Last checked: 7 October 2026 (buyer terms and news quoted on this page)
Chat history is where a company thinks out loud. That is why AI buyers ask for it, and why it needs more careful scoping than any other data type. This guide covers what buyers value, what to leave out, and how a deal handles export and privacy.
Finished documents show the result. Chat shows how the team got there.
AI labs are training models to do work, not only to answer questions. A policy document tells a model what a company decided. A Slack or Microsoft Teams thread shows the question that started it, the options people raised, the objection someone made, who decided and what happened next. That sequence is hard to find anywhere on the public web, and it is exactly what a model needs to learn how real teams operate.
micro1's data partnership page lists "decision-making patterns" among the things it wants, next to SOPs, project histories and QA processes. Slack and Microsoft Teams both appear in the system lists buyer programs publish.
There is also public precedent. Forbes reported on 16 April 2026 ("AI's New Training Data: Your Old Work Slacks And Emails") that shut-down startups were selling their Slack and email archives, and Fast Company and Gizmodo covered the same trend. Troveo cites roughly $10,000 to $100,000 per archive deal in that closure market. In August 2026, Google agreed to pay $10 million in Spirit Airlines' bankruptcy proceedings for internal data that included emails, Teams messages, spreadsheets and operations files, as reported by ABC, TIME and others; micro1 then made a $12.5 million rival offer, as reported.
Keep those numbers in context. They describe archives from closed companies and one airline in bankruptcy. They are not a price list for a running company's chat data, and nobody can price yours without seeing it.
Threads capture the trade-offs and rejected options that never reach a final document.
A message that points to a Jira ticket, a GitHub pull request or a Drive document ties separate records into one connected workflow.
Shorthand, corrections, handoffs and escalations, the way people actually talk at work.
Mode asks for several years of records the company owns. Years of channel history show how processes changed, not just a snapshot.
Buyers talk about "workflows". In chat, a workflow is a thread that starts with a problem and ends with a decision or a change.
Customer [ID-0193] says order [ORDER-7731] never arrived, but the carrier shows delivered. Second case this week from the same depot.
Pulling the scan history. Linked the support ticket here. Can someone check the dock assignment for Tuesday?
Scanned at dock 4 instead of dock 2. The new temp schedule swapped two loaders. Suggest we reship and fix the roster.
Approved: reship today. From now on, delivered-but-missing claims over the threshold need a photo check before we credit.
Updated the returns SOP and posted the link. Closing this thread.
Every name and number above is a consistent placeholder: Person A stays Person A across the whole archive, so a model can follow who did what without learning who they are. A buyer sees a trigger, an investigation, a root cause, a decision with an owner and a documented process change. Thousands of threads like this, linked to tickets and SOPs, are what turn scattered messages into a training resource.
The strongest candidates: most of the process value, the least personal chatter.
Case by case. Many hold leadership, HR or deal talk. Review each by name.
Exclude by default. They hold the most personal content and the least structured work. Include only with a narrow written scope and employee notice.
Channels shared with clients, vendors or partners contain other companies' words. Leave them out unless those companies agree in writing.
You decide the scope. In most deals the buyer runs the export and the de-identification, under a written agreement.
List workspaces or Teams tenants, channels, date ranges and approximate message counts. This manifest describes the data without containing it.
Name the channels, the years and the exclusions. micro1 states on its page that scope is agreed in writing. Whatever program you use, insist on the same.
Practitioners advise never sending a full dataset before there is a price. A manifest and a small de-identified sample are enough to get more than one offer.
An admin grants access for the agreed scope. Export features and retention differ by plan, so confirm what history actually exists before you promise years of it. Mode describes buying "an agreed copy", with originals staying with the company.
Mode says it de-identifies before onward delivery. micro1 says sensitive and confidential information is scrubbed and originals are deleted after processing. Then comes acceptance against the agreed criteria, and payment. Practitioners cite 60 to 90 days from first contact to close.
The exclusion list protects people and contracts. Write it before the first sample leaves the building.
General information, not legal advice. Talk to your own lawyer before you sign.
Chat is the most personal business record a company holds. Your staff wrote these messages to colleagues, not to an AI lab, and even with names replaced a thread can reveal who was sick or who complained about a client. The employee question deserves its own decision, separate from the legal one. Our page on employees and selling company data covers notice, consent and how to talk about it with staff.
The laws to raise with your lawyer by name: GDPR if any employees or customers are in the EU or UK, CCPA/CPRA for California residents (including employees), HIPAA if health information passes through chat in a healthcare setting, and any confidentiality duties you owe clients. In the EU, works councils may also have a say.
| What appears in chat | Why it matters | Typical handling |
|---|---|---|
| Employee names and @mentions | Identifies staff and links them to opinions and mistakes | Consistent pseudonyms (Person A, Person B) across the whole archive |
| Customer names, emails, phone numbers, order IDs | Personal data of people who never agreed to anything | Redaction or placeholders; whole channels excluded where it is dense |
| Health, family and personal-life remarks | Sensitive categories under several privacy laws | Exclude channels; redact remaining mentions; review samples by hand |
| Client confidential details | Your contracts with clients may forbid any disclosure | Exclude client channels; check each NDA before scoping |
| Credentials and secrets | A security risk, not just a privacy one | Secret scanning before export; rotate anything found |
Described in general terms. Exact menus, plan features and admin roles change, so confirm each point with your own admin.
Use this to describe your data honestly in a manifest. It is not a pricing tool.
| Signal | Stronger | Weaker |
|---|---|---|
| Threading | Discussions happen in threads with a clear start and end | Everything flat in the main channel, topics interleaved |
| Links | Messages link to tickets, pull requests and documents | No references to other systems |
| Channel purpose | One purpose per channel, consistent naming | Most work in a few catch-all channels |
| Time depth | Several continuous years in the same channels | Gaps after a migration or a retention purge |
| Language | Primarily English, which micro1 lists in its eligibility | Mixed languages without a clear majority |
| Human share | Mostly people writing to people | Mostly bot alerts and automated posts |
Each one is easy to avoid before the first sample leaves, and hard to fix after.
A scope nobody can review is a scope nobody can defend. Name channels and years, and say what is out.
Offering eight years when retention settings kept two damages trust at the first sample. Check first.
A sample should be small and already de-identified. A raw sample is a disclosure, not a preview.
Alerts and automated posts inflate message counts and lower the value per message a buyer sees.
Other companies' words are not yours to license. Filter by member type, not only by channel name.
Employees who learn from the news that their messages were sold lose trust fast. Decide on notice before export.
Periods matter as much as channels. Even inside a good channel, some stretches of time are better left out: a restructuring or layoff period, an incident that is still under legal review, the months around an acquisition. Excluding a date range is a normal scope line, and buyers expect to see it in a careful manifest.
Only what each program publishes. Ranges cover whole data packages, not chat alone.
| Program | Published payout | Published eligibility | Relevant to chat data |
|---|---|---|---|
| micro1 Enterprise Data Partnership | "$100k+ qualified", "$500k+ large-scale", "$1M+ highly unique"; referral page "$100K-$2M+ for approved data packages" | 30+ employees, mature operations, documented processes, modern software tools, primarily English; US prioritized, then other Western markets | Names "decision-making patterns" and project histories among what it wants |
| Mode company data | "$100K-$5M" | 20+ full-time US office employees (accounting firms 10+, law firms 6+); several years of records the company owns | Buys "an agreed copy"; originals stay with the company; de-identifies before onward delivery |
| Grepped | "$20K-$5M", "get paid in 7 days" | Any vertical; also pays individual professionals for expertise | Broad intake; confirm scope and timing in your own agreement |
Last checked: 7 October 2026. Sources: micro1.ai/data-partnerships and micro1.ai/company-referral; data.mode.inc; grepped.ai, each as published. Published ranges are not offers, averages or promises.
Independent site. Some links are referral links: if your company signs with a buyer through them, the buyer may pay us a fee. You are not charged, and we never see your data.
Not sure which program fits your headcount and country? Run the eligibility checker or read the full buyer programs comparison.
Work through this list with whoever administers Slack or Teams, and with your lawyer.
Some buyer programs list Slack and Microsoft Teams among the systems they draw data from, and a shut-down startup market for Slack archives was reported in 2026. Whether your company can sell depends on who owns the content, what your client contracts and employee policies say, and which channels you put in scope. Read the legal pages and talk to your own lawyer before you sign.
Most sellers are better off excluding direct messages by default. They hold the most personal content and the least process value. If a buyer asks for them, agree a narrow written scope and give employees notice first.
Nobody can price yours without seeing it. Troveo cites roughly $10,000 to $100,000 per archive deal in the closure market for shut-down startups. Buyer programs publish wider ranges for whole data packages, for example micro1 lists $100K-$2M+ for approved data packages on its referral page. Those are published ranges, not offers or averages.
The answer depends on your jurisdiction, your employment contracts and your internal policies. GDPR applies to EU and UK employees, and CCPA/CPRA covers California residents, including employees. Even where consent is not legally required, notice is the honest minimum. This is general information, not legal advice.
The scoping logic is the same: team channels first, private chats last, shared channels with other companies out unless they agree. Teams often links closely to SharePoint files and meeting recordings, so decide early whether those are in or out.
Mode asks for several years of records the company owns, as published. The other programs we track do not publish a fixed number of years. Longer, continuous history in the same channels usually shows more about how work changes, but only the history that survived your retention settings counts.
Their messages usually remain in the workspace and get the same de-identification as everyone else's. Notice is harder because former staff are not easy to reach, which is one more reason to keep DMs and personal channels out. Ask your lawyer what your policies and local law require.