AI notetaker transcripts: who owns them and can they be licensed?

Who has a claim to a meeting transcript?#

A meeting transcript involves several parties at once: the company that ran the recording, the people who spoke, any outside guests, the notetaker vendor and the meeting platform. Each holds a different kind of interest, and a license has to respect all of them.

The rights map below is a starting point for a review. It describes the usual position, not the answer for any specific meeting.

What the notetaker vendor's terms decide#

Notetaker vendor terms decide whether your company can export transcripts in a usable form and whether the vendor has already used them for its own purposes. Some vendors reserve a right to use customer content to improve their models unless an admin turns it off, and others restrict using exported content to build competing products.

Read the current terms and the settings page together, because the default can differ by plan and the admin may have changed it. Record what the terms said at the time each recording was made, since terms change over time.

• 1Ownership clause: does the customer own its content, or only receive a license to it?
• 2Vendor use: may the vendor use transcripts to train or improve its own models, and was that setting on?
• 3Export: are transcripts, speaker labels, timestamps and summaries available through an API or bulk export?
• 4Downstream restrictions: do the terms limit what you may do with exported content?
• 5Retention: how long are audio and transcripts kept, and does deletion in the app reach backups?
• 6Subprocessors: which third parties processed the audio for transcription?

How recording consent works across states#

Recording consent in the US is set mainly by state law. Some states require every participant to consent to a recording, while others require only one participant's consent. A call that includes people in different states can bring the stricter rule into play, which is why many companies apply an all-party notice by default.

At least one widely used notetaker vendor has been sued in the US over bots that allegedly recorded people who had not agreed, and over the use of those recordings to train the vendor's own models. Those are allegations against a vendor, not findings about their customers, but they show why both the recording notice and the vendor's training setting belong in any review of an archive.

Notetaker bots that join a meeting and announce themselves, or show a recording banner, are one way companies give notice. Whether that notice was adequate for a given call, and whether people outside the company had a meaningful chance to object, is assessed with counsel.

Consent to record is not consent to license. An employee who knew a standup was recorded for note-taking did not necessarily expect the transcript to reach an outside AI developer. That gap is closed with clear employee notices going forward, de-identification, and careful scoping of historical recordings.

Which transcripts are worth licensing?#

Transcripts are most useful to AI developers when they capture how a team reasons through a problem and connect to a record of what happened next. An incident review that references the Jira issue, the pull request and the fix is far richer than an unlinked conversation.

Preparing transcripts for a license#

Preparing notetaker transcripts means removing personal and confidential details while keeping the structure that makes the conversation useful. Names become role labels such as engineer or support lead, customer names become placeholders, and email addresses, phone numbers and account identifiers are removed. Ticket and pull request references can stay if they point to records in the same package.

Automated detection is a first pass, not a finish line. The open-source Presidio project states in its own documentation that automated detection gives no guarantee of finding all sensitive information and that additional protections should be used. Spoken language makes this harder, because people spell names aloud, mention customers casually and speak over each other, so sampled human review is part of the process.

Generated summaries and action items need the same treatment as verbatim text. A summary can name a customer or a struggling employee in a single line.

Illustrative: an analytics software company sorts its notetaker archive#

Illustrative: a fictional analytics software company has used a notetaker connected to Google Meet and Zoom for several years. The archive holds standups, incident reviews, customer calls and interviews in one workspace.

The general counsel starts with the vendor terms and confirms the company owns its content and can bulk export transcripts. Using calendar metadata, the team keeps only meetings where every attendee had a company email address, drops anything titled or tagged as HR, legal, interview or board, and excludes recordings from before the employee notice was added to the handbook.

What remains is a set of engineering and product meetings, many referencing Linear issue keys. After de-identification and sampled review, the company decides to include those transcripts in a package alongside the linked issues, and to keep customer-facing calls out entirely.

How SourceX approaches meeting records#

SourceX starts with a metadata-only fit check: which notetaker and platforms were used, the meeting categories, the date range and the notices in place. No transcripts are shared at that stage. In the Rights step of the SourceX five-step transaction, vendor terms, employee notices and outside-party exposure are reviewed before any Preparation work begins.

The de-identification method, review sampling and exclusions are recorded in the SourceX Evidence Packet as the privacy record, alongside provenance, licensing rights, permitted use and release authorization. The supplier signs off on the final scope before Delivery begins.