Recorded meetings and calls capture reasoning that is never written down. They are also the company data with the most consent questions attached. This page separates what is usually workable from what usually is not.
Last checked: 7 October 2026. Buyer terms are quoted as published on that date.
Recorded internal meetings on Zoom or Microsoft Teams, sales calls captured by conversation-intelligence tools, recorded support lines, training sessions and interviews. Each can exist in two forms: the audio or video file, and a transcript, often with an automatic summary attached.
Buyer programs list “Zoom recordings and transcripts” among the communication sources they accept. Spoken work is attractive for an obvious reason: people explain, argue, troubleshoot and decide out loud in ways they never type. A design review or a hard customer escalation shows reasoning in real time. That same quality is why calls carry more risk than any written record.
Most sellers think only about the first question. The second one decides more deals.
US federal law and many states allow a call to be recorded when one participant consents. A smaller group of states, usually called two-party or all-party consent states, require every participant to agree. When people join from different states, which rule applies is itself a legal question.
Outside the US, recording rules and data protection law, such as the GDPR for participants in the EU and UK, add their own requirements. If any recording in your archive was made without the consent its location required, it should not be in scope.
“This call may be recorded for quality and training purposes” was said for a reason at the time, usually coaching staff. Whether licensing that recording to an AI company falls inside that purpose is not something a seller can assume.
The same applies to employees in internal meetings: they knew the meeting was recorded, not that the recording might be licensed. See our guide on employees and selling company data for notice and communication.
Other laws to name to your lawyer: state biometric privacy laws, if a voice could be treated as a biometric identifier; HIPAA, if calls touch patient information; GLBA, if they include customers of a financial institution; and attorney-client privilege for anything involving counsel. Ask which apply, what notice or consent each would require, and whether the answer changes for transcripts versus audio. Our page on whether it is legal to sell company data covers the broader picture.
| Question | Transcript | Audio or video |
|---|---|---|
| Who can be identified? | People named in the text, plus anyone recognizable from context. | Everyone who speaks, by voice alone; on video, by face too. |
| How hard to de-identify? | Names, numbers and places can be replaced with consistent pseudonyms. | Spoken names can be muted, but the voice itself remains. |
| What is lost? | Tone, pauses, interruptions and anything shown on screen. | Little, which is also why it exposes more. |
| Extra legal questions | Fewer: mostly what was said. | Possible biometric questions for voices and faces. |
| Accuracy | Automatic transcripts contain errors, including misheard names that slip past redaction. | Not applicable. |
Ask each buyer whether it wants audio at all. If transcripts meet its needs, there is little reason to hand over voices. If it does want audio, ask how voices are handled after delivery, and what the agreement says if a participant is later identified. Who carries that risk is a contract question, covered in our guide on de-identification before selling data.
Many meeting tools now produce summaries and action lists on their own. Those summaries were written by software, not people, so they say little about how your team thinks. They can also carry errors that a reader would take as fact. If you include them, label them as machine-generated in the manifest so the buyer can treat them separately.
The more interesting record is what people did with them. micro1 lists “AI performance feedback,” meaning human feedback on AI outputs, among the data it wants (as published, checked 7 October 2026). If your staff routinely corrected AI notes, rated them or rewrote them, the before-and-after pairs are a different kind of data from the calls themselves. Ask whether a buyer values those corrections, and whether they can be offered without the underlying recording.
A final practical point: the recording platform may hold transcripts, summaries and audio in separate places with separate export tools. Check what your plan lets you export, and in what format, before you describe the archive to anyone.
A rough order of difficulty. Your contracts and locations can move any row.
Stand-ups, design reviews, planning sessions. Only employees present, so the main tasks are notice and removing personal side talk.
Recorded walkthroughs of how work is done. They sit close to SOPs and often contain little personal data.
Outside participants did not agree to your sale, and your contracts with them may include confidentiality terms.
Prospects and customers are third parties, and their words are customer information. Many sellers leave these out entirely.
Callers read out account numbers, addresses and payment details. Redaction must be near perfect, and often is not.
Law, accounting, healthcare and M&A calls carry confidentiality duties and sometimes privilege. Treat them as excluded.
A transcript on its own shows people talking. A transcript next to the ticket, document or project it was about shows a decision being made and then carried out. micro1 names “decision-making patterns” among the things it wants, and its source list puts Zoom recordings beside Jira, Confluence and Slack. That suggests where call data adds most: as the spoken layer of work that is also written down elsewhere.
In practice, that means scoping by project or team rather than by recording tool. If the engineering team’s tickets and design documents are in scope, their internal design-review transcripts for the same period may add context. If a team’s written records are excluded, its calls probably should be too, because the same sensitive topics will come up out loud.
It also means recordings rarely need to be the first thing you offer. Many companies start with documents and tickets, see how a buyer handles de-identification and acceptance, and only then decide whether transcripts are worth the extra consent work.
No buyer we track publishes a rubric for recordings. These contrasts follow from what buyers say they want: records of how work was actually done and decided.
| Signal | Stronger | Weaker |
|---|---|---|
| Content | Design reviews, incident calls, planning and problem-solving sessions. | Status updates, social chat, meetings that end without a decision. |
| Continuity | A recurring meeting on one project over months, so decisions can be followed. | Scattered one-off calls with no thread between them. |
| Speaker labels | Each speaker consistently labeled, so a pseudonym can replace each one. | One undivided block of text with no speakers marked. |
| Context | Meetings tied to tickets, documents or projects in other systems. | Recordings with no title, agenda or link to the work. |
| Accuracy | Clear audio and transcripts that read correctly. | Garbled automatic output that a reviewer cannot follow. |
| Consent trail | Notices and policies documented for the whole period. | Nobody knows what participants were told. |
One workable order of operations for a first deal built on internal meetings.
Manager and employee calls carry performance, health and personal matters. They are rarely labeled as such in the recording tool.
“Weekly sync” can turn into a conversation about a client or a colleague. Titles filter the obvious; samples catch the rest.
A vendor who joined for ten minutes is still an outside participant who did not agree to your sale.
Some exports bundle audio or video with the transcript. Check the delivery format, not just the plan.
Recordings kept past the period your policy promised are a problem before any buyer sees them.
Employees who learn from outside that their meetings were licensed lose trust fast. Tell them first.
No price is given: nobody can value a recording archive without reviewing it. The company could revisit the excluded sales calls later, once counsel has reviewed the notices prospects heard and the contracts that governed those conversations.
Quoted from each program’s own pages, checked 7 October 2026. Ranges are what programs publish, not what any company will receive.
micro1 publishes “$100k+ qualified,” “$500k+ large-scale” and “$1M+ highly unique,” with 30+ employees and primarily English; it states that scope is agreed in writing and sensitive information is scrubbed. Mode publishes “$100K-$5M,” with 20+ full-time US office employees, and says it de-identifies before onward delivery. Grepped publishes “$20K-$5M” for any vertical. Ask each how it treats recordings specifically.
Independent site. Some links are referral links: if your company signs with a buyer through them, the buyer may pay us a fee. You are not charged, and we never see your data.