Yes. But to use its answer in a case file, you need to be able to check every statement against a real source: a statutory provision, a judgment or a document in the file.
A tool’s price does not guarantee its reliability. The difference lies mainly in its method: some tools search for sources first, then draft. Others produce an answer directly from what they learned during training.
How does a chatbot produce an answer?
An AI that does not consult sources builds its answer from the information it acquired during training.
As a result, it may cite a reference that looks correct but does not exist. Its knowledge may also leave out the most recent laws or decisions.
An assistant that consults sources works differently. It first searches a legal database for documents, then builds its answer from the results.
This method does not eliminate every error, but it allows the legal professional to open the sources and verify the answer.
Three types of chatbot to tell apart
Not all tools described as “legal chatbots” work in the same way.
Before choosing a tool, you should therefore look at the sources it uses. The other criteria are set out in our guide on how to choose a legal AI assistant.
Why references must always be checked
A false reference can be hard to spot, as it can follow the exact format of a genuine judgment.
This is why a simple reread is not enough. Before using a citation, you need to open the source and check:
- that it really exists;
- that it actually says what the AI claims;
- that it is still applicable;
- that it fits the right legal context.
This check must take place before anything is sent to the client or filed with an authority.
The method is explained in more detail in our article on AI and Swiss case law.
What Swiss case law shows
Checking AI-generated answers has already been an issue before the Federal Supreme Court.
In judgment 4A_231/2026 of 3 July 2026, a party had used AI to prepare its appeal. Among other things, the text contained incorrect references and statements that were not properly linked to the documents in the case file.
The problem, then, was not the use of AI itself, but the use of a text that had not been sufficiently checked.
The rule to remember is simple: an AI-generated answer must be treated as a draft to be reviewed, not as a legal source.
An internal legal chatbot meets a different need
An internal chatbot works mainly from the organisation’s documents: policies, contracts, templates or procedures.
It can, for example, find a standard clause or an internal rule.
But an internal rule is not necessarily a rule of Swiss law. The two must therefore be clearly distinguished.
Access rights must also be respected. A user must not be able to use the chatbot to view a document they would not normally have access to.
Conversation history should likewise be handled with care, as it may contain legal or confidential information.
These questions are covered in detail in our article on legal AI in companies.
Protecting data before sending it to an AI
Before sharing information relating to a matter, find out what the tool does with the data.
Professional secrecy is protected in particular by Article 321 of the Swiss Criminal Code and by Article 13 LLCA.
You therefore need to know:
- where the data is stored;
- in which country it is processed;
- who can access it;
- how long it is kept;
- whether it is used to train models.
Anonymising a case file is not just a matter of removing the client’s name. A date, an amount, a place or a combination of details can sometimes make it possible to identify a person.
To find out more, see our article AI and data protection.
How to ask a legal chatbot a good question
A question that is too general often produces an answer that is too general.
To get a more useful answer, state clearly:
- the applicable law;
- the canton concerned;
- the reference date;
- the key facts;
- any points that are still uncertain.
If the tool works with your documents, also ask for the exact page, article or clause used in the answer.
Another useful step is to ask which elements could contradict the analysis, in order to bring out opposing arguments or points that remain uncertain.
This method is presented in more detail in our article on AI-assisted legal analysis.
How to test a legal chatbot
The best way to assess a tool is to test it on case files you already know.
For example, you can:
- choose a few questions from closed case files;
- ask the chatbot the same questions;
- open every source cited;
- check that the answer takes account of the right canton and the right law;
- measure the time actually saved, including verification.
If the time saved on drafting is then lost correcting errors, the tool adds little value.
To identify the solutions to compare, see our article on legal AI alternatives.
Setting simple rules within the firm
A good tool is not enough. The firm also needs to set clear rules for using it.
A simple policy can answer four questions:
- Which tools are authorised?
- What data can be shared with them?
- Who must check the result?
- What record of this check must be kept?
Team training should then emphasise source checking and the limits of AI.
The uses specific to each profession are covered in our articles on AI for lawyers and AI for legal professionals.
Where does Silex fit in?
Silex is one of the sourced legal assistants. It searches Swiss law and can also work with the documents in the file.
Research can be limited to a canton, a field of law or a type of source.
With Silo, users can search contracts, notes or correspondence while linking these documents to Swiss law.
For large case files, the file analysis can bring out the main facts, missing information and the points to examine.
The professional nevertheless remains responsible for the final analysis.
According to Silex, files are hosted in Switzerland with Infomaniak, are not used to train the models and are not visible to other organisations.
Details are available on the security and product pages.
Silex pricing
Silex offers several monthly plans:
- Solo: CHF 149 per month for one user;
- Team: CHF 490 per month for up to five users;
- Enterprise: CHF 895 per month for up to ten users.
The plans are detailed on the solutions and pricing pages.
In short
A legal chatbot can speed up research or help prepare a text, but its answer must always be checked before it is used.
Before choosing a tool, look above all at:
- which sources it uses;
- whether these sources are accessible and verifiable;
- where the data is processed;
- who can access it;
- whether documents are used to train models.
And before using a citation, always open the source.
To test Silex on a real question from your firm, you can book a demo.
FAQ
Can a legal chatbot replace a lawyer?
No. It can help search for, summarise or structure information, but it does not take responsibility for the advice or the strategy of the case.
Can you use a general-purpose assistant if you anonymise the case file?
With caution. Removing the client’s name is not always enough. Other information can allow a person or a case to be recognised.
What is the difference between a legal chatbot and an internal chatbot?
A sourced legal assistant works from the law and, depending on the tool, from the documents in the case file. An internal chatbot works mainly from the organisation’s own documents and rules.
