AI-powered legal research for Swiss law uses semantic search and large language models to scan statutes, Federal Court and cantonal court decisions, and legal commentary. Rather than scrolling through keyword lists, lawyers receive structured, source-based first assessments in a fraction of the time required by traditional manual research.
Why legal research in Swiss law is particularly demanding
The Swiss legal system combines federal law, 26 cantonal legal frameworks, and three official languages. A lawyer clarifying an employment-law question for a client in Canton Graubünden may need decisions in German, Romansh, and Italian. Most general-purpose AI tools – and even some specialised platforms – do not fully solve this multilingual retrieval problem. The ZHAW's JuRAG research project, funded by Innosuisse, identified exactly this issue: meaning-based multilingual retrieval across Swiss Federal Court decisions is technically feasible, but performance gaps between languages remain unresolved.
There is also a structural characteristic of the Swiss legal corpus that rarely gets mentioned: many cantonal and lower-court decisions from before 2000 exist only as scanned PDFs with OCR errors. Semantic indexing models trained on clean running text stumble over these recognition errors. A platform advertising an index of 21 million decisions has not thereby solved the OCR problem.
Finally, attorney-client privilege under Art. 13 BGFA and Art. 321 StGB extends to auxiliary persons and systems used within a law firm. Feeding mandate data into an AI tool running on US servers, or one that uses queries for model training, creates a real professional-liability risk.
How AI research works technically
At their core, all major legal research platforms combine two building blocks: Retrieval-Augmented Generation (RAG) and a large language model (LLM). The RAG system searches a pre-built index (statutes, decisions, commentaries) for semantically similar passages and passes these to the LLM as context. The LLM then formulates a structured answer with source references.
The decisive quality difference lies in the index, not the LLM: which decisions are covered, from which year, in which languages, and whether only headnotes or also the legal reasoning sections (Erwägungen) are indexed. CASUS, for instance, highlights the relevant Erwägungen directly in search results, so no separate click through to the original document is needed. For time-pressured research, that is a real practical difference.
A further technical problem that gets underestimated in daily practice: many decisions cite each other, but most platforms do not trace the citation chain. A lawyer who stops at the first decision found may miss a later leading case that shifted the legal position.
The Swiss market at a glance
Several platforms compete for the Swiss legal research segment. They differ not only by database size but by a factor that most comparisons overlook: whether research integrates directly into the downstream document workflow.
Platform | Decisions in index | Multilingual | Post-research workflow (document/drafting) | Hosting |
|---|---|---|---|---|
CASUS | 660,000+ (cantonal + federal) | DE, FR (IT/EN via Chat) | Yes – research feeds directly into document workflow | Switzerland/EU |
Legislator.ch | 21 million (broad) | Not specified | No – pure research platform | Switzerland |
DeepLegal | Federal + cantonal from 1954 | DE, FR, IT, EN | No – pure research platform | Switzerland (ISO 27001) |
Lawsearch AI | 735,000+ decisions | Not specified | Limited – no document integration | Switzerland |
Silex.legal | Not published | Primarily FR | No – pure research platform | Switzerland |
The table makes one thing clear: lawyers who only need to research have several solid options. Those who want to move research results directly into a contract draft, an internal assessment, or a memo without switching between applications have far fewer choices.
AI research in practice: a concrete example
A four-lawyer employment-law boutique in Zurich handles wrongful-termination disputes regularly. Previously, an initial research pass on a new situation – say, a dismissal on grounds of illness and whether it was abusive under the relevant provisions of the Code of Obligations – took around three and a half hours: opening the case-law database, testing search terms, reviewing decisions one by one, marking relevant sections, then writing a short summary for the file.
Using CASUS's legal research mode, that step now takes under thirty minutes. The platform searches across more than 660,000 cantonal and federal decisions, surfaces the relevant Erwägungen directly in the result, and produces a structured assessment with pro and contra argument lines and concrete recommendations. The output can immediately feed into a clause draft or client correspondence via a chat action – all within the same workflow, without copying and pasting between applications.
This is not an edge case. Law firms that conduct mandate-related research daily consistently report that the largest time saving is not in the search itself but in converting the research result into a usable working document. An integrated workflow closes exactly that gap.
What AI research does not do – and where errors arise
The most common source of error in practice is not a wrong answer from the AI but a missing verification step by the lawyer. Anyone who relies on the platform-delivered Erwägungen without checking the original document risks overlooking a restrictive obiter dictum that substantially qualifies the main holding. In firms that have used AI research for some time, a pattern emerges: junior staff who have grown up with AI summaries open the original document less often than experienced colleagues – with a correspondingly higher risk of missing nuances.
Three further limitations are structural:
First, even the largest indexes do not cover all cantonal lower-court decisions. Cantonal superior court and district court decisions from smaller cantons are patchy.
Second, case law evolves. A 2019 decision may have been superseded by a later leading case. Platforms that offer no citation tracking do not flag this.
Third, AI research does not replace legal judgment. Structured assessments, pro-and-contra lines, and action recommendations are tools for the lawyer, not substitutes for professional analysis.
No reputable platform claims its results are "always complete" or "guaranteed correct." Anyone reading such promises should be cautious.
The regulatory framework: what lawyers need to know in 2026
The Swiss Bar Association (SAV) adopted recommendations on AI use in legal practice in summer 2024. The core message: lawyers remain personally responsible for the accuracy and completeness of their work, even when AI systems are involved. Professional secrecy under Art. 13 BGFA and Art. 321 StGB extends to all auxiliary persons and systems.
The EU AI Act has been in force since August 2024; the GPAI obligations under Art. 53 EU AI Act for foundation-model providers have applied since August 2025. For Swiss law firms that serve EU clients or that might themselves qualify as operators of an AI system, the classification question is increasingly relevant. The Swiss Federal Council adopted the cornerstones of a Swiss AI regulatory framework in February 2025; a consultation draft is to be prepared by end of 2026.
In practical terms, choosing a platform that hosts data in Switzerland or the EU, does not use data for model training, and offers Zero Data Retention is not just a matter of comfort – it is professional-liability risk management.
CASUS Legal Research: how it works and how it differs
CASUS's legal research mode is embedded within the AI Chat. When a legal question is entered, the response is grounded in statutes and court decisions, with a direct preview of the relevant Erwägungen – no leaving the chat window to open source documents. The database covers more than 660,000 cantonal and federal decisions.
What sets the integrated approach apart: the research result is not a final product. It can feed directly into a clause rationale, an internal assessment, or a document change via Agent Mode. Once the research is done and a clause needs adjusting or a contract section needs redrafting, there is no application switch – the next step happens in the same chat window.
CASUS hosts all data in Switzerland and the EU, does not transfer data to the US, and offers Zero Data Retention: document content is not used for model training and is not stored after the session. The platform is available as a Microsoft Word add-in and as a web app.
The legal research function for German and Austrian law is launching in the coming weeks (as of June 2026); Swiss law coverage is already live.
For firms that want to cover their entire document workflow – from research through AI-assisted contract review to benchmarking against internal playbooks – CASUS is designed as an all-in-one platform, not a standalone research add-on.
Firms that want to test the integrated workflow directly can get started with a free trial. A full-featured account costs CHF 100 per seat per month on annual billing, or CHF 125 on monthly billing (regular price CHF 145), as of June 2026.
FAQ
What is AI-powered legal research for Swiss law?
AI-powered legal research combines semantic search with a language model to scan statutes and court decisions and deliver structured, source-based assessments. In the Swiss context, this covers federal law, cantonal law, and decisions across multiple languages.
How many decisions does CASUS cover?
CASUS searches across more than 660,000 cantonal and federal decisions. Results display the relevant Erwägungen directly within the chat interface, without requiring a click-through to the original document.
Is AI legal research compatible with attorney-client privilege?
It depends on the platform. Art. 13 BGFA and Art. 321 StGB require that auxiliary persons and systems also maintain professional secrecy. Platforms with Swiss or EU hosting, without US data transfer, and without using queries for model training meet this requirement structurally better than general-purpose AI tools.
How reliable are the results?
AI legal research provides source-based, structured first assessments. No reputable platform guarantees completeness or accuracy. Review by the responsible lawyer – including reading the original document – remains essential to avoid missing obiter dicta and more recent leading cases.
What does CASUS cost?
CHF 100 per seat per month on annual billing, CHF 125 on monthly billing (regular CHF 145). As of June 2026.
Does CASUS cover French- and Italian-language cantonal court decisions?
The current database includes cantonal decisions from all cantons. Multilingual research is possible via the chat mode. Fully resolved multilingual retrieval across all language regions remains – as academic research (JuRAG/ZHAW) confirms – a technical challenge that all platforms are still working on.
How does CASUS differ from pure research platforms such as Lawsearch or DeepLegal?
The main difference is the post-research workflow. Pure research platforms deliver answers that the lawyer then manually transfers to working documents. CASUS connects research directly to document editing, clause suggestions, and Agent Mode actions within the same workflow – no media switch required.
Does the EU AI Act apply to Swiss law firms?
The EU AI Act is not directly applicable in Switzerland. Swiss law firms serving EU clients, or using AI systems that could fall under EU regulation, should monitor developments. The GPAI obligations under Art. 53 EU AI Act for foundation-model providers have applied since August 2025, which indirectly affects the tools firms use.
