AI and data notice

1. What is sent to AI, and when

voye sends text to AI only when you use a feature that needs it:

• a review of a document (general, for an audience, or against a brief), together with the brief, the audience and any knowledge-base passages the review needs;
• an AI edit of a passage (Shorten, Make stronger, Rephrase);
• a new document written from one of your documents;
• importing a file, where the file's text is converted into a document;
• formatting a document with AI.

The text is sent to OpenRouter, which passes it to the AI model provider that answers the request. These providers may be located outside the European Economic Area; see section 8 of our Privacy Policy for the safeguards we rely on. The text is processed to return the result to you. We do not use it to train AI models.

Your documents are always saved to your account as you wrote them, whether or not anonymisation is on. Anonymisation applies only to what is sent to AI.

2. How anonymisation works

When "Anonymise everything sent to AI" is on in your account:

• before any text leaves your browser for AI, voye looks for personal details in it: people's names, email addresses, phone numbers, links and places;
• detection happens on your device, using fixed patterns together with an on-device language model (or your browser's built-in AI, or in the desktop app Apple Intelligence or macOS's own name finder, where available); the details found are not sent anywhere to be detected;
• the on-device model is made mainly for ten languages: English, Dutch, German, French, Spanish, Italian, Portuguese, Latvian, Arabic and Chinese; text in other languages is still checked, but names and places are found less reliably;
• each detail found is replaced by a label such as [Person 1] or [Place 2], used consistently across the text, and a name found once is also replaced wherever it appears again;
• the AI answers using the labels, and voye puts the original details back on your screen only;
• if the text cannot be anonymised, for example because the on-device model could not be loaded, it is not sent and the request fails instead;
• formatting with AI is unavailable while the setting is on, because it works from the saved, non-anonymised text.

When the setting is off, text is sent to AI as you wrote it.

3. Limits of anonymisation: no guarantee

Automated detection is imperfect. Although we make every reasonable effort, we do not represent or warrant that anonymisation will remove all personal data, or that the text that is sent cannot be used, alone or combined with other information, to identify you or someone else. In particular:

• unusual names, nicknames, names that are also ordinary words, misspellings and initials may not be detected, and detection is less reliable in languages other than the ten listed in section 2 and in text that mixes languages;
• details that are not names, contact details or places are not removed, including employers, job titles, schools, dates, project and product names, figures and distinctive achievements, and these can identify a person in combination;
• identifiers written in an unusual form (for example an email address split over lines, or a phone number written in words) may not be detected;
• the detector may occasionally replace text that is not personal, which can make the AI's answer less useful;
• anonymisation does not apply to what you send outside AI features, such as feedback messages to us, or to documents stored in your account.

Anonymisation reduces the personal data that AI providers receive; it is not a guarantee of anonymity within the meaning of data protection law, and anonymised text may still be personal data.

4. What you can do

• Keep "Anonymise everything sent to AI" on if you prefer that personal details are replaced before text is sent.
• Read your text before using an AI feature and remove anything you do not want to share, especially special categories of data (such as health information) and data about other people.
• Use placeholders such as [Name] or [Company] for details that the review does not need.
• If you include personal data about other people, make sure you are allowed to share it.

5. About AI output

AI output can be inaccurate, incomplete or unsuitable for your purpose. voye instructs the AI not to invent facts and to mark missing information as [placeholders], but it cannot guarantee that the output is correct. Review every suggestion before you apply it. Output is not legal, career, financial or other professional advice.

6. Responsibility

To the fullest extent permitted by law, and as set out in section 16 of our Terms of Service, we are not liable for loss arising from personal data that anonymisation did not remove from text you chose to send to AI, or from your reliance on AI output, except where caused by our intent or gross negligence or where liability cannot be excluded by law. Nothing in this notice limits your rights under the GDPR or other data protection law.

7. Questions

If you have a question about how voye handles data sent to AI, or believe anonymisation missed something it should have found, contact us at [privacy@your-domain]. See our Privacy Policy for your rights.