How to sell Slack data for AI training

Why AI buyers want your Slack workspace

Most of the public internet has already been used to train models. What labs can’t find online is how a real team makes decisions: who asked for what, how a bug got triaged, how an outage got handled, how a launch slipped and got back on track. Slack holds years of exactly that.

The hottest use is training AI agents. Buyers build reinforcement learning gyms, simulated workplaces stocked with a real company’s Slack threads, emails and documents, where an agent practices workplace tasks and gets graded (Gizmodo). An agent that will work alongside a team needs to see how teams actually talk: short requests, half-finished threads, hand-offs and follow-ups.

Chat alone is worth less than chat that connects to other records. A Slack thread that references a Jira ticket, a pull request and a customer support case lets a buyer build a task with a start, the steps people took and an outcome. If your Slack links out to tickets and code, plan to offer those too.

What Slack data has sold for

Public numbers are few, but the deals reported so far give a range:

• The former CEO of Cielo24, a company that shut down, sold its Slack messages, internal email and Jira tickets as training data for “hundreds of thousands of dollars” (Gizmodo).
• SimpleClosure, which helps startups wind down, closed about 100 data deals in a year at $10,000 to $100,000 per company (SimpleClosure). Its AssetHub launched with workspace data, including Slack, email, Notion and Google Drive, in beta (Business Wire).
• Data vendors are approaching companies directly. Mercor has been chasing shuttered startups for their Slack and Jira archives (AI Weekly).

Most of these sellers were closing down, but you don’t need to be. An operating company can license a cleaned copy and keep using Slack as before. For a wider view of prices, routes and buyers, see how to sell company data to AI companies.

How to export Slack data, by plan

Slack’s own workspace export is the starting point. What it covers depends on your plan (Slack Help Center):

Three details matter for a sale:

• Exports contain file links, not files. Attachments stay in Slack. If a buyer wants documents, export them from where they live and clean them separately.
• The wider exports have a stated purpose. On Business+, Slack describes the self-serve tool for private channels and DMs as being for legal and regulatory compliance. On Free and Pro, owners have to apply to Slack. Ask a lawyer whether using that export to license data fits the terms you applied under.
• Public channels are usually enough. They hold most of the work buyers care about and carry the least privacy risk.

Why the export method matters: Slack’s API terms

In May 2025 Slack updated its API terms. Apps may no longer bulk export data through Slack’s APIs, keep persistent copies, archives or long-term stores of it, or use it in large language models (Hunton Andrews Kurth, Slack API Terms). Cloud Wars reported the change as a ban on bulk exports and on using API data for LLM training (Cloud Wars).

That rules out the easy-looking shortcut: a buyer installs an app in your workspace and pulls everything through the API. It also gives you a second reason to avoid it. A buyer with live access sees raw data before anything is removed.

The safer pattern is that your own admin runs Slack’s workspace export, the data gets cleaned, and buyers judge it from redacted samples. The API terms govern apps, but your customer agreement with Slack still applies to your workspace. Have a lawyer read the terms for your plan before you license anything.

What to remove before a buyer sees it

Buyers want to learn how work gets done, not who said what. Clean the export before any sample leaves your hands.

Direct messages and sensitive channels

Leave DMs out by default. Do the same for HR, legal, finance, health, hiring and any “random” or social channel where people talk about their lives. Keep the channels where work happens: engineering, support escalations, incidents, product and sales process.

People

Replace names, handles, emails and phone numbers with consistent placeholders, so “Person 14” stays the same person across threads. Then read whole threads, not just fields. A job title, a city and a date can identify someone even with the name gone.

Customers and partners

Chat is full of customer names, deal sizes and other companies’ confidential details, often covered by NDAs. Removing a name doesn’t fix a contract that forbids sharing, so carve those threads out entirely. Shared Slack Connect channels contain messages written by people at other companies; exclude them unless you have a clear right to share.

Secrets

Engineers paste API keys, passwords, tokens, database URLs and customer data into Slack all the time. Scan for them with a secrets detector and remove them. Rotate any live credential you find, whatever you decide about the sale.

Microsoft Teams and Gmail: the same idea, different tools

Teams. Admins export Teams chats through Microsoft Purview eDiscovery. eDiscovery (Premium) can rebuild threaded Teams conversations, so you can review and export them as whole conversations (Microsoft Learn).

Gmail and Google Workspace. A super admin can use the Data Export tool to export the organization’s data to Google Cloud Storage. The admin account must be at least 30 days old with 2-Step Verification on, and an export typically takes about 72 hours but can take up to 14 days (Google Workspace Admin Help). Google Vault can also search and export Gmail for specific accounts (Google Vault Help).

The same cleaning rules apply. Email has more outside parties than chat, so customer and partner carve-outs take longer. As with Slack, check the terms of any tool or API you use to pull the data.

Risks to weigh before you sell

Privacy advocates have already questioned these deals. In Forbes’s report on companies selling old Slack messages and email to AI labs, one said:

1. Your people. Employees wrote these messages without expecting them to train AI. Expect staff and the press to ask what was shared and why.
2. Rights you don’t have. A law firm’s review of these deals found most companies can’t sell everything a buyer asks for (Frankfurt Kurnit). Chat holds other companies’ secrets and vendor material.
3. Redaction that misses. Automated tools catch names and emails. They miss nicknames, inside jokes and unique details. A human review of a sample is worth the time.
4. Losing track of the data. Models trained on your data stay trained. Approve each buyer and use, set a deadline for deleting the raw copy, and limit resale.
5. Tool terms. Breaching Slack’s or another vendor’s terms could put your account at risk. Use the export path your plan allows.

This guide is not legal advice. Privacy, employment and contract rules vary by country and state, so get a lawyer to review your scope and the buyer’s contract.

With Smelted, identities come out first, buyers judge quality from redacted previews, and you approve every buyer. You’ll never be asked for passwords or access to your live systems. Licenses typically land between $20K and $5M, with payment in as little as 7 days after you sign. Our 60-second estimate gives you a range for your workspace.

Questions founders ask