Call Transcription and Privacy: Who Reads It, Where It's Stored, and When It's Deleted

Why a transcript is more sensitive than a recording

A recording is an audio file. To know what's in it, you have to sit and listen from start to finish. A transcript is text. You read it in half a minute, search it for a word, copy a sentence from it and forward it by email.

That's exactly what makes transcription so useful: a manager can go through ten calls in the time it used to take to listen to one. But it's also what calls for care. An ID number a customer read out, a medical detail a mother shared, a family's financial trouble — in a recording they're buried inside an hour of audio. In a transcript they're a single line that a search can find.

So the rule is simple: everything you know about the privacy of recordings applies to transcripts too. And on top of that, you need to think about where the text is sent and who reads it.

When is a call transcribed at all

With us, a call is never transcribed just like that. There are three ways, and in each one someone in your organization explicitly chose it:

• On-demand transcription — an authorized user picks a recorded call and asks to transcribe it.
• Automatic transcription — turned on for a specific line: an extension, a number, a queue, a hunt group, a code or a recording group. From then on, recorded calls on that line where someone spoke with an agent for 20 seconds or more are transcribed and sent by email.
• Voicemail transcription — turned on for a specific voicemail box, and every new message is transcribed and sent to the box's addresses.

There is no single setting that transcribes the whole account. Automatic transcription also works only from the moment it's turned on, and does not go back to older calls.

Where transcription happens

Under Kesher's privacy policy, the recording is transcribed by our own transcription engine, which we run with server providers working for us. The recording is not handed to a third-party AI service, and the transcription engine doesn't keep it — it only passes through to be turned into text.

Why does this matter? Because there's a big difference between a service that transcribes for you and a service that uses your calls for its own purposes. With us, recordings and transcripts are not used to improve the transcription engine without the customer's consent.

What is kept, and for how long

• The transcript text is kept with the call ID, its date and time, and its length, in a separate folder for each account.
• In the "Transcription" screen, the "My Transcripts" list shows all of the account's transcripts from the last 90 days: on-demand, automatic and voicemail.
• After 90 days the transcript is deleted automatically. If you need it again, transcribe it again, as long as the recording itself still exists.
• The listening link sent in the automatic transcription email stays active for 90 days.

Note the difference between the transcript and the recording. The recording is kept according to its recording group's retention period, which you set: 1 to 180 days. The transcript is kept for 90 days. If the recording was deleted earlier, the transcript may still be available until the end of its 90 days, and vice versa.

Deleting early

You don't have to wait 90 days. In the "My Transcripts" list, every transcript has a delete button that removes it after you confirm. This is useful when a transcript was created by mistake, or when a customer asked that their call not be kept as text.

Who may read, request and turn it on

The Kesher control panel has a separate permission for each transcription action, so you can give each person exactly what they need:

• Reading an existing transcript — same as listening to recordings. Someone blocked from listening to recordings can't read their transcripts either. The logic is simple: the transcript is the same call, just in writing.
• Requesting a transcript for a call — a separate permission. A read-only user can't request a transcript.
• Managing transcription — turning automatic transcription on and off for lines and voicemail boxes, choosing the addresses, and the "Transcription" screen. This permission is off by default, and your organization's system administrator turns it on for whoever needs it.

For example, a secretary can listen to and read transcripts of office calls without being able to turn on automatic transcription for the management line or change where the emails are sent.

Transcripts by email: who else reads them

Automatic transcripts and voicemail transcripts are sent by email. Once the email goes out, it sits in the mailboxes of whoever received it, and is kept there under that mailbox's rules, not the PBX's. We can't delete an email that has already reached your mailbox.

So the most important choice in automatic transcription is where it's sent:

• The personal address of whoever is responsible for the line, not a shared inbox that ten people read.
• On a management, finance or counseling line — only to people who would be expected to hear these calls even without a transcript.
• When an employee leaves, remove their address from every line. The advanced search in the control panel finds every place an address appears.

One more thing that helps: each address gets one email per call, even when the call passed through several lines with transcription turned on. That way there are no duplicate copies in the inbox.

What is not transcribed automatically

Even on a line with automatic transcription, not every call becomes text:

• A call that wasn't recorded — no recording, no transcript. The line has to be assigned to a recording group, and the group has to record that direction: incoming, outgoing or both.
• A call nobody answered — not transcribed and not sent.
• A short call — a call where someone spoke with an agent for less than 20 seconds isn't transcribed automatically.
• Calls from before it was turned on — not sent.

From a privacy standpoint, this is an advantage: less text is created, and short "sorry, wrong number" calls don't end up in anyone's inbox.

Voicemail: a special case

With voicemail, it's the other way around. No recording group is needed, because the message itself is the recording, and there's no minimum length: every message is transcribed and sent. While transcription is on for a voicemail box, the regular email the PBX sends from that box is turned off, so each address gets only one email.

Voicemail messages are often more personal than calls: people leave a message when they can't reach a person, and tell the whole story in it. So it's especially worth checking where each box's transcripts are sent. A personal box — to the personal address. A department box — only to whoever handles the requests.

Corrections: what happens when you send one

Transcription is never perfect. Names, terms in Lashon Hakodesh and passages in Yiddish are where it makes the most mistakes. That's why you can send a correction for a sentence in a transcript.

When you send a correction, the audio clip of that sentence and the corrected text are saved — only to improve transcription. Corrections are reviewed by the Kesher team. In other words, a correction is the only time a short clip of your call leaves the usual path, and it happens only when a user on the account chose to send it.

If something especially sensitive was said in a particular call, it's better not to send a correction for that sentence. You can fix it for yourself in the text you copied, without sending anything.

Notice to callers and internal policy

We don't give legal advice, and for questions about what is allowed and what is required, it's best to consult a lawyer. But a few practical steps help any organization:

• If you announce that calls are recorded, think about transcription too. A customer who knows the call is recorded doesn't always assume it will also become text sent by email.
• A written one-page policy. Which lines are transcribed, where the text is sent, who reads it, and what to do when a customer asks for deletion.
• Don't transcribe what you don't need. Automatic transcription suits a sales line or a support line. For the line of a rav who answers personal questions, it may be better without it.
• Review once a quarter. The "Transcription" screen shows every automation running on the account, each with its line, addresses and the date it was turned on. Whatever is no longer needed — turn it off.

Example: a chesed organization with a help line and a management line

A chesed organization runs two lines. On the help line, families call to ask for help, and sometimes describe a health or financial situation. On the management line, staff talk with donors and suppliers.

The director decides:

1. On the management line — automatic transcription, sent only to his personal address. That way, in the evening he reads the main points of the day's calls.
2. On the help line — no automatic transcription. When a coordinator needs to go back to a particular call, she requests an on-demand transcript for that call only.
3. The coordinators have permission to listen and to request transcripts, but not to manage transcription. They can't turn on transcription for the management line or change where it's sent.
4. When a family asks that their call not be kept, the coordinator passes the request to the director, and he deletes the transcript from the "My Transcripts" list, since that screen requires the transcription management permission.

The result: transcription saves time where it really helps, and the families' sensitive information doesn't float around in email inboxes.

Five questions before you turn on automatic transcription

1. Is this line recorded at all, and does it need to be?
2. Who is supposed to read the transcripts, and who definitely isn't?
3. Which address are they sent to, and who else reads that inbox?
4. What do you do when a caller asks to delete their call?
5. When will you check again whether transcription is still needed on this line?