CaseGroundDeutsch

Legal analysis you can verify

CaseGround is built for lawyers and tax advisers. You enter a set of facts and receive a legal review memo in which every step names its source: statute, case law or administrative guidance. All processing takes place exclusively on our own servers.

The law thinks in connections. So does CaseGround.

If you are examining a unification of shares, you need at least five sections of the German Real Estate Transfer Tax Act: § 1 for the taxable event, § 8 for the tax base, § 13 for the persons liable and §§ 18 and 19 for the notification duties. CaseGround’s norm graph records how these provisions are connected. For every connection, it shows the relevant passage in the statute.

ExemptionsCore provisionsConsequences and procedure
§ 1(3)Unification of shares“at least 90 per cent of the shares in the company in the hands of the acquirer”

Connected with

§ 1(3) Unification of shares“at least 90 per cent of the shares in the company in the hands of the acquirer”§ 1(3) no. 1

How a review works

You describe the facts as you would in a file note. CaseGround breaks them down into individual facts and applies the law to them, element by element. Within a few minutes the review memo is ready. It supports every step with a citation and flags every gap. We are starting with real estate transfer tax.

The case becomes a structure

Facts

Holding AG has held an interest in Zwischen GmbH since 2015, as has Mr Ott. Since 2010, Zwischen GmbH has held 100% of the shares in Grund GmbH, which owns a property in Nuremberg. On 1 August 2026, Holding AG buys a further 20% of Zwischen GmbH from Mr Ott and receives the shares the same day. Is real estate transfer tax triggered?

Structure

Holding AGacquires 20%
Mr OttShareholder
Zwischen GmbHholds Grund GmbH
Grund GmbHowns property
PropertyNuremberg
??100%ownsPurchase on 1 Aug 2026

Review

GrEStG

Purchase agreementHolding AG buys from Mr Ott.

The object is a propertyShares are bought, not a property.

Follow-up question

How large was Holding AG’s interest in Zwischen GmbH before the acquisition?

Choose an answer. It changes the structure and the memo.

Memo

Review memo, draft, interest before the acquisition: null%

The acquisition triggers real estate transfer tax under § 1(3) no. 1 GrEStG.

With the acquisition, Holding AG indirectly holds, through Zwischen GmbH, 100% of the shares in Grund GmbH, which owns the property. The threshold of at least 90% is reached.

§ 1(2b) is not met because only 20% of the shares change hands and Holding AG, as an existing shareholder, is not a new shareholder.

The tax is based on the assessed property value.

Holding AG and Grund GmbH are liable for the tax.

Edit paragraph Ask about it

Language modelFactsReview memoResult and reasoningRule engineElements, thresholds, deadlinesNorm graphStructure of the statuteOfficial sourcesStatutes, rulings, decrees

How CaseGround is built

  1. 1Capturing the factsCaseGround breaks the text down into individual facts and identifies the parties and assets involved.
  2. 2Linking to the norm graphEach fact is assigned to the provisions of the statute where it matters.
  3. 3Targeted support from the language modelThe language model’s main task is to interpret indeterminate legal concepts.
  4. 4Calculation in the rule engineThe rule engine resolves elements that call for little judgement as well as unambiguous calculations. The same facts always lead to the same result.
  5. 5Assigning citationsEvery result rests on a verified source from our collection, not on the language model.
  6. 6Writing the memoThe end product is a fully drafted memo with result, reasoning and citations, ready for your review.

Why we are building CaseGround

In 2012, lawyers in Germany were 47.5 years old on average. Ten years later, the figure was 51.7.1 Over the same period, the share of lawyers over sixty rose from 16.6 to 27.3 per cent, while the share of 31 to 40-year-olds fell from 27.6 to 13.4 per cent. The number of admitted lawyers has been falling since 2017. At the start of 2026 it stood at 138,420.2

The picture in tax advice is similar. The average age is 53.6 years, and 30.7 per cent of tax advisers are over 60.3

In 48.9 per cent of law firms, there is only a single lawyer.1 Anyone who works alone has to get up to speed in unfamiliar areas of law first or pass the matter on.

Even so, clients expect more. 73 per cent of the in-house lawyers surveyed for the Legal Tech Monitor 2026 want to concentrate their work with law firms that demonstrably cut costs with AI.4 Demands are rising and new talent is scarce, so legal reviews have to get faster. Yet the professional remains responsible for them, which means they have to be verifiable.

51.7years on average

Training and vacancies in law and tax firms

New training contracts for legal and notarial assistants56

8,930in 2002

2,885in 2025

Positions left unfilled at sole-practitioner tax firms7

35.9%in 2018

59.1%in 2024

The obvious answer would be more staff. Because they are so hard to find, we are building CaseGround.

Running on our own hardware

Client confidentiality is protected by professional law (§ 43a(2) BRAO, § 57(1) StBerG) and by criminal law (§ 203 StGB). External service providers may only be given access under the conditions of § 43e BRAO and § 62a StBerG.8 The German Federal Bar (BRAK) therefore advises putting only abstract questions to language models wherever possible.9 But that makes it almost impossible to review a specific set of facts.

That is why the language model at CaseGround runs on our own hardware. The model calls stay within our network as well. CaseGround is therefore your only service provider within the meaning of these provisions.

More than a language model

The German Federal Bar puts it this way: language models calculate probabilities and do not understand what they write. That is why they can produce statements that sound logical and are still wrong.9

Researchers from Stanford and Yale have measured this. On questions about decisions of US federal courts, general-purpose language models were wrong in 58 to 88 per cent of cases.10 Specialised research products with a legal database did better, but in a later study they still hallucinated in 17 to 33 per cent of queries.11

CaseGround’s advantage lies in how it is built. A pure language model can simply assert, whereas CaseGround has to prove. The structure comes from the norm graph and the evidence from official sources. Anything that cannot be supported is marked as unsupported in the review memo.

ChatGPT 4, general-purpose language model58 out of 100 wrong

Llama 2, general-purpose language model88 out of 100 wrong

Legal research products, lower bound17 out of 100 wrong

Legal research products, upper bound33 out of 100 wrong

Studies of US law using products from 2023 and 2024. The figures are not directly comparable with one another. We are not aware of any such measurement for German law.

No citation without a source

Invented citations have already come before German courts. In July 2025, the Local Court of Cologne criticised a brief containing fabricated quotations and pointed to § 43a(3) BRAO.12 In September 2025, the Regional Court of Frankfurt found a breach of a lawyer’s basic professional duties, because AI output has to be checked before it is used.13 In November 2025, the Berlin Court of Appeal made clear that lawyers have to check the citations in their briefs themselves.14

A draft therefore only helps if you can check it faster than you could do the research yourself. That is what CaseGround is designed for: every citation in the review memo is backed by an official source.

CaseGround compared

Legal AI platforms and publisher assistants can do a great deal. CaseGround does one thing very precisely: it checks whether a specific set of facts meets the elements of a legal provision and shows what that conclusion rests on. The table is simplified and describes product categories, not individual vendors.

General-purpose language modelsLegal AI platformsPublisher assistantsCaseGroundBuilt forText and knowledge work of every kindResearch, document review and drafting across many practice areasResearch and work within the publisher’s contentReviewing a set of facts through to the review memo, one area of law at a timeBasis of the answerLanguage model, sometimes with web searchLanguage models with agents, integrations and firm knowledgeLanguage model on curated publisher contentNorm graph and rule-based evaluation, with the language model answering individual questionsOrigin of citationsModel or web search, not verifiedConnected data, depending on setupPublisher database including commentaryVerified collection of official sourcesBreadth todayVery broadVery broad, many marketsBroad, German legal systemReal estate transfer tax, further areas of law to follow

As of November 2026

Register your interest

If you would like to use CaseGround, please sign up. We will get in touch as soon as there is news.

By registering, you agree that we may contact you about CaseGround by email or phone. You can withdraw your consent at any time. Details are in our privacy policy.

Or email us directly at info@caseground.de.

Sources

  1. Institut für Freie Berufe Nürnberg for the BRAK, STAR 2025, Statistisches Berichtssystem für Rechtsanwälte, February 2026, figs. 1.2.4 and 2.4.7.
  2. BRAK, BRAK-Magazin 2/2026, membership statistics as of 1 January 2026.
  3. Federal Chamber of Tax Advisers, Berufsstatistik 2024, as of 1 January 2025.
  4. beck-aktuell, Rechtsabteilungen nutzen KI und erwarten das auch von Kanzleien, 17 September 2026, on the Legal Tech Monitor 2026 by Legal Tech Verband Deutschland.
  5. beck-aktuell, Ausbildungszahlen sinken weiter, aber immerhin langsamer, 23 January 2026.
  6. beck-aktuell, Was tun gegen die rückläufigen Azubi-Zahlen?, 14 September 2023.
  7. STAX 2024, Digitalisierung und Fachkräftemangel, IfD Allensbach, supplement to DStR 4/2025, pp. 15 et seq.
  8. § 43a and § 43e BRAO, § 57 and § 62a StBerG, § 203 StGB.
  9. BRAK, Hinweise zum Einsatz von künstlicher Intelligenz, as of December 2024.
  10. Dahl, Magesh, Suzgun, Ho, Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models, Journal of Legal Analysis 16 (2024), 64 ff.
  11. Magesh, Surani, Dahl, Suzgun, Manning, Ho, Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools, Journal of Empirical Legal Studies 2025.
  12. AG Köln (Local Court of Cologne), order of 2 July 2025, 312 F 130/25, reported by beck-aktuell, 8 July 2025.
  13. LG Frankfurt am Main (Regional Court), order of 25 September 2025, 2-13 S 56/24, reported by beck-aktuell, 26 September 2025.
  14. KG (Berlin Court of Appeal), order of 20 November 2025, 17 WF 144/25, reported by beck-aktuell, 8 May 2026.

Photos from Pexels, edited by us. All information as of November 2026.