AI training-data buyers license the documents a company writes about its own work: SOPs, policies, manuals, playbooks and wikis. Names come out first. We tell you which buyers take them and on what published terms.
Advertised by each company itself. Whole-dataset figures, not per page, and none is an offer.
Anything the company wrote about its own processes is a candidate. The trouble starts with material written about people, for clients, or by someone else.
Receiving procedure
Step 3If the count is short by more than 2%, phone supplier contact before signing the delivery note.
Version history
v4Short-count threshold set at 2%editorv3Photo of damaged pallets requirededitorv2Supplier call moved before sign-offeditorv1First drafteditorInvented example. The black bars stand for names removed; each says what kind of detail sat there.
HR case files and reviews, which are about identifiable people. Client deliverables, which the services contract usually assigns to the client. Vendor manuals, paid courses and purchased templates, which someone else owns.
Every buyer page we read says it de-identifies. Do your own pass first anyway: you know which names matter. The redaction guide
Notion, Confluence, Google Drive and SharePoint each export differently. Count pages per space first, and export once a buyer has scoped what it wants.
The figures are for whole datasets. Labelbox Prism leans to reasoning traces, support logs and robot telemetry rather than documents, so it is not among them.
$100K to $3M+
Names SOPs, process docs and knowledge bases outright. Wants 30+ full-time staff.
$10K to $1M+
Files with metadata, Jira and Confluence. Anonymizes on your systems. Range is illustrative.
$100K+
More records than documents. US firms, 20+ W-2 staff, three years trading.
Not published
Workspaces of companies closing down: Slack, Drive, mail, Notion, Figma.
Disclosure: some of these buyers pay a referral fee when an introduction of ours ends in a signed license. Companies pay us nothing, and no fee changes what these cards say. Independent: none of these companies owns, funds or endorses this site.
Files move only after a signed agreement, and only the ones it names.
Types, rough counts and the tools that hold them. No files at this stage.
The buyer asks for a small cleaned sample, or reviews one inside your environment.
Scope, rights, acceptance criteria and price go into a written license your counsel reads.
You go first, the buyer second. At Scale AI nothing ships until you approve the package.
Only micro1 states a window: 30 days from acceptance. The others leave timing to the contract.
The route depends on whether the company will exist after the deal.
Operating company
Winding down
What has been reported. Spirit Airlines' business data, including staff email, Teams chats, spreadsheets and calendars, drew a $10 million bid from Google against $7.5 million from Mercor, with no court approval as of Reuters' August 2026 report. At the small end, MIT Technology Review described a closed startup whose code fetched about $25,000.
Most companies should expect less than the headline ranges. The document value check lists the advertised ranges your headcount and years qualify for.
They apply before any price is discussed.
Yes, when the company wrote them or holds the right to license them. micro1 names SOPs, process documentation and internal knowledge bases as material it wants, and Scale AI lists files and documents that carry metadata. Procedures written for a client, or lifted from a vendor, are not yours to license. See which documents sell.
SOPs and work instructions, runbooks, internal wikis, company policies and project documents are the types buyer pages name or come closest to. HR case files, client deliverables and purchased third-party material stay out of the inventory.
No buyer we read publishes one. The published figures (micro1 $100K to $3M+, Scale AI an illustrative $10K to $1M+, Nyne commonly $100K or more) are per partnership. Our document value check shows which of those published ranges your company's size even qualifies for.
No. Scoping runs on a description: kinds of document, rough counts, which tools hold them, how much personal detail they carry. Scale AI goes further and does its anonymizing on your own systems, signing off each package with you. At the first stage nobody needs a single file.
Yes, and it is the clearest case for documents. SimpleClosure's AssetHub takes workspace data from Google Drive, Notion and Slack for founders winding down. Pay for workspace data is not published.
It can, if the documents are about people. A procedure about receiving freight is the company’s; a performance review is about an employee. Keep HR case files out, strip names from the rest, and have counsel check any confidentiality promises in your handbook.
You do, on every page we read that addresses it. micro1 states that the underlying data stays yours. Nyne leaves ownership and the rights you grant to the signed agreement. A license hands over use, so the scope of that use is the clause your lawyer should read hardest.
Tell us what kind of documents the company holds and roughly how many. We check which buyers’ published rules the collection meets, then tell you who they are.
Note: we make introductions, and the buyer at the other end may pay us for them.
Keep documents, exports, share links and samples to yourself. We work from your description alone.