Quick answer
Artificial intelligence (AI) should not be written off when it comes to legal matters. AI tools handle simple everyday questions well, and they are good at checking legal documents that already exist. The more complex the matter and the higher the cost of a mistake, the less you should rely on AI legal advice. AI has no personal experience and bears no responsibility, so the decision has to stay with a person, and a complex matter needs a lawyer.
AI-Generated Legal Documents: What Clients Bring Me
My inbox looks different these days. More and more often, private clients and companies alike send me a document with a note attached: “ChatGPT wrote this. Could you check that everything is in order?” From private clients it tends to be a complaint or a tenancy agreement. From business owners it is a supplier contract or a reply to the tax authority.
The trouble is that a sound document and a dangerous one look equally convincing: fluent language, a confident tone, references to specific sections of the law. Without legal training, you cannot tell them apart.
People ask me more and more whether they can use ChatGPT for legal advice. What follows is my personal view as a practising lawyer.
Using AI for Legal Questions: Complaints and Claims
Let me start with the good news. There is a whole category of tasks where AI is genuinely useful and where paying a lawyer often makes little sense.
Think of a complaint to a shop about a faulty product, a compensation claim against an airline for a delayed flight, a letter asking a landlord to return a deposit, or a request to a property manager. For a business, it might be a polite payment reminder, a reply to a simple customer complaint, or a routine letter to a business partner. The same goes for simply asking what an official letter means and what it wants from you.
In matters like these the rules are simple and well documented, the stakes are low, and a mistake can be corrected. An AI tool will produce a clear, level-headed draft of a complaint or claim in a couple of minutes, and often that is all you need.
AI Contract Review: What It Finds
AI’s second strength matters just as much to individuals as to companies: it is good at working with a document that already exists. Ask a chatbot to check a contract for contradictions and it will often spot what tired eyes miss: one clause sets a 14-day payment term, while another has a penalty that starts on day ten. Ask it to play the other side and pick your position apart, and you get a quick rehearsal before a negotiation or a dispute. It can also tighten the structure and tone of a legal document or letter, translate it, and help you prepare questions for your lawyer.
Why AI Cannot Replace a Lawyer: It Has No Personal Experience
AI has read more about law than any lawyer will read in a lifetime. But it has never done any of it. It has never been in court, never filed documents with a register, never negotiated with a difficult counterparty, and never explained to a bank where a company’s money comes from.
To my mind, this is the single most important point in this article. The law on paper and the law in practice are further apart than most people outside the profession realise. The law may permit something that no bank will actually accept. A technically correct document can sit with an authority for a month because the official is used to seeing it in a different format. A rule can be on the books and yet go unenforced for years.
When a client asks me a question, my answer draws on dozens of similar cases I have handled myself, and I remember how each one ended. An experienced lawyer also develops legal judgment and instinct: the sense that something is off about a deal before you can say exactly what.
Experience also tells you what to ask. Half my work in a first consultation is drawing out what the client did not think to mention: who the real owner is, where they pay tax, whether the business needs a licence. A chatbot answers the question it was asked, even when a different question should have been asked.
AI has no way of building experience. It gives advice and, as a rule, never finds out what happened next: whether the document was accepted, whether the dispute was won, whether the bank froze the account. That is why the knowledge of an “AI lawyer” remains book knowledge, however vast. Experience is precisely what a client pays a lawyer for, even when the invoice just says “contract drafting”.
Courts, Transactions, Taxes: Where You Need a Lawyer
The further you move from an everyday complaint, the more that lack of experience shows. I am talking about court claims and litigation, share deals, disputes between business partners, inheritance involving assets in several countries, group tax structures and licensed activities. These matters have many interconnected parts, and a mistake in one can bring down the whole structure. AI can describe each part competently in isolation, but it cannot put them together into a solution that works for a particular person or company.
A mistake like this often comes to light a year later, when it is too late to fix. It is worth keeping the cost in mind: legal advice costs tens or hundreds of euros, whereas undoing the consequences costs thousands, if it can be done at all. We cover the typical situations in a separate article on when you need legal advice.
AI Hallucinations in Law: Fake Case Citations and Invented Rules
Generative AI tools such as ChatGPT build their answers mainly on open and popular sources: legislation, articles, online discussions. Much of legal practice and case law sits elsewhere. Court decisions are held in separate databases with their own search tools, often only in the local language. Guidance from authorities is buried in documents few people read. Many working rules exist only in the heads of the people who apply them.
AI does not always reach these layers, and it does not tell you when it has not. A lawyer who does not know something says, “I need to check.” AI rarely does, and sometimes it cites a provision or a court decision that does not exist. These fabrications are known as AI hallucinations. This is not a theoretical risk: a public database of such incidents already lists more than a thousand court cases worldwide, and in many of them the fake case citations were filed by professional lawyers.
Legal advice has to be jurisdiction-specific, and this is where general-purpose AI is weakest. In my experience, the problem is worse for small countries such as Estonia: there is little material to draw on, so AI often fills the gap with the logic of German or American law.
Legal Advice From a Chatbot: Why It Agrees With You
If you ask, “Can I do it this way?”, a chatbot will often find a way to say yes. How you phrase the question has a big effect on the answer. A good lawyer sometimes has to tell you what you do not want to hear: you cannot do that, and I would not sign this. When lawyers get it wrong, they answer for it with their reputation and their money. When AI gets it wrong, there is in practice no one to hold accountable.
AI in Your Company: Confidentiality and Personal Data
Businesses face an additional risk. An employee who pastes a contract with a business partner, or customer data, into a chatbot may be breaching a confidentiality agreement or data protection rules. It is the company that will be held responsible. If people in your firm use ChatGPT, Gemini or similar services, you need a few simple internal rules: which services are approved and what must never be uploaded to them.
How to Fact-Check AI Legal Advice
I realise that many people will put their legal question to AI anyway. If you do, at least follow a few rules.
- State the country, the date and your role up front. “Estonia, I am a tenant, a private individual” or “I am a company director” changes the answer considerably.
- Insist on sources and check against them in several passes. Ask for the specific section of the law, its exact wording and a link. Then ask the AI to verify the facts, and separately the conclusions, against those sources. Repeat two or three times. Simply asking “Are you sure?” is almost useless: AI is as likely to confirm a mistake as to abandon a correct answer. Open the key links and read them yourself.
- Check in a fresh chat and ask for the weak points. Within a single conversation, AI is swayed by what has already been said and by your tone: it may defend a mistake, or give way to you where it was right. Open a new chat or a different service and ask: “In what ways could this answer be wrong?”
- Strip out personal data and trade secrets before uploading a document.
Important
These steps reduce the risk considerably, but they do not eliminate it: the tool checking the answer is the same tool that got it wrong.
AI or a Lawyer: How We Can Help
One caveat: everything above describes things as they stand today. The tools are improving fast, and some of these weaknesses will fade with time. The lack of personal experience and responsibility will not.
So my rule is simple: AI helps you prepare and check, while a person decides and takes responsibility. If money, deadlines, your property or the status of your company depend on a document, show it to a lawyer before you sign or submit it.
At Eesti Firma, every document is prepared and checked by a lawyer who is responsible for the result. If you have a document or a situation you are unsure about, including a contract or a claim drafted by AI, get in touch with our lawyers and we will go through it together.
Private clients with family and civil law matters can also turn to the law office in our group, Kalashnikov Õigusbüroo.
Frequently Asked Questions
In practice, no one. The developers of these services limit their liability and state plainly in their terms of use that the answers are not a substitute for professional advice. The consequences fall on whoever signed or submitted the document.
There are three signs: the mistake cannot be undone, the sum involved is significant for you, or the matter involves a court, a public authority or more than one country. Any one of them is enough.
It can draft a simple standard contract, and for a small everyday transaction that may be enough. But AI does not know your circumstances or local practice, so a contract involving a significant sum or a long term is worth showing to a lawyer before you sign. I would not trust it with a court claim: deadlines, jurisdiction and evidence all matter there, and the mistake only comes to light in court.
It depends on the service and its settings. What you type into a public AI tool is not covered by any professional duty of confidentiality. Before uploading a document, remove names, personal identification codes, bank details and anything that is a trade secret.
It is already changing the routine part of the work. Risk assessment, knowledge of how things work in practice, and responsibility for the result remain with people.
