Judykatura 2.0 searches by wording and by meaning at the same time — which is why a question written as a plain sentence works better here than a string of keywords. Below: what the search engine looks for, how to phrase questions, what to avoid, and what else comes with your access.
An ordinary search engine compares characters: type „cookies” and it will miss a decision that speaks of „trackers”. Ours runs two tracks at once — and decides for itself which result belongs at the top. You never have to pick a mode.
Finds exactly what you typed — and such hits always go to the top of the list.
Finds documents that say the same thing, even in completely different words — and in a different language.
Each rule with an example: on the left a question that weakens the result, on the right the same question asked well.
You do not have to hit the exact phrase used by the regulator. Describe the situation the way you would explain it to a colleague.
Three topics in one sentence blur the result — the engine splits its attention between them. Ask three questions in turn; each will collect its own set of documents.
Common words („data”, „company”, „issue”) appear in tens of thousands of documents and narrow nothing down. Specialist terms do the opposite: they point to the right shelf.
A case number, reference number or ECLI is searched literally and lands at the top of the list. If we do not hold that document, you will be told so plainly — instead of getting an answer built around something else.
A passage from the reasoning placed in quotation marks is searched word for word. It is the fastest way to the document a sentence in someone else's brief came from.
Your question is matched against the corpus regardless of the language of the document. An English sentence will surface a Spanish authority's decision and a German court judgment — including those nobody has translated. The text stays in the original, so that you read what was actually written.
Keywords are enough when you are looking for documents to read yourself. When you want a written answer with citations, phrase it as you would to a person — with a verb and a question mark.
Six things in a question raise accuracy the most.
The rarer and more technical the term, the stronger the signal. „Cookie wall”, „impact assessment”, „joint controllership” work; „a GDPR case” does not.
Add whatever narrows the field: the article, the type of infringement, the sector, the role (controller, processor), the stage of proceedings.
A short description of the facts carries more meaning than isolated words — and meaning is precisely what the second track compares.
Ask in the language you think in. The language barrier applies to reading a document, not to finding it.
The box takes 600 characters. It is for a question, not for pasted pleadings — longer text blurs rather than narrows.
The first question is often reconnaissance. See what came back, then ask a second, narrower one using a term the results suggested.
Worth knowing, because it explains why some questions work better than others.
The question goes in parallel to the literal and the meaning-based search. Any case number detected in it is searched word for word.
A set of full documents is assembled — with a cap per source, so that no single authority dominates. EDPB guidelines and CJEU judgments carry more weight, and separate slots are reserved for legislation itself.
The answer is written solely from the text of those documents. Nothing is added „from memory” — where the material does not answer the question, you are told so.
Case numbers, references and amounts appearing in the answer are checked against the sources. Every citation opens the document, so you can read it in the original.
The AI answer is an addition to the search, not a condition of using it — the document list works without it.
Click one to see what a good question looks like in practice. An answer with citations requires signing in.
Behind the search runs continuous, automated monitoring of official sources across the Union. Every day, with no human involved, it checks whether something new has appeared — and brings it into the database in full text. That is why the results are current, rather than „as of the last review”.
118 official sources checked automatically, several times a day.
PDF, HTML, XML, even scans — the whole text of the document enters the database, never just a link or a summary.
Material outside the scope is filtered out, completeness of the text is verified, the date, authority and case number are put in order.
The document becomes searchable on both tracks — literally and by meaning — alongside the rest of the corpus.
New items go into the morning briefing and into News, so that you know about them before your competitors.
