Legal research tools for Switzerland should be selected through a controlled evaluation, not a universal ranking. The right choice depends on the authorities a team needs, how results can be verified, where research fits into the working process and which governance requirements apply. A fixed test set reveals more than a long feature comparison.
Begin with the work the tool must support
“Legal research” covers several different jobs. A litigator tracing a line of decisions has different needs from an in-house lawyer checking a recurring contractual issue. A team working mainly with federal authorities requires a different source mix from one handling cantonal matters or multilingual questions.
Define two or three recurring research jobs before comparing products. For each, state the expected work product: a source list, an internal note, a first assessment or material ready to support drafting. This prevents the evaluation from rewarding features that look impressive but do not reduce friction in the team’s actual work.
The method for AI legal research in Swiss law explains the underlying research process. This article focuses on choosing a tool for that process.
Criterion 1: relevant source coverage
Database size alone is a weak proxy for usefulness. The important question is whether the sources needed for a defined task are available and searchable at the necessary level of detail.
Ask vendors to describe coverage in categories that can be tested:
legislation and official materials;
Federal Supreme Court and other federal decisions;
relevant cantonal courts and instance levels;
languages and historical depth;
full text, headnotes or metadata only;
update process and visible source date.
Then test the description against completed matters with known authorities. A single correct leading case proves only that the individual source can be found. It does not establish completeness for a canton, time period or source class.
Criterion 2: verifiability of every material proposition
A research answer should provide a stable route to the supporting authority. The evaluation should check whether a user can identify the source, open the original, locate the relevant passage and understand why it supports the answer.
Useful features may include direct links, clear source identifiers, passage previews and separation between quoted material and generated explanation. The interface matters because verification that takes too many steps is less likely to happen consistently. Convenience, however, does not replace opening the original where the proposition is material.
Test failure cases as well. What happens when a source is unavailable, two authorities conflict or the system is uncertain? A trustworthy workflow exposes those limits instead of turning them into confident prose.
Criterion 3: quality of retrieval, not fluency of the answer
Fluent text can hide a weak search result. Score whether the tool finds the expected authority before assessing the writing. Record decisive sources found, important sources missed and irrelevant results introduced.
Use questions with factual variations that change the legal analysis. This shows whether the system responds to the relevant distinction or merely repeats a general explanation. Include at least one question with no clear answer in the available material. A system should be able to state that evidence is insufficient.
Search quality and synthesis quality should receive separate scores. That distinction makes remediation possible: a team can adjust the query when retrieval is weak, or change the review template when synthesis is the problem.
Criterion 4: fit with the downstream workflow
Research rarely ends with a list of sources. The result may need to become an internal assessment, a clause rationale, a client message or a section of a longer document. Count the steps between verified source and finished work product.
Consider whether the workflow supports structured export, collaboration, retention of source references and reuse in the document environment. The value is not that a product contains the most functions. It is that the necessary handoffs remain clear and do not break the audit trail.
CASUS connects legal research with broader document work. Other teams may prefer a dedicated database combined with their existing drafting environment. The evaluation should reward the arrangement that fits the team’s process, not a category label.
Criterion 5: governance and access
Before real matter data is used, the organisation must understand what data enters the service, who can access it, how long it is retained and which administrative controls are available. These questions should be answered through current contractual and technical documentation, not assumptions based on the provider’s location or marketing language.
Include access management, logging, deletion, incident handling, subprocessors and model-training terms in the review. The exact requirements depend on the organisation, the data and the intended workflow. Record the responsible decision-maker and the evidence used for approval.
A safe evaluation can begin with synthetic or already public material. Moving to confidential documents should be a deliberate gate after the relevant review is complete.
Criterion 6: measurable implementation effort
Licence cost is only one part of the decision. Capture setup time, training, query adjustment, verification and transfer into the final document. A tool that generates an answer quickly may still be expensive to use if every result needs extensive repair or cannot be moved into the working environment cleanly.
Measure both the new workflow and the existing baseline. Use the same task boundaries. If the manual route includes source checking, the tool-assisted route must include it too. The guide to AI time-savings measurement provides a more detailed scorecard.
A practical scorecard
Use a weighted scorecard and publish the weights internally before testing. An example structure is:
Area | Evidence from the pilot |
|---|---|
Source fit | expected source classes and decisive authorities found |
Verification | original accessible and supporting passage easy to locate |
Research quality | relevant distinctions, counterauthority and uncertainty handled |
Workflow fit | steps from source to approved work product |
Governance | documented answers and controls for the intended data |
Effort | total time including review, repair and transfer |
Weights should follow the use case. A litigation team may put more weight on source depth and citation paths. A high-volume contract team may give more weight to workflow integration, provided the required legal sources remain verifiable.
How to run the pilot
Select a small test group and a fixed set of completed matters. Do not let each provider choose different demonstration questions. Give every product the same task specification and expected output. Record the first attempt as well as any additional prompting needed.
Review the results independently where possible. The person who configured the test should not be the only evaluator. Discuss disagreements: they often reveal that the team has not yet agreed on the quality threshold.
At the end, choose among three outcomes for each use case: approved, approved with conditions or not approved. A product can be suitable for one workflow and unsuitable for another. That is more useful than naming one “best” tool for every Swiss lawyer.
What this framework deliberately avoids
This page does not publish a vendor league table. Provider features, source collections, prices and contractual terms can change. A fair comparison requires the same current evidence and the same criteria for every provider.
It also avoids treating adoption, database size or answer speed as proof of quality. Those measures answer different questions. The decision should rest on the team’s verified results, review burden and operating requirements.
Teams that want to include CASUS in a controlled evaluation can test it with their own source set. Keep the expected authorities and scoring criteria fixed before the first run.
Frequently asked questions
What is the best legal research tool for Switzerland?
There is no universal winner. The best fit depends on required sources, verifiability, workflow, governance and the results of a controlled test with the team’s own matters.
Is the largest database automatically the best?
No. A large collection can still miss the relevant source class or rank an important authority poorly. Test coverage and retrieval separately.
How many products should a team pilot?
Two or three serious candidates are usually enough for a focused comparison. Use the same questions, source expectations and scoring method for each.
Should confidential documents be used in the first test?
Not before the relevant governance review. A pilot can start with synthetic, public or appropriately prepared material and add confidential data only after approval.
When should the comparison be updated?
Review it when the source scope, material product behaviour, contractual terms or the team’s use cases change. A date alone does not keep a comparison current.
