Notaries
AI for notaries: preparing files within professional secrecy
Summaries, drafts and answers to clients, on a setup where the files never leave the office.
A notary may use AI, within limits
AI for notaries works for preparation and summaries, on a setup where files do not leave the office and with a notary who reads every text that goes out. What does not work is a deed that a model writes and nobody reads.
A notary's office runs on text: documents from clients, searches, draft deeds, emails asking about the steps and the costs of a sale. Much of that text follows a fixed pattern, and that is where AI is strong. At the same time everything falls under professional secrecy, and that decides where a model may run.
This page is written for notaries and office managers in Belgium who work with international clients, and for readers abroad who want to know how a Belgian office can use AI. It covers five tasks where AI saves time and one where it does not belong, where the files then sit, what the EU AI Act and the GDPR ask, and what a document assistant costs against the hours it gives back.
Tasks by risk, and who checks
Low risk
An email to a client about the next step
Who checks: Staff before sending
Client questions about the steps of a sale
Who checks: Set answers the firm approved
Medium
A summary of a long document or file
Who checks: The file handler, with a source per line
Preparing and sorting searches
Who checks: The file handler
High risk
A draft deed from a template and file data
Who checks: The notary, line by line
The deed itself, unchecked
Who checks: Do not
The firm's documents
Templates, files, emails
Index on a private environment
In the EU or at the firm
Assistant
Answers with the source
The ranking is an example of how a firm can order its tasks; what the profession allows is for the notary and the professional body to decide.
Five tasks where AI saves time
These are the tasks that take a lot of time in most offices and follow a fixed pattern. For each, what AI does and who checks it.
| Task | What AI does | Check |
|---|---|---|
| Summarising documents | A long document or a full file in half a page, with the source next to every sentence | The file manager reads the source |
| Preparing a draft deed | Putting the data from the file into the office's own template, and showing what is missing | The notary, line by line |
| Questions from clients | Answers about the steps of a sale or a gift, drawn from texts the office approved | Fixed answers; anything else goes to a staff member |
| Ordering searches | Sorting and summarising what comes in per file, with a list of what is still missing | The file manager |
| Emails | A first draft of a reply, in the office's tone | The staff member, before sending |
What AI does not do here: decide, advise or finish a deed. It prepares, the notary decides. A model that searches your own documents and shows its source works through RAG; how to build such an assistant is on our page about a custom ChatGPT for business.
Where the files sit
The biggest question about AI in a notary's office is not what the model can do, but where the documents go. Three routes, from open to closed:
- A public chatbot on a free account. Not for files: you do not know what happens to the text. Why, is on is ChatGPT safe for business.
- A business plan with a data processing agreement, such as ChatGPT Business at 21 euros per seat per month or Microsoft 365 Copilot. The supplier does not train on your data and signs as a processor. Fine for emails and general texts, less so for files with sensitive documents.
- An assistant of your own on a setup in the EU or in the office itself. The documents are indexed on your setup, the model gets only the documents a question needs, and every answer shows its source. This is the route for client files.
Under the GDPR every supplier that processes personal data for you is a processor, with an agreement under article 28; what that agreement holds is explained under data processing agreement. If you put a chatbot on your site that answers clients, the AI Act asks that it says it is an AI. What to put in an office policy is in our guide to the EU AI Act for business. What your profession allows, you settle with your professional body.
Clients who do not read Dutch
Many offices in Flanders now see buyers, sellers and heirs who live and work in English: expats buying a house, a foreign parent in an estate, an international company buying a building. In Flanders the deed itself is drawn up in Dutch. The work around it, though, can follow the client's language: an email about the next step, a summary of the draft, a list of the documents still needed.
That is a task where an assistant helps without touching the deed. It summarises the draft in English with a reference to the article each sentence comes from, a staff member checks it, and the client arrives at the signing knowing what the deed says. The same holds for French, which a large share of clients in and around Brussels read first.
| What the client gets | Who makes it | Who checks it |
|---|---|---|
| An English email on the steps of a purchase | the assistant, from texts the office approved | the staff member |
| A summary of the draft deed, article by article | the assistant, with the source per sentence | the notary or the file manager |
| A checklist of the documents still needed | the assistant, from the file | the file manager |
| The deed itself | the notary | the notary |
If your clients also find you through your website, a separate English page per service, written for that reader, works better than a translate button; how that works on one domain is on multilingual website development.
A worked example for an office of ten
Take an office with two notaries and ten staff that opens 60 new files a month. The figures are assumptions; put your own in their place.
| Task | Time today per file | With an assistant | Saved per month |
|---|---|---|---|
| Summarising and ordering documents | 45 minutes | 15 minutes | 30 hours |
| Preparing a draft from the template | 60 minutes | 30 minutes | 30 hours |
| Emails and questions about the steps | 40 minutes | 25 minutes | 15 hours |
| Total | 75 hours |
At an assumed average cost of 70 euros an hour, that is 5,250 euros a month, or 63,000 euros a year. The model cost is small: summarising a file of 200 pages with Claude Sonnet 5.5, at 2 dollars per million input tokens and 10 dollars output, costs about 0.29 dollars. Check your own volume with the AI cost calculator.
Building an assistant around your files, with the index, rights per staff member and the link to your practice software, then sits in the same range as what the first year returns. An office often starts with one task, the summaries or the client questions, and pays less accordingly. VBS works with a fixed price per project, agreed after a first conversation.
How we build an assistant for an office
- Pick one taskThe task with the most hours and the lowest risk first, usually summaries or client questions.
- Documents and rightsWhich documents the assistant may read, and which staff member may see which file.
- A trial on closed filesTwenty closed files, with what the assistant gives next to what the office did.
- In use, with a logEvery question and every answer kept with its source, so the office can trace what happened.
Close to lawyers and property managers
A notary's office shares a lot with a law firm: professional secrecy, files, deadlines and clients who want to know where things stand. What AI does there is on AI for law firms; which software a firm uses and where a client portal fits, on legal practice management software. A portal where clients upload their own documents is on client portal.
We have no notary's office as a client. What we did build, and what works the same way: a management platform for Business Centrum with files, documents, a log and a portal with a magic link, and AMOSI, our own AI platform for small businesses, which went live in July 2026. How to run such a project is on AI for business.
What notaries ask us about AI
- May a notary use ChatGPT?
- For general texts without file data, a business plan with a data processing agreement can be enough. For client files, choose a setup where the documents do not leave the office. The difference between the models is on ChatGPT vs Claude.
- Does a notary's office need an AI policy?
- A written policy makes clear which tool may be used for which task and who checks. Whether it is required is on do I need an AI policy.
- Can a chatbot on our site answer questions about costs?
- Yes, from texts the office approved, and with a clear limit: for a calculation in a specific file it refers to a staff member. How such a bot works and what it costs is on AI chatbot development and AI chatbot cost.
- Where does an office start?
- With a task that takes many hours and carries little risk, tested on closed files. How a business brings in AI step by step is on implement AI in your business.
- Can the assistant sort client emails too?
- Yes: ordering incoming emails per file and preparing a reply that a staff member reads first. That works like AI customer service, with the office's own documents as the source.
- Can we work with you in English?
- Yes. Meetings, analysis and the written proposal can be in English or Dutch, and the assistant can answer in the languages your clients use.
AI for business
How VBS helps you with this
We build AI into your own work: it drafts emails, reads documents and answers questions, and you stay the one who decides.
- First a task that costs time, then a working prototype
- On your own documents and your own data
- An AI policy your team understands
How we approach it is on our page about AI for business.
An assistant for your office, within professional secrecy?
Tell us which task costs your office the most time and which practice software you use. After a conversation you get a proposal for a first application and a fixed price.
Rather get in touch right away?