Selling sales call recordings and sales emails for AI training

Yes, a company can license its recorded sales calls and its sales emails to an AI buyer, if each call was lawfully recorded, the conversations are its own rather than its customers', and the prospects' personal data comes out before anyone reads them. What it cannot do is sell them as a category of their own: no buyer page we read names outbound sales calls or sales emails with their outcomes as something it buys, and none publishes a price for them. The buyers that come closest buy a company's whole working record. HUMXN Research Labs lists " pipelines, call notes, emails and quotes, with the deals that were won and lost" and says its deals typically start around $250,000; Polyshares names CRM history, email and call recordings among what it buys, at $100K to $2M and above a licence; and Mercor licenses " recorded customer conversations" from call centres and " real usage data from an AI app, agent or developer tool", the second at $20K to $480K a company. Sold alone, as audio, the same calls meet off-the-shelf call-centre recordings that license for $5 to $25 per hour of audio. Sales conversations are worth most joined to the deals they were about, offered with the rest of the company's records.

This guide is for the founder, sales leader or officer of a company that holds recorded sales calls (from a dialer, a meeting tool or a conversation intelligence product), call transcripts, or the outbound emails its sales team or its software sent, with the replies, meetings and deals that followed. It also covers a voice or sales-software platform whose servers hold calls and emails it handled for its customers, and calls placed by AI voice agents. It covers what buyers value in sales conversations, who buys them and what they are worth, whether each call was lawfully recorded, whose conversations a platform holds, the rules for AI-agent calls, sales email, de-identification, preparation, and how a sale through AIData.Broker works. Every quote comes from the page, statute or ruling it names, read on the day shown in its record.

This guide is research from public sources, not legal advice.

Contents

1. What a sales conversation archive holds, and what buyers value in it
2. Who buys sales calls and sales emails
3. What sales calls and sales emails are worth
4. Was each call lawfully recorded?
5. Whose conversations are they? Platforms that hold their customers' calls
6. Calls placed by AI voice agents
7. Sales emails and sequences
8. De-identifying calls and emails
9. Preparing a sales conversation lot
10. How a sale through AIData.Broker works
11. Questions people ask
12. What this means for you

1. What a sales conversation archive holds, and what buyers value in it

A company's sales conversations are never in one place. The calls sit in a dialer or phone system, in a meeting tool such as Zoom, Google Meet or Teams, or in a conversation intelligence product, software that records every call, transcribes it and scores it. The emails sit in the mailboxes of the people who sent them and in the sales engagement tool that ran the sequences (a sequence is the timed series of emails, calls and tasks a rep sends one prospect). And the result of every conversation sits in the CRM: the stage a deal reached, the meeting that was booked, the deal won or lost and the reason given. Each system holds part of the story, and none holds all of it.

What makes this material worth training on is the outcome. A model learning to qualify a lead, answer an objection, write a first email or book a meeting needs to know which conversation worked, and the outcome is the label that tells it. Buyers of company records say so in general terms. HUMXN writes that records which clearly document " real problems, procedures, and outcomes are worth the most", and that buyers care " far more about depth, outcomes and connections between systems than about tidy formatting". Corpus Bridge counts outcomes attached and clean threading among what adds value to conversations. And micro1's revenue chief told reporters that if a company will not sell its Slack, CRM or email records, the rest of the corpus is less valuable.

The companies that sell AI to sales teams show what such data trains. Gong says its product runs on models trained on more than 3.5 billion sales interactions, and Landbase says its model for go-to-market work was pre-trained on over 40 million B2B campaigns and sales interactions. Both describe data they or their customers already hold, not data they bought (section 2). A voice-data company, Liva AI, publishes a sample of unscripted sales calls between a salesperson and a prospect as a speech dataset, each speaker on a separate track, which is how speech buyers want a conversation delivered.

Audio and transcripts go to different buyers

Audio and text are two different products. Speech buyers price by the hour of audio (section 3) and want each speaker on a separate channel, clean sound and accurate transcripts beside it; for them the outcome of the call adds little, and a transcript without its audio is of little use. Buyers of company records want the words joined to the deal, and can take transcripts alone. Most companies hold both: the decision is which buyers each part goes to, and whether the audio is worth the extra consent work that section 4 describes.

2. Who buys sales calls and sales emails

No buyer we found lists sales calls or sales emails as a kind of data it buys, in the way buyers list support tickets, code or CRM records. The public wording that comes closest is from the buyers of a company's whole working record, and from one buyer's listing for call centres. The table is every buyer whose own page names one of these parts of a sales archive, with what it says it pays. For the full list of buyers and their terms, see who buys AI training data.

Two things stand out. First, every one of these buys the conversations as part of something larger: HUMXN and Polyshares buy a company's record of work, with sales as one part beside tickets, code and documents; Mercor's call-centre listing is for a contact centre's archive of customer calls, and its AI product listing is for a product's usage data. None prices a sales call on its own. Second, the wording is about outcomes and links: HUMXN's phrase is the deals that were won and lost, which is the CRM, not the recording.

The clearest sign that sales calls sell inside a company's records is a purchase reported by a buyer-side referral network in October 2026: a lab buyer took a creator marketplace's corpus that included about 83K sales-call recordings beside its code, CRM, support tickets, Drive files and Slack messages. The calls were one part of a whole company's history, which is how this guide recommends offering them.

Who trains on sales conversations but does not buy them

The demand for models that sell is clearest at the companies building them, and none of them shows a way in for a company's archive. Gong's and Landbase's figures (section 1) describe what their own platforms recorded or ran. Liva AI, which delivers voice data to labs, says it collects voice and video entirely in house and makes no third-party purchases. Speech-to-text providers take customer audio by agreement rather than by purchase: Deepgram says the only customer data it uses in future model training is data contractually included in its Model Improvement Partnership Program. Approaching any of these companies is a cold offer, not an application to a programme, and a holder should expect to be asked why its calls are worth more than what the company already has.

Real phone conversation is wanted for voice models more broadly. Neon, a consumer app, pays people $0.15 a minute (reported) to record their own phone calls and sells the audio to AI developers. That is a supply of consented two-party calls built from scratch, and it shows the bar a company's archive has to clear: every voice on it there agreed to the sale.

3. What sales calls and sales emails are worth

There is no public price for a sales call with its outcome, per call or per hour, and none for a sales email. So a holder prices its sales conversations by the route it sells them through, and the two routes give very different numbers. The rates page has every public price behind the figures below.

As part of the company's records. Buyers' published offers for one company's records have a median of $280K per company, and the middle half of all public prices for a company's records runs $132K to $485K per company. Completed sales are fewer and lower, with a median of $72.5K per company across the handful on record, so read the offer figures as what buyers say before they have seen the data. Mercor's listings above sit in the same range, and the most a buyer has offered in public for one company's archive is $12.5M, in a bankruptcy auction. On this route, sales calls and emails raise the price of the whole record by adding outcome-labelled conversations to it; they are not priced separately.

As audio alone. Recorded calls sold for their speech meet a market of call-centre recordings that already exist in quantity. AIxBlock puts off-the-shelf call-centre audio at $5 to $25 per hour of audio, WiserBrand lists US customer-service calls at $25 an hour on a usage licence, and a German call-centre set on Opendatabay lists at $1,320 for 212 hours. Across all audio, asking prices off the shelf have a median of $119 per hour of audio and completed sales $73 per hour of audio, while speech recorded to a buyer's order costs $124 per hour of audio. The gap is the point: AIxBlock puts custom collection at 5 to 60 times the off-the-shelf price, and an archive of existing calls sells at the off-the-shelf end. Our speech data cost guide sets out the per-hour market in full.

What moves the price on either route

• Outcomes joined to each conversation. A call or thread linked to the meeting it booked or the deal it lost is the version the record buyers describe. Without the CRM join it is a transcript.
• Rarity. Less common languages and regulated domains sit at the top of AIxBlock's range ( scarcity beats volume), and FileYield, a listing site, says " multi-language call data is extremely valuable for conversational AI training". Enterprise sales in a specialist field (medical devices, industrial equipment, financial software) is rarer than consumer telesales.
• Two clean channels. Each speaker recorded separately is worth more to a speech buyer than one mixed track.
• Proof of consent. A lot where every call carries its recorded notice can go to more buyers than one where nobody can show it (section 4).
• Who removes personal data. In the Spirit Airlines auction, one bidder offered more if it could de-identify with its own tools, and the estate chose the lower bid with outside de-identification. Who de-identifies is a term with a price.

A worked example

An example, not a real holder. A software company with a mid-sized sales team holds four years of calls in a conversation intelligence tool, its outbound sequences in a sales engagement tool, and its pipeline in a CRM. Sold as audio, its calls would be priced by the hour at call-centre rates, its English is the most common language there is, and every prospect who joined from an all-party state would need proof of notice before the audio could go. Sold with its CRM history, tickets and code to buyers of company records, the same calls, transcribed, de-identified and joined to each deal's outcome, become one of the reasons the record is worth more than a record without them. The second route is usually the better one for a company like this, and the audio can be held back without losing what those buyers came for. The valuation gives a range for the second route from the company's own profile, with no upload.

4. Was each call lawfully recorded?

Before anything about AI, a recording has to have been lawfully made, because a defect in how it was made follows it into every sale. The federal Wiretap Act is a one-party rule: it is not unlawful for a person to record a call " where such person is a party to the communication or where one of the parties to the communication has given prior consent", unless the purpose is criminal or tortious. A company whose own rep placed and recorded the call is a party. But several states require every party's consent, and the same Act separately punishes anyone who intentionally discloses or uses the contents of a call knowing it was unlawfully recorded. Our audio guide has the state-by-state table, each row read in the statute's own text. For sales calls, four points from it matter most.

Prospects are everywhere, and mostly on mobile phones. A sales team calls people in every state, and California's rule for calls with a cellular or cordless phone on either end forbids recording " without the consent of all of the parties to a communication", with no exception for calls the parties might expect to be recorded. Most CRMs record where a prospect's company is, not where the person sat during the call, so the careful reading is to treat a call as falling under the strictest state any participant could have been in.

The notice has to be on the recording. Where all-party consent applies, a recorded announcement is usually what proves it. In Washington, " consent shall be considered obtained whenever one party has announced to all other parties" that the call is being recorded, if the announcement is itself recorded; Connecticut's civil statute accepts a verbal notice at the start or " an automatic tone warning device". A dialer that played a recorded notice on every call, and kept it on the recording, is the strongest evidence a holder can give. A policy saying reps were told to mention it is weaker.

Two states wrote rules for exactly these recordings. Illinois lets a business engaged in marketing or telephone solicitation record calls on one party's consent for quality control, staff training and internal research, but nothing obtained that way may be " divulged to any third party". Pennsylvania's exception for telemarketing and customer-service calls says " Any communications recorded pursuant to this paragraph may only be used by the business for the purpose of training or quality control", and requires them to be destroyed within a year. Licensing such recordings to an outside buyer is neither. Sales calls recorded under these exceptions stay out of a sale.

Consent to record is not consent to train. Every recording statute governs the act of recording; none we read says whether a lawful recording may later be licensed to train a model. Privacy law comes closer: California's rules require the consumer's consent before processing personal information for any purpose that people would not reasonably expect, and reusing a prospect's recorded voice to train someone else's model is the kind of use those rules have in view. This is why a buyer's lawyers ask whether calls and meetings were recorded with every participant's consent, and whether that consent covered use beyond the meeting, and why most sales calls sell as de-identified transcripts rather than as audio. Our meeting recordings guide applies the same rules to calls held in Zoom, Google Meet and Teams, and our employees guide covers the reps' own side of each call.

5. Whose conversations are they? Platforms that hold their customers' calls

A dialer, a voice platform, a conversation intelligence product or a sales engagement tool holds a great many calls and emails, almost all of them made by its customers' sales teams. Those conversations are its customers' records, held on their behalf, and the platform is not free to license them. This is the first question for any software company that thinks of its servers as an archive.

Under California law a platform holding customers' data is usually their service provider. A service provider holds personal information under a written contract that must prohibit it from " Selling or sharing the personal information" and from using it for other purposes, and the regulations list the purposes it may use that information for: improving the service it provides is one, and licensing it to an outside AI buyer is not. Under the GDPR it is a processor, which acts only on documented instructions from the controller; a processor that decides for itself to use the data for a purpose of its own shall be considered to be a controller in respect of that processing, and carries a controller's duties for it. Buyers check this before price: the rights check asks whether your customer contracts and data processing agreements allow using it for anything other than serving them.

What a platform can license is narrower but real:

• Its own sales team's calls and emails, under the same rules as any company's.
• Its customers' conversations where each customer agrees, in writing, for this use. Deepgram's programme is the model: it trains only on data contractually included by the customer, and a customer can opt each request out. A platform that wants to license its customers' calls needs the same: an opt-in in its contracts, a record of who opted in, and a share for the customers who did.
• Usage data about its own product, the logs of how its software and agents behaved, which Mercor's AI product listing names, once anything belonging to customers is removed from it.

The reverse check matters for every company. If your calls were transcribed through a provider with a training programme, read its terms and your settings: part of your audio may already be inside that provider's training rights, and a buyer will ask whether you are offering something another company has already used.

6. Calls placed by AI voice agents

More outbound sales calls are now placed by AI voice agents, software that dials, speaks and answers in a generated voice. Their recordings raise one question that human calls do not: whether the call itself was lawful. The FCC ruled in 2024 that the Telephone Consumer Protection Act's restrictions on an "artificial or prerecorded voice" cover AI-generated voices, so that such calls " require the prior express consent of the called party" on the lines the Act covers, absent an emergency or an exemption. The ruling governs how the calls were made, not whether their recordings may be licensed later; but a buyer's counsel can ask how that consent was obtained, and an archive of calls placed without it is one a careful buyer will not take.

Recording an AI agent's call adds a second question no statute we read answers: whether software counts as a party that can consent to the recording. The safe basis is the one that works everywhere, a recorded notice at the start of each call that the call is being recorded, which is what Washington and Connecticut accept as consent (section 4).

What such an archive is worth depends on which side of the call a buyer wants. The person's side is human speech and human objections, worth what any consented call is worth, under every rule above. The agent's side is generated speech, of no use to a buyer of human voices, but its decisions, joined to the outcome of each call, are the usage data that Mercor's listing for AI products asks for: " real usage data from an AI app, agent or developer tool", at $20K to $480K a company. A voice platform whose agents run on its customers' behalf is back in section 5: the calls are its customers'.

7. Sales emails and sequences

A company's outbound sales email is two things at once: its own writing (the templates, the sequences, the order and timing that worked) and a record of conversations with people who never became customers. The first is the company's to license. The second carries those people's names, addresses and words.

Export it yourself. Mail sent from Google Workspace or Microsoft 365 comes out through the administrator's own export: Google's takes a full export of an organization's data, and our email archive guide covers both systems, their deadlines and what each export holds. Sequences, steps, opens and replies sit in the sales engagement tool, which has its own export and its own terms. Read them for a clause on AI training before anything leaves: HubSpot, for one, bars an app using data accessed through the APIs to train, fine-tune, or improve AI or machine learning models, so the route out has to be the company's own export, never a buyer's connector. The HubSpot and Salesforce terms pages say what each allows.

The contact list stays out. Buyers of email for training are not buying lead lists, and the list is the riskiest part to license. Pangeanic, which sells business email corpora for AI training, says it does not provide prospect databases, address lists, marketing leads or data intended for unsolicited outreach, and its specification asks the buyer to choose the balance of inbound and outbound mail and whether replies are in. A holder delivers threads with their outcomes, without the addresses; our CRM guide explains why the contact records are removed from a CRM export too.

Prospects' personal data needs the same care as customers'. Prospects signed no contract with the company, but what they wrote is their personal data, and the privacy promises the company made when it collected it travel with it: the rights check asks that the policies in force when the data was collected allowed this kind of transfer, or affected people gave fresh affirmative consent, or the data is fully anonymised. A California business that licenses identifiable personal information must sign an agreement that specifies that the personal information is sold or disclosed by the business only for limited and specified purposes. In practice, de-identified threads are what sells.

Templates are the easy part. The company's own templates, with the reply and meeting rates each produced, carry no one's personal data and show exactly what worked. They are the least risky part of a sales archive and, joined to their outcomes, one of the most useful.

8. De-identifying calls and emails

Text can be de-identified; a voice is harder. In transcripts and emails, the names, phone numbers, email addresses, street addresses and account numbers of reps, prospects and customers come out, and so do the details that identify a prospect's business: its name, its deal size, its internal problems as described on the call. Buyers check for the second kind specifically, asking whether a holder can list which counterparties appear in what it offers. Our de-identification tool does the text work on your own computer, and our published standard sets out the rules it follows, each with its source.

Audio keeps the voice, which identifies the speaker even when every name is bleeped. Some laws treat a voiceprint as biometric data, and buyers who take call audio ask for consent from each identifiable speaker. Our meeting recordings guide covers voices, voiceprint law and what buyers accept. For most sales archives the practical answer is that transcripts are de-identified and sold, and audio goes only from calls where every voice can be shown to have agreed.

Decide early who does the work. A buyer that de-identifies with its own tools sees your data first; an independent provider, or your own team, keeps it in your hands until it is clean. Section 3 showed the choice has a price, and our guide on proving data quality without sharing it covers what a buyer can be shown before anything leaves.

9. Preparing a sales conversation lot

1. Inventory by system, with counts. List each system (dialer, meeting tool, conversation intelligence product, sales engagement tool, CRM), the years it covers, how many calls, hours and threads it holds, and how many reps produced them. Buyers price from a profile before any file moves: no buyer of company records states a sample size, and several price from a description alone.
2. Join each conversation to its outcome. Match calls and threads to CRM opportunities by account and date, and record the stage reached, the meeting booked and the result. This join is what turns a pile of calls into the record the buyers describe.
3. Sort calls by notice and by place. Separate calls with a recorded notice from calls without one, and calls with participants in all-party states from the rest. Take out calls recorded under the Illinois and Pennsylvania telemarketing exceptions (section 4).
4. Separate your conversations from your customers'. A platform licenses only its own team's calls, and its customers' only where each has agreed in writing (section 5).
5. Take the list out. Remove contact records, personal addresses and anything bought from a list vendor (section 7).
6. De-identify, then check. Remove people and counterparties from transcripts and emails, read a random sample by hand, and decide which audio, if any, can go (section 8).
7. Export before subscriptions end. Conversation tools and CRMs keep data only so long after a contract ends: Salesforce makes data available for 30 days after termination on request. A company that is closing should export first and decide later; our deletion countdown shows how long the common systems keep data.
8. Describe it. One document a buyer reads first: systems, years, counts, languages, how outcomes are joined, how consent was handled and how data was de-identified. Our datasheet builder writes it, and the rights check goes through the questions a buyer's lawyers will ask.
9. Offer it with the rest of the company's records. Sales conversations are worth most beside the CRM, tickets, code and documents they connect to. Our buyer match shows which buyers' published rules fit what you hold.

10. How a sale through AIData.Broker works

AIData.Broker represents companies selling their data to AI buyers. We act only for the holder, as its agent in the sale, for holders based in the United States, the United Kingdom and Canada. For a holder of sales calls and sales email, a sale through us runs like this.

The valuation comes first, with no upload. The valuation asks about your systems, years and people and returns a range from the public records in section 3. If you add files, they are measured on your own computer. The valuation binds no one.

You keep the data and deliver it yourself. We do not hold your recordings or your mail, show any part of them to a buyer, or license them in our name. You, or a de-identification provider you choose, deliver directly to the buyer, and the buyer pays you directly.

Every buyer that fits sees it at once. Every buyer whose published requirements fit your records receives a description that does not identify you; each that gives its confidentiality undertaking receives the same datasheet, with one deadline for offers. Offers come back on the same terms, so you can compare price, who de-identifies, payment timing and exclusivity side by side. Our free offer check reads any offer you already have against the published terms of the buyers we track.

The fee is due only out of money you receive. It is a success fee on what you are actually paid, band by band: 12% of the first $250,000, 10% up to $1 million, 8% up to $3 million and 6% above $3 million. On a $300,000 sale the fee is $35,000. Mandates are not exclusive: our fee is due only on a sale to a buyer we introduced, during the mandate and for a tail of 12 months after it, and we do not take mandates where the expected sale is below $23,000. How it works sets out every step and term.

Questions people ask

Can I sell my company's sales call recordings to AI companies? Yes, if the calls were lawfully recorded, they are your company's own calls, and the people on them are protected. No buyer lists sales calls as a category of their own, so they sell best as part of the company's records, de-identified and joined to the deals they were about. Audio goes only where every voice can be shown to have agreed (section 4).

Who buys sales call data for AI training? Buyers of whole company records: HUMXN names " pipelines, call notes, emails and quotes, with the deals that were won and lost", Polyshares names CRM history, email and call recordings, and Mercor licenses recorded customer conversations from call centres. The companies building sales AI, such as Gong, train on what their own platforms hold and show no door for buying an archive.

How much are sales call recordings worth? There is no public price for a sales call. As audio they meet call-centre recordings at $5 to $25 per hour of audio; as part of a company's records, buyers' published offers have a median of $280K per company a company. The outcome joined to each call is what makes them add to that figure.

Do I need consent from prospects to sell recorded sales calls? To record, federal law needs one party's consent, but several states need everyone's, including California for any call with a mobile phone on either end ( Penal Code 632.7). To license, recording law is silent, but privacy law may need fresh consent for a use people would not expect, and Illinois bars giving telemarketing recordings to a third party. De-identified transcripts avoid most of this; audio does not.

Can a voice AI or call platform sell its customers' calls? Not on its own say-so. A platform holding customers' calls is usually their service provider or processor, and California's rules list improving its own service, not licensing to an outside buyer, among its permitted uses. It needs each customer's written agreement for this use, or it licenses only its own team's calls and its product's usage data.

Can I sell our outbound sales emails and sequences? Yes, the company's own templates and sequences with their results are among the safest parts of a sales archive. The sent mail and replies carry prospects' personal data and come out de-identified, and the contact list stays out: email corpus sellers do not provide prospect databases or marketing leads. Export through your own administrator, never a buyer's connector.

Do buyers take calls made by AI voice agents? Some do, as usage data: Mercor's listing asks for " real usage data from an AI app, agent or developer tool". Expect to show that the calls were lawful, since the FCC treats an AI voice as an artificial voice that needs the prior express consent of the called party, and that each recording carries a notice.

What this means for you

1. Decide which route your calls take. For most companies the better route is with the CRM and the rest of the company's records, de-identified and joined to outcomes; audio alone sells at call-centre prices.
2. Check each call's consent before its value. Recorded notices, the states your prospects were in, and the Illinois and Pennsylvania telemarketing exceptions decide what can go at all.
3. If you are a platform, separate your own conversations from your customers'. Their calls need their written agreement; your own team's calls and your product's usage data do not.
4. Take out the contact list and de-identify the rest with our de-identification tool, on your own computer.
5. Get a range before any buyer sees anything. The valuation is free and needs no upload; our hub guide on selling company data to AI labs covers the whole sale, and what labs buy shows where sales conversations sit among the kinds of data in demand.

AIData.Broker acts for holders in a sale, never for the buyer: if you want several buyers to compete for your company's records, sales conversations included, start with the valuation.