A practical guide to using AI in legal work

AI has, in a short space of time, become a natural part of the working day in many companies. Tools like ChatGPT, Claude, and Gemini get used to analyze information, develop ideas, write drafts, and prepare decisions.

Legal questions increasingly end up in these tools too. A contract clause needs to be understood. A policy needs drafting. A customer request needs assessing before tomorrow morning's meeting.

The value is obvious. AI can give you a faster overview, better structure, and a more thought-through starting point. For companies without an internal legal function, the technology can shrink the gap between a question coming up and the company understanding what needs to be clarified.

Legal work is different from a lot of other use cases, though. The answer can affect agreements, employees, ownership interests, personal data, and financial risk. A convincingly worded answer isn't enough on its own. It also has to be correct, tailored to the situation, and based on a sound factual foundation.

So the central question isn't whether AI can be used in legal work. It can. The question is which tasks the technology is suited to, where the limits are, and when the judgment call needs to go to a legal advisor.

This guide explains how companies can use AI effectively and responsibly, which tools fit which needs, how to get better answers, and what the rules actually say about accountability, privacy, and confidentiality.

Can you use Claude and ChatGPT as a legal sparring partner?

Yes, and that framing is the whole point. A sparring partner helps you train, but it doesn't fight the match for you.

General-purpose AI (Claude, ChatGPT, Gemini) is excellent at the work that surrounds a legal question. Explaining a concept, summarizing a long agreement, listing the questions you should be asking, drafting a first version of a policy, or pressure-testing your own reasoning before a negotiation. For a founder without in-house legal expertise, that's a real advantage.

It shouldn't, however, be the sole basis for a decision that could carry legal or commercial consequences.

The difference comes down to accountability and judgment. An AI tool produces text based on patterns in the data it was trained on. A legal advisor assesses the facts, asks follow-up questions, sees the issue in the context of your business, and stands behind the advice.

What's the difference between an AI tool and a legal advisor?

Four things matter in particular: accuracy, context, accountability, and confidentiality.

Accuracy

Large language models are built to produce answers that sound plausible and are well-written. That doesn't mean the answers are always legally correct.

The model can cite a legal provision that doesn't apply, draw a conclusion that doesn't actually follow from the sources, or reference a court decision that doesn't exist. The error can be hard to spot, because the answer is often delivered with a lot of confidence.

A 2024 Stanford RegLab study found that general-purpose AI tools produced fabricated or incorrect information in a significant share of legal queries, including invented statutory references and court decisions that were never handed down.

The Mata v. Avianca case is a well-known example. A US lawyer filed a court submission citing several fictional rulings generated by ChatGPT. When the lawyer asked the tool whether the rulings were real, it confirmed that they were. The information was never checked before the document was submitted to the court.

The example points to a fundamental limitation: AI doesn't always reliably distinguish between what's correct and what merely looks correct.

Even lawyers need to check AI-generated content against reliable sources. For people without a legal background, the risk is greater, because it can be harder to spot which part of the answer is misleading.

Context and legal judgment

A legal question can rarely be assessed in isolation.

A contract clause has to be read together with the rest of the agreement. The parties' negotiating positions can matter. So can the company's risk appetite, business model, prior communications, and the practical goal of the agreement.

AI only sees the information you give it. If important context is missing, the answer can end up wrong even if the legal explanation looks reasonable on its own.

An advisor can ask follow-up questions, surface unstated assumptions, and recognize that the real problem is something other than what was originally asked.

When shouldn't AI be used alone?

AI can still be useful in preparation, but it shouldn't be the sole basis for a decision once the matter carries real consequences. A legal advisor should normally get involved when:

• The company is about to sign, send, or commit to something.

• The matter involves a termination, warning, restructuring, or other employment-related decision.

• Significant financial value or ownership interests are on the line.

• The question touches regulatory requirements or possible action by an authority.

• There's a dispute, or a clear risk of one.

• The decision could meaningfully affect a customer, employee, or other individual.

• A useful answer would require sharing personal data or confidential information.

• The company isn't sure whether the AI's answer is grounded in correct law or practice.

Norwegian employment law is a good example. A termination or restructuring requires both a sound factual basis and a correct process. A general AI answer can't confirm whether the company has met every requirement in the specific case.

The same goes for investment documents, shareholders agreements, and important commercial contracts. AI can help explain the terms and identify questions, but shouldn't be the sole decision-maker on which obligations the company takes on.

The rule of thumb is simple: use AI to prepare the decision. Use an accountable advisor to quality-check what the company actually does.

Which legal AI tools fit your company?

The right choice mainly depends on who's using the tool, and what the company actually needs.

General-purpose AI tools

ChatGPT, Claude, and Gemini are a relevant starting point for companies without an internal legal function.

They're flexible and useful for a wide range of tasks, including summarizing, explaining, idea development, structuring, and first drafts.

For founders and leaders, this is normally the most accessible category. The tools can make it easier to understand a problem and prepare the groundwork before a legal advisor gets involved.

They should still be used as general working tools, not as independent legal decision-makers.

Specialized legal AI tools

Specialized platforms are built for law firms and in-house legal teams. They can be used for contract analysis, legal research, document review, and due diligence. Some tools generate answers grounded in defined, controlled legal sources, and are built for professional users who can assess the quality of the output.

These tools can deliver real efficiency gains for legal teams. They also require legal expertise, solid implementation, and typically a much bigger budget.

For a company that only occasionally needs to check a contract clause, this category is usually more than the need calls for.

For companies without in-house legal expertise, the most practical combination is often general-purpose AI for preparation and drafting, alongside a dedicated legal advisor who quality-checks the decisions.

The technology doesn't replace the advisor. It can make the advisor faster and more effective.

How do you get better answers?

The quality of the answer largely depends on the quality of the instructions.

Vague questions tend to produce general, not-very-usable answers. "Is this clause fine?" gives the model limited information about what it's actually supposed to assess.

A better prompt describes the company, the situation, your role in the agreement, and what you want help with.

• Give it context and a role. Open with who you are and the situation: "We're a Norwegian SaaS company with 15 employees. We're entering a customer agreement where we're the supplier. Explain the clause below and assess what commercial and legal risks it could carry for us." Specific context produces specific answers.

• Be precise about what you want. "Explain this clause and flag anything unusual or risky for the customer" beats "is this ok?" Ask it to show its reasoning. Tell it to think step by step and explain why, not just conclude. You'll catch faulty logic faster.

• Ask it to flag uncertainty. A line like "Tell me explicitly which parts of this you're unsure about, what information is missing, and which questions should be confirmed with a legal advisor" turns the model into a more honest collaborator.

• Give examples and specify format. Want a comparison table, a list of questions for your lawyer, a plain-language summary? Say so.

• Treat the answer as a draft. The first response should be treated as a working document. Push back. Correct facts. Ask for alternative assessments. Explore how the answer changes if the assumptions change. A good AI process is iterative. The goal isn't a finished legal conclusion from one prompt, but a better foundation for the assessment that follows.

A good prompt doesn't just get you a better answer. It gets you a better-prepared conversation with the advisor who'll make the final call.

Are you allowed to use AI-generated results for legal matters?

There's no general prohibition in Norwegian law against using AI as a tool in a company's legal work.

The company is nonetheless responsible for anything it does on the basis of the output. Responsibility for a contract, termination, investment, or other decision can't be transferred to the model.

AI-generated content also isn't legal advice in the professional sense. The tool has no duty to safeguard the company's interests, verify the facts, or flag anything missing from the assessment.

At the same time, the rules around AI use are evolving.

The EU AI Act (AI-forordningen) is being phased in across the EU and is expected to be implemented in Norway through the EEA agreement. Among other things, it requires companies using AI systems to ensure staff have an adequate level of AI literacy.

That means the company should have a handle on:

• Which AI tools are being used

• What the tools are being used for

• What information can be entered into them

• How the results should be checked

• Which matters require human judgment

• Who's responsible for overseeing the use

AI systems used in connection with recruitment, hiring, and other decisions about employees may be subject to stricter requirements. These are areas where technology can directly affect individuals' rights and opportunities.

The AI Act comes on top of existing rules. GDPR continues to apply in full whenever AI tools process personal data.

Privacy, security, and confidentiality

The most avoidable mistakes usually happen because employees share more information with AI tools than they should.

A practical rule of thumb: personal data and confidential business information shouldn't go into open, consumer versions of AI tools.

That includes things like:

• Names of customers, employees, and job applicants

• Salary information and other HR data

• Health information

• Customer lists

• Internal conflicts

• Non-public contract terms

• Trade secrets

• Investment plans and financial information

• Documents tied to an ongoing dispute

If the company needs to use AI on material that includes this kind of information, it needs a solution and a contractual setup suited to professional use.

Practical guardrails:

Use the right tier. Consumer versions of AI tools may use your inputs to train future models unless you opt out. Business tiers (Teams, Enterprise) and APIs typically don't, and they're the right choice for company use.

Check the data processing agreement. If a tool processes personal data on your behalf, you need one, and you should confirm where the data is stored (ideally the EU/EEA).

Anonymize before you ask. Strip out names and identifying details, ask your question about the generic situation, and apply the answer yourself.

Train your team. Decide which tools are approved, what's allowed in them, and who to escalate to. This isn't just good practice, under the AI Act, basic AI literacy is becoming a requirement.

The ethical version of all this is simple. Treat anything you put into an AI tool as if it could be seen by someone else, and don't let a machine make a call that affects a real person's job or rights without a human reviewing it.

The bottom line

AI can give companies faster access to information and make legal work more efficient. It can cut down the time it takes to understand an issue, improve the quality of first drafts, and leave the company better prepared before an advisor gets involved.

That's a real advantage, especially for companies without their own legal department.

The value depends on the technology being used within clear boundaries, though.

AI can help the company understand, structure, and prepare. It can't take responsibility for a decision, guarantee the legal sources are correct, or weigh every commercial and human consequence of the advice.

The most effective model, then, isn't AI or a legal advisor.

It's AI and a legal advisor, each used for what they're best at.

The technology handles the information work, the structure, and the first draft. The legal advisor assesses the risk, sees the full picture, and stands behind what the company actually signs, sends, or decides.

FAQ

Can I use Claude for legal advice?

You can use it to understand legal concepts, summarize documents, and prepare for a conversation with an advisor. You shouldn't treat its output as advice you act on directly, it carries no professional responsibility and can produce confident, fabricated answers.

Am I legally allowed to use AI for my company's legal matters?

Yes, no law in Norway forbids it. But you remain fully accountable for anything you act on, GDPR applies whenever personal data is involved, and the EU AI Act introduces new obligations (including AI literacy) phasing in through the EEA from late summer 2026.

What's the single biggest mistake to avoid?

Pasting personal or confidential data into a consumer AI tool. Anonymize first, use business-tier tools with a data processing agreement, and keep a human in the loop on any decision that affects someone's rights or your company's risk.