A company can license its internal chat history, from Slack channels to Teams chats, to AI developers and the programs that supply them, but only for the messages it may export and only on the terms its chat tool allows. Slack's API Terms, effective October 10, 2025, bar apps distributed outside the organization from training a large language model on API Data, and no party has published a price for chat messages as such.
Chat is one data type within AI training data licensing, where a license is one outcome and keeping the archive is another. At least nineteen programs and intermediaries name chat in their published asks (checked November 9, 2026), and the largest chat figure on record is 500,000,000 Teams items inside one USD 10,000,000 successful bid for a whole dataset, not approved as of November 9, 2026.
Each program term, platform term and court figure below carries its check date, and the DataPacts Editorial Team explains how we source, date and check every figure.
What Chat Messages Are as AI Training Data
Chat messages as AI training data are a company's internal conversations, held as channels, threads, group chats and direct messages, licensed as a de-identified copy while the company keeps its own records.
Chat is common among the sellers that programs see. Mercor's Brendan Foody, who describes himself as its CEO, states that Slack appears at 55 % of more than 300 partner enterprises (LinkedIn post, September 2026, checked November 9, 2026), and SimpleClosure, a shutdown service, states that 95 % of the workspaces on its AssetHub include Slack (published September 3, 2026, checked November 9, 2026).
Programs ask for chat by tool name (Slack, Teams, Google Chat) or as "team chat", "internal chat" or "chat and messaging". Customer-facing chats belong to customer support tickets and chats, and the logs a chatbot provider keeps from its own users are a different subject.
What Makes a Chat Archive Worth Licensing
A chat archive is worth licensing when it shows real work over years: threads in which people ask, decide and hand over tasks, written by people rather than bots. Programs state their entry traits in four groups, each as the program's own claim (checked November 9, 2026).
- History: Nyne states 3+ years of operational workflows, and Collinear AI states "Several years of history, written by people".
- Team size: Scale AI prefers 40+ employees, Nyne asks for 20+ full-time U.S. W-2 employees, and The Corner asks for "substantial Slack or Teams history" from closed companies.
- Language: Collinear AI asks for English data, and The Corner states 70 %+ English.
- Connected tools: Mercor states "Connecting more tools generally results in a higher payout", and Centific states that compensation depends on the number of connected applications.
Thresholds differ by program and are entry rules, not offers. The thresholds that programs publish are compared under data sale readiness.
Programs and Intermediaries That Ask for Chat Data
At least nineteen programs and intermediaries name chat in their published asks; programs acquire a license, while brokers, referral pages, marketplaces and seller's agents handle the sale and never acquire the data. The table compares them alphabetically as of November 10, 2026, with each term taken from the program's own page, checked November 9, 2026.
| Name | Role | Chat sources named | One stated term relevant to chat | Source |
|---|---|---|---|---|
| Avelence | intermediary (broker) | "Slack/Teams" | de-identification on a separate copy; fee agreed separately | avelence.com |
| Centific | program | "email, calendar, Teams, Slack, Google Chat" | owner chooses applications, teams and periods | centific.com/enterprise-data |
| Collinear AI | program | "Slack" | one-time snapshot; seller keeps ownership | collinear.ai/data-acquisition |
| Data Licensing Advisors | intermediary (seller-side adviser) | "internal chat", "email and messages" | seller signs directly with the buyer; the adviser never takes custody | datalicensingadvisors.com |
| Handshake AI | program | Slack among sources named, "+1K more" | "All data is scoped, de-identified, and confidential" | joinhandshake.com/ai/data-partnerships |
| Labelbox Prism | program | "cross-system interaction data (Salesforce, Jira, Slack)" | exclusive by default; models already trained cannot be unlearned | prism.labelbox.com/faq |
| Mercor | program | "Slack", "MS Teams" among 100+ tools | no AI lab buyer or third-party vendor sees raw data | mercor.com/data |
| Mode | program | "chat" | "Mode buys the filtered copy" | data.mode.inc |
| Nyne | program | "chat (Slack, Teams, Google Chat)" | data used for training and evaluation only | nyne.ai/data-partnerships |
| Ooak Data | program | "Slack", "Teams" | raw data deleted within 30 days of delivery | ooakdata.com/partner |
| OpenTrain AI | intermediary (referral page) | "team chat" | never receives or holds data | opentrain.ai/company-data-licensing |
| Pangea.ai | intermediary (marketplace) | "chat and email" | non-exclusive license for a set use and term | pangea.ai/data-licensing/company-data |
| Polyshares | program | "Slack and Teams" | "Defined material for a defined term" on its home page vs "Perpetual exclusive license to the operational record" in its track record | polyshares.com |
| Scale AI | program | "chat (Slack, Teams, Google Chat)" | anonymizes inside the company's environment before extraction | scale.com/data-partnership |
| SimpleClosure | intermediary (shutdown service) | WorkspaceData (Slack) | license or outright sale; sensitive employee data excluded | simpleclosure.com/product/assethub |
| SuperAnnotate | program | "Slack threads, email chains, handover notes" | seller approves each buyer | superannotate.com/data-partnerships |
| The Corner | intermediary (micro1 referral page) | "substantial Slack or Teams history" (closed companies) | raw data kept at most 180 days | thecorner.ai/data/partnerships |
| Troveo | intermediary (marketplace) | "chat and messaging" | licenses on the seller's behalf; seller keeps ownership | troveo.ai/business-data |
| Turing (Project Lazarus) | seller's agent | "team communication (Slack, Teams)" | non-exclusive; Turing is the only seller for the first 90 days | lazarus.turing.com |
Program = acquires a license itself; intermediary and seller's agent = handle the sale, never acquire the data.
Polyshares shows two license wordings on its own pages, so a seller asks which applies. The wider field of programs, beyond those naming chat, is compared under company data partnership programs.
What a Chat Archive Is Worth: Claims and One Court Record
No party has published a price for chat messages as such; the figures on record are program-wide ranges stated by the programs themselves and one court record in which chat sits inside a larger dataset.
The ranges cover company data in general, not chat alone, and each is the program's claim (checked November 9, 2026): Scale AI states "$10K–$1M+ illustrative value per data partnership", Polyshares states "Typically $100K to $2M", Ooak Data states "Deals to date range from $20K, the smallest closed, to several hundred thousand dollars", and Labelbox Prism states deals "typically start around $50K".
The court record is the Spirit Airlines sale in Bankr. S.D.N.Y. 25-11897 (SHL). The Dkt. 1463 asset schedule lists 500,000,000 Teams items as included, inside the successful bid of USD 10,000,000 from Google LLC for the whole de-identified dataset, with an alternate bid of USD 7,500,000 from Mercor.io Corporation after the auction of August 14, 2026. The sale was not approved as of November 9, 2026, with a hearing set for November 17, 2026 (Dkt. 1779). The bid covers every record type in the schedule, so it carries no value per Teams item and none can be derived.
Estimators are indicative: Nyne labels its output "indicative and not an offer". Methods for pricing, and what the figures above can and cannot support, are set out under what company data is worth to AI companies.
Platform Terms That Decide How Chat Leaves Each Tool
The chat tool decides the route out: its export reaches only some message types, and its API or platform terms can bar training or resale through connectors. The table compares five tools as of November 10, 2026, from the vendors' own pages checked November 9, 2026.
| Chat tool | What the export reaches | What the vendor's terms say about training or resale | Source |
|---|---|---|---|
| Slack | private channels and DMs only on Business+ (approved application) and Enterprise (via Support); Free and Pro export public channels | apps distributed outside the organization "must not… use API Data to train a large language model"; bulk export only under an additional agreement; effective 2025-10-10 | slack.com |
| Microsoft Teams | Teams Export APIs, with an active Teams license for each user whose data is exported | no published term on AI training with exported Teams data on record here | learn.microsoft.com |
| Google Chat (Google Workspace) | Data Export tool for super admins | Workspace API user data policy: using user data to create, train or improve an AI model is prohibited, beyond a user's personalized model | developers.google.com |
| Discord | not on record | Terms s. 9 bar selling, licensing or otherwise commercializing content or data obtained from the services; terms effective 2025-09-29 | discord.com |
| app: per-chat export, view-once and disappearing messages excluded | Business Platform terms (last modified 2026-09-23): no sale or licensing of Platform Data, including anonymous, aggregate and derived forms (s. 4.1(b)); no AI training except fine-tuning for the business's exclusive use (s. 4.7) | facebook.com/legal |
The Teams row records an absence, not a permission. The WhatsApp rows separate the Business Platform, which carries the terms, from the app's per-chat export. Plan limits, retention and the mechanics of each route are covered under export rights by source system, tool by tool.
People in a Chat Archive: Staff, Guests and Customers
Every chat archive holds other people's words: staff in channels and direct messages, guests from other organizations, and customers quoted in threads.
The Irish Data Protection Commission states in its AI Insights Report (September 2026) that it is "unlikely that a controller will be able to demonstrate that data subjects expect the usage of their private messages or other private data for AI training". In the United States, 18 U.S.C. 2702 bars providers of public communication services from divulging stored content without lawful consent (2702(b)(3)), and a company's internal systems are generally not services offered to the public. GDPR Art. 88 (in force May 24, 2016, applying from May 25, 2018) lets member states and collective agreements set more specific employee-data rules with suitable safeguards.
The notice and consent steps are not repeated here and sit under employee notice and consent.
Scoping a Chat Export: Channels, Periods and Exclusions
Chat licenses are scoped by channel, team and period, and the programs that publish their process let the company exclude material before anything leaves. Five programs describe how, each as its own claim (checked November 9, 2026).
- Centific: the owner chooses applications, teams and periods, and exclusions are enforced in extraction code.
- Ooak Data: the seller chooses systems and date ranges, exclusions are annexed to the agreement, and raw data is deleted within 30 days of delivery.
- SimpleClosure: sensitive employee, compensation, health and personal information is excluded.
- Scale AI and Mercor: Scale AI anonymizes inside the company's environment before extraction, and Mercor states that no AI lab buyer or third-party vendor sees raw data.
The chat tool narrows scope first. Slack exports private channels and direct messages only on Business+ by approved application or on Enterprise through Support, so the plan already limits what a license can cover. What remains is cleaned under PII removal and de-identification of message bodies.
Licensing Chat Data: Benefits, Risks, Downsides and Alternatives
Licensing chat turns an archive the company already holds into license income, but the words belong to many people and a model trained on them cannot give them back.
- Benefits:
- Income from a record already held: Polyshares and Scale AI state ranges for company data (claims, checked November 9, 2026).
- Ownership stays with the company under a license: Collinear AI states the seller keeps ownership, and SuperAnnotate states "You're licensing rights for a set period, not selling the data".
- Scope can be limited by team, system or period (Centific, Ooak Data).
- Risks:
- Labelbox Prism states "Models already trained on the data cannot be unlearned".
- A connector can breach a chat tool's terms (Slack API Terms, Google Workspace API policy, WhatsApp Business Platform terms).
- Regulators doubt that people expect private messages to train AI (Irish DPC, September 2026).
- Downsides:
- Preparation and review of message bodies take time before any sample leaves.
- Payout is uncertain, because estimators are not offers.
- Alternatives:
- License one team or one period only.
- License through a seller's agent or intermediary instead of direct (Turing, Troveo).
- Keep the archive for the company's own use.
- Not licensing at all, covered under whether to sell or license your data.
No option is a default; the choice depends on the plan, the rights review and tolerance for an irreversible grant.
Where Chat Data Meets Email, Source Systems and the Seller Guide
Chat is one of several workplace records: email and meetings are separate data types, each chat tool has its own export page, and the seller guide sets the order of steps.
- Parent: every data type AI companies buy
- Email: licensing a company email archive
- Slack: Slack data and AI training
- Teams: Microsoft Teams data export
- Staff rules: employee data and AI training rules
- Seller guide: how to sell or license company data
- Meetings: selling meeting recordings and transcripts
- Google Chat exports: Google Workspace exports
This page is information, not legal advice.
Chat Archives From a Closed or Insolvent Company
Closed and insolvent companies license chat too, and the largest chat figure on record comes from a bankruptcy sale. That sale is the Spirit Airlines data archive sale, not approved as of November 9, 2026 (hearing November 17, 2026, Dkt. 1779). Labelbox Prism states that distressed and wound-down companies are eligible, Ooak Data states that wind-down deals are signed by a director, liquidator or administrator, and The Corner asks for closed companies with substantial Slack or Teams history (checked November 9, 2026).
The route for a company that has stopped trading is set out under selling data from a shut-down or defunct company.
Program Pages Named on This Page
Each program or intermediary in the table has its own profile, listed here alphabetically where one is published.
Avelence, Centific, Collinear AI, Data Licensing Advisors, Handshake AI, Labelbox Prism, Mercor, Mode, Nyne, Ooak Data, OpenTrain AI, Pangea.ai, Polyshares, Scale AI, SimpleClosure, SuperAnnotate, The Corner, Troveo, Turing (Project Lazarus).
Frequently Asked Questions
Can a company sell its Slack or Teams messages to AI companies?
Yes, within its export scope and its chat tool's terms, usually as a license of a de-identified copy rather than a sale. The plan sets what can be exported, and the tool's terms set what a connector may do.
What are a company's chat logs worth to an AI buyer?
No chat-only price is on record; the program ranges are claims about company data in general. Polyshares states "Typically $100K to $2M" and Scale AI states "$10K–$1M+ illustrative value per data partnership" (checked November 9, 2026).
Can direct messages be part of a chat data license?
Only where the export reaches them, and only after the staff steps. Slack reaches direct messages on Business+ and Enterprise, and the Irish DPC doubts that people expect private messages to train AI (September 2026).
Do AI companies train on chatbot conversation logs?
That is a different subject: an AI provider's use of its own users' chats, not a company licensing its internal chat. Why services train on their users' chats is covered under user data and platform terms.