the 2026 toolkit et AI Maturity Grid: Assess your organization's readiness in 90 minutes.

The entire series of free kits. To drive adoption: AI charter template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit et AI Governance KitFor the AI ​​project sponsor: AI Act kit, GDPR kit, Security and data kit, AI and Social Dialogue Kit, frugal AI kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI finance kit, AI kit for salespeople, AI toolkit for project managers, AI innovator kit et communicator's AI kit.

9. sources

The figures and references cited on this page and in the kit are based on the following sources, consulted in September 2026.

• Thomson Reuters Institute, Future of Professionals Report 2026, June 22, 2026.
• Wolters Kluwer, Future Ready Lawyer Survey Report 2026, March 10, 2026.
• LegalOn and In-House Connect, The 2026 State of AI for In-House Legal, 2026.
• National Bar Council, guide "Ethics and AI", March 2026, available among the CNB practical guides ; commentary by the Justice Village.
• Regulation (EU) 2024/1689 on artificial intelligenceArticle 4 (control of AI), and its reformulation by Omnibus Regulation (EU) 2026/1744 (adopted on 8 July and published on 24 July 2026), presented by personaldata.fr.
• Morgan Lewis, The Risks of Hallucinations and Misuse of Generative Artificial Intelligence Before French CourtsMarch 2026.
• Damien Charlotin, AI Hallucination Cases Database.
• GC AI, AI Prompts for Lawyers: 15 Templates In-House Teams Use Daily, 2026.
• Village of Justice, AI and law: the 5 favorite prompts of the Augmented Lawyers TeamJanuary 2026, updated September 2026.
• Microsoft Adoption, Copilot Scenario Library, legal function.