In this lesson, you will learn how to hide selected terms and contact details, test replacement, and recognize which information a model needs for the task.
Try it: draft a message about a fictional client
Pretend you need a generic reminder for a fictional client called “Practice Client Maple”.
Open the Settings gear menu at the top and choose Anonymize.
Under Obstruct words, add “Practice Client Maple”. Separate multiple terms with commas and click Save. Add name variants separately when needed.
Under Contact details, review Email addresses and IBANs. Turn on the rules you need. These switches are not automatically enabled for every account; unlike the term list, changes to these switches save automatically.
Start New chat and send:
Write a polite reminder to Practice Client Maple that our meeting is tomorrow at 10:00. Do not invent any other details.Where restored text is marked, hover or tap it to inspect what the model actually saw. Confirm that replacement works before relying on it with real information.
Remove the fictional practice term afterwards and Save if you no longer need it. Workspace rules may still apply even when your personal switch is off.
Keep necessary meaning visible
If you add Swisscom to hidden terms and then ask about Swisscom offers, the model sees a substitute token instead of “Swisscom”. It cannot reliably use that name to identify the provider or research the right offers. A generic reminder may work with a hidden company name; a question about that company's products needs the real identity. Decide which information is necessary before hiding it.
Text replacement does not guarantee redaction of image pixels, audio, every attachment, or external tool inputs. Connected tools may need restored values to address real records. Remove sensitive content manually where that boundary matters; do not assume every personal detail is detected.
Check your result
You have tested one harmless term and can explain when to hide an identity and when that identity is essential to the task.