Anonymization
Feature video
Who is this for?
- Organizations with strong confidentiality requirements (regulated industries, human resources, healthcare, legal) that cannot keep named content from their conversations.
- Teams that want the AI insights of their meetings without naming the people mentioned in them.
Permissions and licenses
- Requires a license that includes the feature (Business).
- Enabling it at company level requires the right to edit the company settings (administrator role).
- By default, only an administrator can set anonymization. A team can set it for itself only if an administrator has enabled "Allow users to change these settings" in the company settings.
How to use it
Enable anonymization
The setting is in the "Anonymization" section of Settings > General > Meeting rules, under "Automatically anonymize generated content". It is available on two tabs:
- Company: applies to every meeting of the company;
- Team: applies to the team's meetings.
A meeting is anonymized as soon as one of these levels is enabled: if the company enforces anonymization, teams cannot disable it. The setting is off by default.
Check that the setting applies to the next meetings
Anonymization is not retroactive. Each meeting keeps the state of the setting at the time it is recorded: enabling anonymization protects the following meetings without touching the ones already recorded, and disabling it does not restore the names of past meetings, which stay anonymized.
Read an anonymized transcript
Speakers stay identified: each turn carries the participant's name, as in a regular meeting. Everything spoken that could identify a person, however, is replaced with a label:
[CITED PERSON 1]for a person's name, whether they attend the meeting or not;[EMAIL 1]for an email address;[PHONE 1]for a phone number;[REDACTED NUMBER 1]for any other identifying number: IBAN, payment card, customer, contract or ID number.
Numbering is consistent across the whole meeting: the different spellings of the same name receive the same label, even when the automatic transcription misspelled it.
Find the labels in the generated content
The summary, chapters, tasks, translations, subtitles, Ask Leexi answers, exports and share pages are produced from the anonymized transcript: the people mentioned appear under their label. The meeting participants, however, stay identified: the summary can name them, as the call page does. There is nothing else to configure.
Going further
- Only the content of the conversation is anonymized. The participant list, the meeting title, the speaker attribution, the note sent to the CRM and the notifications still show who attended. To also hide the participants' identity, restrict access to the call with the access rules.
- Snippets inherit the setting. A snippet created from an anonymized meeting is anonymized too.
- Reverting an already anonymized meeting is not possible from the interface. If an anonymized transcript is unusable, contact Leexi support.
Frequently asked questions
Things to watch out for
- The audio is not anonymized. If your confidentiality requirement also covers the recording, restrict access to the call or enable the media purge.
- Participants stay identified. Anonymization protects the people mentioned in the conversation, not the identity of the meeting participants.
- The setting is not retroactive. Enabling anonymization does not protect the meetings already recorded.
- Anonymization is permanent. An anonymized meeting cannot be brought back to its named version from the interface. Test the setting on a low-stakes meeting before enabling it for the whole company.
- Detection is not infallible. A misspelled name can escape masking, a common word can be masked by mistake, and a long number (seven digits or more) can be replaced even when it identifies nobody. Reread a few transcripts after enabling it.
- Speaker identification is reduced on some calls. On VoIP calls, manually imported files and recordings from the mobile and desktop apps, Leexi can no longer name the speakers from the transcript: they appear as "Speaker 1", "Speaker 2" and can be renamed by hand.