Last updated: September 12, 2026
Everything below is about Genesys Cloud. The older on premises line, PureConnect and the Engage side, runs a different recording stack with its own policy editor, and none of it was verified here.
How long Genesys Cloud keeps a call recording
Genesys Cloud publishes no single retention figure for call recordings. Retention is whatever a quality management policy sets, and the documented options are to keep a recording indefinitely, keep it for a set time and then delete it, or delete it. Turning on recording at the SIP trunk retains every external call by default.
That default is where the confusion starts. An organization that never wrote a policy is keeping everything, which surfaces as a storage problem rather than a compliance one.
The other direction hurts more. Somebody wrote a policy in 2023 with a delete action on it, and the calls you wanted for a client dispute aged out on a schedule nobody has checked since. Policies run forward only, applying to interactions that occur after you publish them, so you cannot write one to clear out existing recordings either.
One ceiling sits above all of it. A separate setting called the maximum interaction data retention time deletes recordings past its window whether or not a quality management policy governs them. Screen recording policies cap at one year, while call and digital recordings have no published upper limit.
Deleting audio deletes more than audio. A transcript carries the same retention period as its recording, and it goes when the recording goes, along with the topic and sentiment events detected inside it. Six years of scored history can evaporate behind one policy edit.
Say you run 40 agents taking 60,000 calls a month. Three years of indefinite retention is north of two million files you are paying to keep, and nobody notices because it arrives as a line under fair use rather than a folder anyone browses.
Where the recordings actually sit
Policy based recording captures two ways. Trunk side recording, also called line recording, is the main mechanism and works from the SIP trunk. Station side recording runs on a user's station and is the route to internal calls, which the trunk never sees. Enable both for the same user on external calls and you get duplicates.
A finished recording lives on the interaction's detail page and on the page for an evaluation, which answers where you browse them.
Recording Management holds the choice between storing call recordings in your home region or in the region set by your Global Media Fabric trunk configuration. The same screen flags orphaned recordings, meaning ones existing only on a Genesys Cloud Edge device. Read that setting before somebody with a data residency clause reads it for you. What you are obliged to keep is covered in call recording retention requirements by industry.
Parts of a call can be missing on purpose too, since recording is suppressible at IVR, on hold, and during queue wait.
Getting recordings out of Genesys Cloud
One at a time, the documented path is Performance > Workspace > Interactions, open an interaction, then Download in the Recording Information section of the Details tab. The File Format list offers .ogg_opus, .mp3, .wav, and .webm, with OPUS downloading fastest when the trunk recorded in OPUS. Generated URLs expire on a timer that defaults to 60 minutes and cannot be set above it.
At volume there is an AWS S3 recording bulk actions integration. You create IAM resources in your own AWS account, add the integration, then either enable Export Recording with Integration on a policy or invoke the Recording Bulk Job API. Exports land in a folder tree keyed by organization ID, then year, month, day, hour, and conversation ID, each file named for its recording ID with a matching _metadata.json beside it carrying start time, duration, media type, direction, ANI, and user IDs. A given recording exports once and gets skipped on a duplicate request.
The API looks like the heavy option, and for a first evaluation pull it usually is, since a policy exporting to your own bucket does the same work without anybody writing code. That flips on old audio. Policy driven bulk export is documented only for interactions under 1.5 years, while bulk export, delete, and archive through the Bulk Action API keep working at every age, as does the GET Recording API. Archived recordings export without a restore, though a batch download of them needs one first.
Endpoint paths and limits sit in the Developer Center, which did not return a readable page on 2026-09-12, so nothing about their behavior is claimed here.
Age degrades what you can do with a recording that survives, too. Past five years you keep the audio, playback, and retention controls, with details thinned to interaction ID, type, participant names, and the transcript if transcription was on.
Dual channel or mixed, which decides speaker attribution
This is the fact that matters most here, and almost nobody checks it until scoring is already live.
Dual channel is a checkbox in the external trunk settings, and its default setting is disabled. Enable it and the system saves each channel in a separate stream, using audio channel 0 for the external participant and channel 1 for the internal one. It becomes available only when the codec is PCMU, PCMA, L16, or Opus, and transcription requires one of those four with dual channel turned on.
A mixed recording puts the agent and the caller in one stream, so software has to work out who was speaking from voice characteristics alone. That job is speaker diarization, and it separates knowing a disclosure was read from knowing the agent read it.
Turning the checkbox on is the right first move. It does less than it sounds like, because a mixed file never becomes two channel later, so the history you already have runs through diarization anyway.
Which is survivable, and worth saying plainly rather than selling around. Criteria that hang on who said a specific phrase get shakier on mixed audio, and that holds for any tool reading the file, ours included. A do-not-say term or a missing TPMO disclaimer still gets found. Attribution of that line to the agent rather than the caller is where confidence drops, which is why every flag should carry a timestamp a supervisor can jump to and a confidence score worth sorting by.
What an administrator can pull out today
The practical question is whether you can evaluate an outside scoring tool without opening a professional services engagement. On Genesys Cloud you can.
An administrator with download permissions pulls individual conversations from the Interactions view in four formats, today, with no project. For a real sample the S3 bulk export is the better route, since one export gives you audio plus a metadata file per call carrying the conversation ID, the ANI, the direction, and the duration. Check the trunk's dual channel setting first, because one checkbox decides more about an evaluation than the export method does.
Recordings from Genesys Cloud reach CallSniper by export and upload today, and a direct connector is not shipped. Upload is how nearly every account starts, and the current platform position for everything else lives on the integrations page.
For a read on your own audio instead of a description of one, bring a day of exported conversations and their metadata files to a working session and we will score them against your rubric on screen.
Frequently asked questions
How long does Genesys Cloud keep call recordings?
Retention runs as long as your quality management policy says, and indefinitely if no policy touches them, since recording enabled on the SIP trunk retains all external calls by default. A separate maximum interaction data retention time deletes recordings past its window even when no policy governs them.
Does Genesys Cloud record in stereo?
Only where somebody enabled the dual channel checkbox in the external trunk settings, which ships disabled. With it on, channel 0 holds the external participant and channel 1 the internal one, and the codec must be PCMU, PCMA, L16, or Opus.
How do I export Genesys Cloud recordings in bulk?
Through the AWS S3 recording bulk actions integration, driven either by a quality management policy or by the Recording Bulk Job API. Policy driven export is documented for interactions under 1.5 years old, while the Bulk Action API works at any age.
Can CallSniper connect directly to Genesys?
No. Recordings reach CallSniper by export and upload, which works on day one and needs nothing from your phone platform beyond the files and their metadata.
Sources
- Genesys Cloud Resource Center, all read 2026-09-12: "Call recording overview in Genesys Cloud", "Manage recordings", "Retention period for analytics data and recording", "External trunk settings", "Download a recording", "About the AWS S3 recording bulk actions integration", and "Work with exported recordings in AWS S3 bucket". No vendor last updated date appeared in the captured region.
- Genesys Cloud Developer Center, developer.genesys.cloud. Returned no readable body on 2026-09-12, so API specifics are unverified.
- PureConnect and Engage documentation on help.genesys.com and all.docs.genesys.com. Not read, so the on premises line is unverified here.
- Fetch record and captures at evidence/genesys-2026-09-12/.