Are old company emails worth anything to AI companies?

Often, yes, when the archive holds years of answered threads: requests, quotes and approvals. Less so when it is mostly newsletters and alerts, or old mail is gone or stuck on staff PCs. As of October 2026, 17 of the 25 providers Data Licensing Report tracks name email.

What do AI companies want old business email for?

To train and test models on real work, written by the people who did it. Mercor says “Your internal data captures how organizations actually operate, which is exactly what labs need to build AI that can handle real enterprise tasks”, and describes what it takes from Gmail as “Email messages and metadata”. Replay’s page for AI buyers shows a sample license whose permitted use is “training, evaluation and model development”: teaching a model, and checking its work against real cases (both as of October 2026).

Email suits that because each message was written at the time, dated and addressed. Why would AI companies want our emails? traces one job through a mailbox.

What makes an email archive worth more or less?

Replies, attachments, years and links to other records, in providers’ own words (as of October 2026):

What counts for more	How a provider puts it
Replies kept with the message they answer	Nyne: “Email threads and event details, senders, recipients, invitees, and the responses that connect them”
Attachments kept with the thread	Telegraph Lab (affiliated with this site) lists “Messages · Email threads · Attachments” under work conversations
Years of continuous history	License My Data: “Years of real correspondence and records”
Links to other records	Scrimdata: “Interconnection matters more than volume”

Three things lower what a mailbox holds without anyone noticing:

• Mail nobody answered. Newsletters, alerts and receipts add gigabytes without a single decision in them.
• One side of the conversation. Under POP, Microsoft explains, messages are deleted from the email service once downloaded, and “Sent mail is stored locally on your PC or Mac, not on the email server”. Years of POP can leave history split across PCs, or gone with a replaced laptop.
• Missing years. A gap at a change of system shows as soon as messages are counted by month; Does my company’s data qualify? explains how to trace a migration gap.

We deleted a lot of old email. Is what’s left still worth selling?

Usually, because deleted mail rarely disappears from every place at once. A message one person deleted normally still sits with whoever sent or received it. Scale AI looks for “Years of records in the tools your teams already run on”, so what matters is how many years of threads survive across the company, not whether every mailbox is complete. Check four places before writing anything off:

1. Colleagues’ mailboxes. Every internal message, and every outside thread with a colleague on copy, exists more than once.
2. Outlook archive files. Where classic Outlook’s AutoArchive ran, it moved Inbox items older than six months, by default, into an archive file in the PC’s Documents\Outlook Files folder (Microsoft). It can be set to delete old items instead, so open the file before counting on it.
3. The recovery window. Google and Microsoft keep deleted mail briefly, and longer under a retention rule or hold; the Google Workspace and Microsoft 365 pages give the days.
4. Old servers and backups. A mail server the company once ran, or its backup drives, may hold the years before the move to the cloud.

Former employees’ mailboxes follow the same pattern: a deleted account’s own copy is usually gone within weeks, but its mail with people still at the company survives in their mailboxes.

How do we export an email archive to sell?

Not all at once. A buyer starts from a description and a small sample, and the delivery export follows the agreed scope (how offers work). Early on, know where each part lives and what file it comes out as:

Where the mail is	How it comes out	File you get
Google Workspace	Takeout for one person; the Data Export tool or Vault for many (Google Workspace page)	.mbox; Vault also offers .pst
Microsoft 365	Purview eDiscovery for many mailboxes, or Outlook’s own export for one (Microsoft 365 page)	.pst or .msg
Outlook data files on PCs	Copy the files from each PC, including any archive.pst	.pst
Apple Mail, for any account it holds	Select the mailboxes, then Mailbox > Export Mailbox; Apple says it “exports the mailboxes as .mbox packages”	.mbox

The free sample tool reads .mbox and .eml files; a .pst needs converting first, as the Microsoft 365 page explains. Decide what stays out before any export leaves the building: the mailboxes, domains and date ranges you name become the scope (leaving out records).

Which buyers say they want email?

Data Licensing Report lists a provider here only if its own pages name email or a mail system such as Gmail or Outlook, and checks every provider against that rule; 17 of the 25 it tracks pass, as of October 2026. Avelence, License My Data, Miro Advisory and Sell My Business Data introduce companies to buyers, and FileYield lists data for sale; none of those five buys data itself.

Provider	What it is	What its own pages say about email
Avelence	Introducer; does not buy data	Gmail and Outlook, for “Email threads, linked replies, and meeting history”
Corpus	Licenses directly	“email and chat threads” first in what it licenses
FileYield	Listing marketplace; buys nothing	Email metadata in its communications category
Handshake AI	Licenses directly	Gmail and Outlook among example sources
idler	Licenses directly	Its CorpLaw benchmark, built from a law firm’s anonymized data, includes “negotiation email threads” from the firm’s record
License My Data	Introducer	“Email, calendar, Slack and Teams” under workplace records
Mercor	Licenses directly	Gmail and Outlook among the tools it connects
Miro Advisory	Introducer; does not buy data	Email among communication systems
Nyne	Licenses directly	Gmail and Outlook under email and calendar
Polyshares	Licenses directly	Email in “the working record of the business”
Replay	Licenses directly	Gmail among the sources of an example dataset on its AI-buyers page
Scale AI	Licenses directly	Gmail and Outlook under email and calendar
Scrimdata	Licenses directly	“inboxes” first in its list
Sell My Business Data	Introducer	Emails first in its FAQ list of records that “may be valuable”
SimpleClosure (AssetHub)	Wind-down platform; said on September 3, 2026 it also takes operating companies	Email among its connected platforms
Telegraph Lab	Licenses directly	Email threads and attachments under work conversations, naming Gmail
Troveo	Licensing agent	“Email threads with senders, recipients, and linked responses”

How each of them pays, and on what terms, is set out in the buyers directory.

Whose information is in our email, and what has to stay out?

Everyone the company ever wrote to, and whatever they attached. Outsiders write into a mailbox freely, and Miro Advisory says of communication systems that “Privacy, employee rights and third-party communications require careful review here”. Corpus asks sellers to keep out items such as “Government-issued identification numbers” and HR files, and says its anonymization covers “body text, headers, attachments and transcripts”.

Where those items hide in a mailbox:

• Attachments. Vendors’ Form W-9s, which exist to provide a taxpayer identification number (IRS), payroll reports, signed contracts and scanned IDs. A tool that cleans message text does not necessarily reach inside a PDF, so ask each provider whether attachments are redacted, dropped or reviewed. The sample tool counts attachments by type but never opens them.
• Customers and suppliers. Every outside thread names people who agreed to nothing. Whether privacy law treats them as consumers or as business contacts changes which rules apply; Is it legal to sell company emails and records? sets out the Texas and California rules and what you may have promised in your privacy policy.
• Your lawyers. Mail with outside counsel may be privileged, and Frankfurt Kurnit warns that a sale “may waive privilege over legal communications” (October 1, 2026). Excluding counsel’s email domain takes those threads out whole.
• Other companies’ confidential material. Drawings, pricing and plans sent to you under an NDA or a contract’s confidentiality clause are governed by that document, whoever owns the mailbox.
• Employees’ personal lives. Medical appointments, family matters and job searches turn up in work mail; what “anonymized” means covers how providers handle them.

Ask your own lawyer which of these apply to your archive, and whether anyone must be told or asked first.

What could our email be worth?

No provider publishes a price for email alone, as of October 2026. The reported sales that included email bundled it with other records from companies that were closing. Bloomberg Law reported on August 17, 2026 that the Spirit Airlines package held about 100 million emails alongside Teams chats and operational records, with Google named winning bidder at $10 million (Bloomberg Law); as of late September 2026 the sale still needed court approval (Moneywise). Forbes reported on April 16, 2026 that cielo24, a transcription company that was winding down, received “hundreds of thousands of dollars” for a 13-year record of Slack messages, Jira tickets and email (Forbes).

Neither is a price for an operating company’s mail; How much is my company’s data worth? sets the published ranges side by side. Have three email figures ready: years covered, number of mailboxes, and threads with replies.