AI and privacy
How a healthcare practice or healthcare administration uses PrivaSearch and its own AI assistant to search intake reports, care contracts and complaint files.
A healthcare practice or healthcare administration uses PrivaSearch to search intake reports, care contracts, protocols and complaint files with an AI assistant, without giving the names, BSNs (Dutch citizen service numbers) and addresses of clients to the provider. A colleague checks every document before it becomes searchable, and every answer comes with its source.
This article is about the administrative side of healthcare: the practice manager, the healthcare administrator and the secretariat, with made-up situations from a physiotherapy practice, a mental healthcare practice and a home care organisation. We wrote earlier about the rules for AI in healthcare in general in AI in healthcare.
Health data is special category data. The GDPR prohibits its use, except in a limited number of exceptions; the Autoriteit Persoonsgegevens (AP), the Dutch data protection authority, explains this on its page about using and sharing health data. An intake report, a referral letter or a letter of complaint almost always contains such data.
Professional secrecy applies as well. For medical treatment, it appears in Article 7:457 of the Dutch Civil Code, part of the rules known as the WGBO. Anyone who pastes a report into a chatbot sends it in its entirety to the provider; the AP warns that this can lead to a data breach.
In February 2025, the Health and Youth Care Inspectorate (IGJ) called on healthcare providers to take care when introducing generative AI. The inspectorate points out that generative AI can produce incorrect or incomplete information, and that the healthcare provider remains responsible for careful use.
PrivaSearch is software that lets your team search its own client files with Claude, ChatGPT or Mistral Vibe, after names, BSNs (Dutch citizen service numbers) and addresses have been replaced by pseudonyms and a colleague has checked that.
Where a letter of complaint contained a client's name, date of birth and BSN, the assistant reads "[PERSOON_B4A1]", "[GEBOORTEDATUM_788C]" and "[BSN_361B]". The healthcare professional simply keeps working in the practice's patient record system; PrivaSearch does not replace it.
A physiotherapy practice has contracts with several health insurers, each with its own conditions on referrals, treatment duration and billing. The practice manager wants to know which contracts require a referral for a particular treatment. Contracts contain little data about clients, but they do contain names of contact persons and IBANs. In a project with the contracts, she finds the answer with the provision and the page attached, and she checks it in the contract itself.
A home care organisation receives a complaint and wants to know what was agreed earlier in the correspondence with the client and the family. The complaints officer puts the letters and reports in a shielded project. After the review, she asks the assistant which arrangements about visiting times are in the documents. Because only the members of that project have access, the rest of the organisation sees nothing of the complaint.
Employees often ask questions about work instructions: what is the procedure for an incident, how long do we keep a particular document, who has to sign a report. Protocols contain hardly any personal data, but they are long and spread across folders. A project with all the protocols makes them searchable in one go, with the source for every answer.
A mental healthcare practice wants to know how often intake reports from the past year refer to a particular form of support, in order to align its services better. This is the most sensitive example. Detection is never complete, and a description of a complaint or a life event can make someone recognisable. That is why the colleague in the review looks extra carefully here, and the practice decides in advance whether such reports belong in a project at all. If in doubt, leave them out.
| Task | PrivaSearch and your assistant | The practice or healthcare professional |
|---|---|---|
| Finding a provision, arrangement or procedure | Searches released passages and gives the source | Rereads in the document itself |
| Shielding personal data | Replaces it with pseudonyms; a colleague checks | Assesses whether it is complete |
| Making a diagnosis or choosing a treatment | Does not do that | Decides as a healthcare professional |
| Keeping the client's record | Does not do that; it is not a patient record system | Works in their own patient record system |
| Keeping track of who did what | Records every reveal in a log entry | Can consult the log |
PrivaSearch is neither a patient record system nor a medical device. It is a search layer for documents. Whether it fits within the professional secrecy of your practice or institution is for you to assess, with your data protection officer.
Choose deliberately which documents go in. Start with documents containing little health data, such as contracts and protocols. Only expand to reports about clients once you know how the review works in your practice.
Work with projects and permissions. Permissions apply per project, and administrators do not get automatic access. You put a complaint file in a separate project with only the people who need it.
Connect only a business account. What PrivaSearch returns to the assistant reaches the provider of that assistant, with pseudonyms instead of names. So connect a business account with a data processing agreement; which subscription qualifies is set out in Which subscription for business use?.
Do not type names into your assistant. What you type into the assistant yourself goes directly to the provider, bypassing the pseudonymisation.
Give every project an end date. After that, access is closed and copies, fragments and pseudonyms are cleared away, unless a legal hold applies.
Document it. With health data, a DPIA is quickly required; how to approach one is set out in The DPIA for AI tools. For this processing, PrivaSearch supplies a DPIA starter document and a register entry, derived from the set-up of your environment. These are starter documents, not completed assessments. Pseudonymised data remains personal data; why is explained in Pseudonymise or anonymise?.
The steps are on the page How PrivaSearch works. If you want to see it with fictitious files, book a demo.
Start with documents containing little or no data about clients: care contracts, protocols, work instructions and the quality handbook. A lot of search time goes into those, and the risk is small. That way the practice learns to work with searching by source before reports about clients go into a project.
Then agree who does the review. For documents about clients, that is someone who knows the practice and sees when a description makes someone recognisable, even without a name. Record which kinds of documents never go into a project.
Discuss the introduction with your data protection officer, and include the processing in your record of processing activities. You supplement the starter documents that PrivaSearch supplies with what applies to your practice.
Make sure that everyone who searches has completed the short AI literacy training. Without a valid training, the workspace does not open and the connection with the assistant is refused as well. Why that training exists is explained in our article on AI literacy.
Evaluate after a few weeks: does it save time, and are the answers right when you check them?
That is possible, but not in every tool and not without safeguards. Health data is special category data and healthcare professionals are bound by professional secrecy. An intake report does not belong in a chatbot where it goes in its entirety to the provider. Choose a way of working in which personal data is shielded first, a human checks and the healthcare professional remains responsible.
Not if you search via PrivaSearch. Names, addresses, BSNs, IBANs, number plates and dates of birth are replaced by pseudonyms within PrivaSearch, and a colleague checks that before a document becomes searchable. The assistant only receives those released passages, never the original files.
Detection is never complete, and a description of a complaint or a life event can make someone recognisable, even without a name. That is why human review is mandatory: a colleague sees the original next to the pseudonymised text and adds what was missed. In addition, decide in advance which kinds of reports belong in a project.
No. PrivaSearch is neither a patient record system nor a medical device. It is a search layer for documents that replaces personal data before an AI assistant searches them. Diagnoses, treatment choices and keeping the record remain with the healthcare professional and their own patient record system.
In February 2025, the Health and Youth Care Inspectorate called on healthcare providers to take care when introducing generative AI. According to the inspectorate, such AI can produce incorrect or incomplete information, and the healthcare provider is responsible for careful introduction and careful use. So check every answer against the source.
No, not if you put it in a separate project. Permissions work per project: only the members of that project see the documents, and administrators do not get automatic access. If a SharePoint library is connected, every request checks live whether the user may still see the document there.
For each Claude plan: what happens to your input, how long it is kept, whether an agreement is in place and where your data is stored. October 2026.
What you need to arrange for each type of data before using an AI chatbot: from names and email addresses to health, criminal records and BSNs.
Where an uploaded file ends up, how long it stays, who can access it and what deletion means. With four approaches for files containing personal data.
We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.
Book a demo →