Selling data
How to sell email archives for AI training
By Gareth HoyleLast updated 5 minute readHow we research
In short
You can license company email archives to AI companies, and Mercor, Polyshares, SellMyData and Troveo all list email among the systems they take. The best-known price is Google's $10 million winning bid for Spirit Airlines' de-identified business data, which included corporate email. Value comes from long threads with outcomes, and the main risks are outside senders' data, legal privilege and client NDAs.
Key takeaways
- Google won Spirit Airlines' archive of about 100 million emails and 500 million Teams messages with a $10 million bid, per MIT Technology Review.
- No buyer publishes a price per mailbox, and AIData.Broker puts the median public sale at $31,600 per company, as reported by ddiy.co.
- Google Vault exports Gmail as MBOX or PST, and Microsoft Purview exports mailboxes as PST or MSG files.
- Leave out mail with your lawyers, HR mail, card data and client mail covered by NDAs.
- Headers, signatures and quoted replies repeat identifiers, so de-identification has to cover all three.
Email is the longest-running record most companies hold. A thread that starts with a customer request and ends with an approved quote, a shipped order or a resolved complaint shows a complete piece of work, which is what AI labs want for training agents. The Spirit Airlines sale showed the scale buyers will consider: MIT Technology Review reported an archive of about 100 million emails.
Who buys email archives for AI training?
Email archives are bought by training-data companies, direct buyers, brokers and licensing agents, usually as part of a wider package of operational records.
| Buyer | What it lists | How it pays |
|---|---|---|
| Mercor | Gmail among more than 100 supported tools | Wire transfer after delivery, typically 2 to 4 weeks after signing |
| Polyshares | Email in the operational record | One fixed payment, net 30 to 60 days after approval |
| SellMyData | Messages from Slack, Teams and Gmail | Commission when a sale closes, target payment within 7 days |
| Troveo | Chat, email, documents and other business systems | Revenue share on every sale, no fees |
| SimpleClosure AssetHub | Email within startup workspace data | A percentage on a completed sale or license |
See our profiles of Troveo and SellMyData for exclusivity and fee terms, which differ more than the headline payouts suggest.
How much is a company email archive worth?
A company email archive has no published price per mailbox or per message. The public figures come from whole-company deals.
- $10 million: Google's winning August 2026 bid for Spirit Airlines' de-identified business data, including corporate email (MIT Technology Review)
- $12.5 million: the competing proposal micro1 then made for the same data (Fortune, September 2026)
- About $15,000, pending at the time of reporting, for one closed startup's workspace data (MIT Technology Review, October 2026)
- $280,000 and $31,600: AIData.Broker's medians for buyers' published offers and for public sales per company (ddiy.co)
The gap between published offers and actual sales is the useful number here. Treat buyer headlines as ceilings, and see the AI data deals tracker for the full record of reported prices.
What makes an email archive more valuable?
An email archive is more valuable when it covers many roles over many years and the threads show what happened in the end.
| Factor | Worth more | Worth less |
|---|---|---|
| Mailboxes | Operations, sales, finance, engineering and support roles | A handful of executives |
| History | Several years, with no retention gaps | The last 12 months |
| Threads | Complete chains with attachments kept | Single messages with attachments stripped |
| Outcome | The thread shows the approval, delivery or fix | No visible result |
| Noise | Newsletters, alerts and notifications filtered out | Mostly automated mail |
How do you export email from Google Workspace and Microsoft 365?
You export email from Google Workspace with Google Vault and from Microsoft 365 with Microsoft Purview eDiscovery. Both produce a fixed copy you control.
| System | Tool | Formats | Useful options |
|---|---|---|---|
| Google Workspace | Google Vault | PST or MBOX for Gmail | Include Drive files linked from messages |
| Microsoft 365 | Microsoft Purview eDiscovery | PST or MSG files | Include or condense the original folder path |
Export by date range and mailbox, record a hash of each file and keep the export offline. Do not give a buyer's tool access to live mailboxes, a point Frankfurt Kurnit Klein and Selz makes for all company systems.
Which emails should you leave out?
The emails you should leave out are the ones where someone else's rights, a legal duty or a security rule outweighs any training value.
| Category | Why it is a problem | Usual approach |
|---|---|---|
| Mail with your outside or in-house lawyers | Disclosure to a third party can raise privilege waiver questions | Exclude |
| HR, payroll and benefits mail | Sensitive employee data with little training value | Exclude |
| Client mail covered by NDAs | Other companies' confidential information | Exclude or get written consent |
| Card numbers and security codes | PCI DSS bars storing sensitive authentication data after authorization | Exclude the messages |
| Personal mail sent from work accounts | Employees' private lives | Exclude |
| Newsletters, alerts and spam | Noise that inflates counts | Filter out |
What rights issues come with email?
Email raises more third-party rights than most data sources, because half of every thread was written by someone outside your company. Frankfurt Kurnit Klein and Selz notes that email often holds other companies' confidential information under NDAs and commercial agreements.
- Customers and suppliers: their names and contact details are personal information, and the CCPA treats a paid transfer of personal information as a sale unless de-identified to its standard, per Frankfurt Kurnit Klein and Selz
- Employees: messages about pay and working conditions may be protected under Section 7 of the National Labor Relations Act
- Privacy promises: FTC staff warned in February 2024 that quietly making data practices more permissive after the fact may be unfair or deceptive
- Professional firms: CPA firms need clients' specific consent under the AICPA Confidential Client Information Rule, according to the Journal of Accountancy in November 2026
Law firms carry extra duties under state rules of professional conduct, covered on our page for law firms. We are a broker, not a law firm, so take independent advice on your own archive.
How do you de-identify an email archive?
De-identifying an email archive means treating headers, bodies and attachments as one problem, because the same person appears in all three.
- Replace addresses and display names in the From, To, CC and BCC fields with stable pseudonyms
- Apply the same pseudonyms inside message bodies and quoted reply chains
- Remove signatures, phone numbers, legal disclaimers and tracking links
- Strip routing headers that carry IP addresses and internal hostnames
- Redact attachments, including scanned PDFs, or leave them out
Quoted replies are where redaction fails most often, because one thread can repeat the same signature a dozen times. Test on a random sample of whole threads, not single messages. Our de-identification guide explains how to document the method.
If you hold UK or EU data
Email to and from people in the UK or EU is personal data under UK GDPR and EU GDPR, including the details of outside senders who never dealt with you directly. You need a lawful basis to disclose it, or a level of anonymization the ICO would accept, and pseudonymized mail stays personal data.
How do you start selling an email archive?
You start by listing the mailboxes, years and roles you hold, and the categories you will exclude, before anything is exported. That list is the first draft of your manifest and the basis for any price.
We do not take custody of your raw data. Buyers you approve see a redacted sample under a mutual NDA, and full data moves only after you sign a license directly with them. Book a free consultation to find out whether your archive is likely to interest buyers.
Frequently asked questions
Can a company sell its employees' work email for AI training?
Do outside senders need to be told their emails are being licensed?
Is one executive's mailbox worth selling?
Do buyers want email in MBOX or PST format?
Sources
- AI labs are buying data from dead startups to train their models, MIT Technology Review, accessed November 2026
- micro1 tries to beat Google's bid to buy bankrupt Spirit Airlines' data, Fortune, accessed November 2026
- Spirit's privacy ombudsman backs Google's $10m data deal, The Next Web, accessed November 2026
- How much is company data worth to AI? (2026 prices), ddiy.co (citing AIData.Broker), accessed November 2026
- Data monetization, Mercor, accessed November 2026
- Polyshares homepage, Polyshares, accessed November 2026
- SellMyData homepage and FAQ, SellMyData, accessed November 2026
- Business data, Troveo, accessed November 2026
- AssetHub, SimpleClosure, accessed November 2026
- Export data from Vault, Google Vault Help, accessed November 2026
- Export eDiscovery search results, Microsoft Learn, accessed November 2026
- Selling company data for AI training: what to know before you say yes, Frankfurt Kurnit Klein and Selz, accessed November 2026
- AI (and other) companies: quietly changing your terms of service could be unfair or deceptive, Federal Trade Commission, accessed November 2026
- Why is storage of sensitive authentication data after authorization not permitted?, PCI Security Standards Council, accessed November 2026
- Asked to sell firm data for AI training? Be very cautious, Journal of Accountancy (AICPA), accessed November 2026