What does the AI see? Pseudonyms instead of names
Short answer
The AI gets your data with pseudonyms: names, companies, emails, phone, ID, customer, contract and account numbers, addresses, links and your secret words become codes such as [PERSON_1] and [VIRKSOMHED_2]. A date of birth becomes an age group, a city becomes a region – in free text too – and a number or date can become a range. Words about, for example, health or trade union membership become [personal matter] by default. Gender is written neutrally. The same name gets the same code everywhere, so the AI can still work. The real values are only shown at GoalCube. In free text, names and numbers are found using lists and patterns, so a detail can occasionally slip through; check the preview before the dataset is saved.
See it with example data
Choose a dataset. On the left is your data, which is stored at GoalCube in the EU. On the right is what the AI sees. Point at a name to see its pseudonym.
Customer enquiries September 20265 rows
| Customer | Location | Topic | |
|---|---|---|---|
| Sarah Murphy | sarah.murphy@example.com | Cork | Refund |
| Daniel Kelly | daniel.kelly@example.com | Galway | Delivery |
| Claire Walsh | claire.walsh@example.com | Limerick | Complaint |
| Thomas Camilleri | thomas.camilleri@example.com | Sliema | Delivery |
| Emma Byrne | emma.byrne@example.com | Dublin | Refund |
Enquiry Hi, my name is Sarah Murphy and I returned order 7012345678 two weeks ago, but I still haven't received my money. Please transfer it to IE00 AIBK 9311 5212 3456 78. Kind regards, Sarah
| Customer | Location | Topic | |
|---|---|---|---|
| [PERSON_1] | [EMAIL_1] | Southern | Refund |
| [PERSON_2] | [EMAIL_2] | Northern and Western | Delivery |
| [PERSON_3] | [EMAIL_3] | Southern | Complaint |
| [PERSON_4] | [EMAIL_4] | Malta | Delivery |
| [PERSON_5] | [EMAIL_5] | Dublin | Refund |
Enquiry Hi, my name is [PERSON_1] and I returned order [ID_1] two weeks ago, but I still haven't received my money. Please transfer it to [IBAN_1]. Kind regards, [PERSON_1]
Sarah Murphy and Emma Byrne are waiting for a refund and should get a reply today. Thomas Camilleri has waited 10 days for a parcel. Suggested reply to Sarah Murphy: "Dear Sarah Murphy, thank you for your patience. Your money is on its way and will be in your account within 5 working days."
[PERSON_1] and [PERSON_5] are waiting for a refund and should get a reply today. [PERSON_4] has waited 10 days for a parcel. Suggested reply to [PERSON_1]: "Dear [PERSON_1], thank you for your patience. Your money is on its way and will be in your account within 5 working days."
Recruitment spring 20274 rows
| Candidate | Born | ID number | Employer | Experience |
|---|---|---|---|---|
| Louise Brennan | 1985-03-12 | 1234567TA | Shannon Timber Ltd | 12 |
| Michael Doyle | 1992-11-02 | 2345678KB | Liffey Logistics Ltd | 4 |
| Maria Borg | 1988-06-21 | 0123488M | Grand Harbour Metals Ltd | 9 |
| David Fenech | 1979-01-30 | 0456779M | Valletta Pharma Ltd | 18 |
CV Louise Brennan has 12 years' experience in budgeting and monthly reporting at Shannon Timber Ltd. She implemented a new ERP system and leads a team of three. Contact: louise.brennan@example.com.
| Candidate | Born | ID number | Employer | Experience |
|---|---|---|---|---|
| [PERSON_1] | 40–49 years | [ID_1] | [VIRKSOMHED_1] | 12 |
| [PERSON_2] | 30–39 years | [ID_2] | [VIRKSOMHED_2] | 4 |
| [PERSON_3] | 30–39 years | [ID_3] | [VIRKSOMHED_3] | 9 |
| [PERSON_4] | 40–49 years | [ID_4] | [VIRKSOMHED_4] | 18 |
CV [PERSON_1] has 12 years' experience in budgeting and monthly reporting at [VIRKSOMHED_1]. The person implemented a new ERP system and leads a team of three. Contact: [EMAIL_1].
Louise Brennan and Maria Borg have more than 5 years' experience and should be invited to interview; ask Maria Borg about experience with ERP systems. Michael Doyle has 4 years' experience; the requirement is 5. A person makes the decision.
[PERSON_1] and [PERSON_3] have more than 5 years' experience and should be invited to interview; ask [PERSON_3] about experience with ERP systems. [PERSON_2] has 4 years' experience; the requirement is 5. A person makes the decision.
Supplier contracts 20264 rows
| Supplier | Contact | Account | Expires |
|---|---|---|---|
| Atlantic Components Ltd | Peter Walsh | IE00 IRCE 9205 0112 3456 78 | 2027-01-31 |
| Malta Data Hosting Ltd | Anna Grech | MT00 MALT 0110 0001 2345 MTLC AST0 01S | 2026-12-31 |
| Midlands Precision Ltd | Simon Nolan | IE12 BOFI 9000 0112 3456 78 | 2028-06-30 |
| Celtic Packaging Ltd | Nicole Ryan | IE02 AIBK 9310 6512 9876 54 | 2027-03-31 |
Terms Atlantic Components supplies specialist components for Project Skellig. The price is adjusted annually by the consumer price index plus 3%. Liability is capped at 10% of the annual contract value. Contact Peter Walsh about any changes.
| Supplier | Contact | Account | Expires |
|---|---|---|---|
| [VIRKSOMHED_1] | [PERSON_1] | [IBAN_1] | 2027-01-31 |
| [VIRKSOMHED_2] | [PERSON_2] | [IBAN_2] | 2026-12-31 |
| [VIRKSOMHED_3] | [PERSON_3] | [IBAN_3] | 2028-06-30 |
| [VIRKSOMHED_4] | [PERSON_4] | [IBAN_4] | 2027-03-31 |
Terms [VIRKSOMHED_1] supplies specialist components for [PROJEKT_1]. The price is adjusted annually by the consumer price index plus 3%. Liability is capped at 10% of the annual contract value. Contact [PERSON_1] about any changes.
Atlantic Components Ltd renews automatically unless it is terminated by 31 October. It is important for an internal project, so renegotiate the price rather than terminate. Celtic Packaging Ltd also renews automatically; the deadline is 31 December.
[VIRKSOMHED_1] renews automatically unless it is terminated by 31 October. It is important for an internal project, so renegotiate the price rather than terminate. [VIRKSOMHED_4] also renews automatically; the deadline is 31 December.
What gets protected
| In your data | The AI gets |
|---|---|
| Names of people and companies | [PERSON_1], [VIRKSOMHED_1] |
| Email, phone, address, links | [EMAIL_1], [TELEFON_1], [ADRESSE_1], [LINK_1] |
| National and other ID numbers from EU countries | [ID_1] |
| Customer, order, contract and account numbers, IBAN | [KUNDENR_1], [KONTRAKT_1], [IBAN_1] |
| Your secret words | e.g. [PROJEKT_1] |
| Date of birth | An age group, e.g. 40–49 years |
| City – in a column and in free text | A region, e.g. Region Syddanmark |
| A number or date about the person, e.g. length of service (if you choose it) | A range, e.g. 3–5 years |
| Words about health, trade union membership, religion, politics, ethnic origin, sexuality and criminal convictions in free text | [personal matter] (you can choose something else) |
| He, she, his, her | The person, the person's |
The protection covers free text too, such as CVs and customer enquiries. In free text, GoalCube finds names, emails, ID, phone and account numbers and addresses using lists and patterns. A detail can occasionally slip through, for example a name that is not on the lists or a number written in an unusual way. That is why the upload shows a preview of what the AI sees before the dataset is saved. Words that can point to a person, such as "only" or "CFO", are marked in yellow, and you choose for each word; the marking works on text in Danish, English, German, French, Dutch, Italian, Spanish, Swedish, Finnish and Polish, and the placeholder is in the dataset's language. See Upload a table. Look through the preview, and add words that must be hidden to your list of secret words. See all the options in Adjust the protection per column.
Each text field is read up to 20,000 characters. If a field is longer, it is cut off, and the upload page shows how many fields this applies to.
Same person, same code
"Louise Brennan", "Louise" and "LOUISE BRENNAN" all become [PERSON_1] throughout the dataset. That way the AI can compare and count without knowing the name.
Pseudonyms are not anonymous
For you, the data is still personal data, because you hold the key. GoalCube reduces the risk, but the rest of your GDPR work still needs to be in order. A person can also be recognised without a name; that is why, for tables about people, GoalCube shows a recognition risk (low, medium, high) for each record, see Datasets.