Revolution AgencyRevolution Agency

Delivered by our team through its European experience (Evolution Agency, France).

Case study: Probate genealogy, France

Answer tax questions and show where every answer comes from.

For Guenifey, a probate genealogy firm, our team built an assistant for notaries. It answered questions on tax and inheritance law from the firm's tax guide, and cited its source in every answer.

Every answer

carried its source: a link to the right page of the guide's PDF (a design rule)

1

knowledge source: the firm's 2025 tax guide

0

questions at the end of an answer: the assistant kept it short and did not prompt back

Methodology: the results on this page are qualitative. They come from operating the tools and from the testing done before each launch. We publish client-specific percentages only once they are measured and approved.

Context

Guenifey is a probate genealogy firm. It produces a tax guide, and the 2025 edition served as the basis for the assistant.

Notaries spend hours searching for tax and inheritance information in complex documents. The firm wanted an assistant that finds the information in its guide, with no room for error. The starting point:

  • Questions on tax and inheritance law, asked by notaries
  • A reference source already written: the 2025 tax guide, as a document and as a PDF
  • A zero-error requirement and full traceability

Challenge

In inheritance tax, an answer has to be verifiable. The firm demanded zero error tolerance, full traceability, and validation by a legal expert at each iteration.

1

Trace every answer

Every answer had to show where it came from in the guide, with a link to check it.

2

Stay on the guide

Answer from the firm's guide, with short answers that go straight to the information.

3

Validate with a legal expert

Every new version of the assistant had to be validated by a legal expert.

Solution: an assistant that answers from the guide and proves it

An assistant built and run by our team. Its rule: answer from the firm's guide, and show where every answer comes from.

  1. 1

    The guide as knowledge base

    The 2025 tax guide was split and indexed in a spreadsheet (Google Sheets) that served as the knowledge base. The assistant looked for the answer in that content.

  2. 2

    A source cited in every answer

    Every answer cited its source. A custom script turned those citations into links to the PDF version of the guide, so the notary could open the document and check.

  3. 3

    Short answers

    The instructions required brief answers, capped in length, never ending with a question. The notary got the information and its source, nothing more.

  4. 4

    Legal review at each iteration

    Each version of the assistant went through a legal expert, who reviewed the answers before validation.

Results

No measured percentages on this page: we only publish a client rate once it is measured and approved. The figures above describe design rules and the scope of the service, not measurements. The findings below describe the service as it ran.

Every answer pointed to its source, with a link to the guide's PDF

Answers drawn from the firm's 2025 tax guide

Review by a legal expert at each iteration, to meet the zero-error requirement

Short answers that did not send the notary a question back

The honest framing: the assistant did not replace a tax adviser's opinion. It found the information in the guide and showed where it came from, so the notary could check it.

Stack

Conversational assistant on web chatClaude language model (Anthropic)Knowledge base: the 2025 tax guideGoogle SheetsAirtableCustom JavaScript: sources turned into PDF links

Your documents are the reference. Your assistant should cite them.

Evaluate your project free with our interactive tools, or book a call: we look at your reference documents and the questions you get, and tell you what an agent can handle, and what it should leave to your legal team.

Book a call
NovaVoice guide
Ready