Are old or outdated SOPs still worth including?

Often, yes. Revision history shows how a process changed, and micro1’s page counts several years of records in a company’s favor. Mark superseded versions so a buyer can tell them apart.

Do we need a minimum number of documents?

No page we read sets a document count. Every published threshold is about the company instead. micro1 wants a full-time staff of 30 or more, Scale AI would rather start at 40, and Nyne asks for a US W-2 payroll of 20 plus three years of trading.

Can we license documents written in Spanish or French?

Possibly, but micro1 says data primarily in English is what it prioritizes. Other buyers do not publish a language rule. Note the language split when you describe the collection.

What about documents built from a paid template?

The parts you wrote are yours; the template text may still belong to its publisher. Check the template license or rewrite those sections before counting them.

Is a Confluence space better than a folder of Word files?

Usually. Linked pages, labels, owners and dates are metadata, and Scale AI lists documents and files with metadata and Confluence knowledge records among what it takes. Loose files can still be licensed; they simply carry less context.

Which company documents sell to AI buyers?

Documents the company wrote about how it works: SOPs, runbooks, policies, manuals, playbooks and wiki pages. Buyers list these by name. Files about individual people, work owned by clients and anything written by a third party stay out.

Buyer lists checked November 7, 2026

Document types, one row each

"Who wants it" names only buyers whose own page lists that kind of material. Where no page names it, we say so.

Company document types, read against buyer pages on November 7, 2026
Type	Example	Who wants it	Personal-data risk	Usually licensable	Why
SOPs and work instructions	Month-end close checklist; warehouse receiving procedure	micro1 (lists SOPs and process documentation)	Low	Yes	Written in-house about how the work is done. Names usually appear only as owners or approvers.
Runbooks and process docs	Incident runbook; new-hire IT setup steps	micro1 (process documentation); Scale AI (documents and files with metadata)	Low	Yes	Step-by-step operational text. Strip hostnames, internal URLs and any credentials first.
Internal wiki or knowledge base	A Confluence space; a Notion team workspace	micro1 (internal knowledge bases); Scale AI (Jira and Confluence knowledge records)	Medium	Yes	Valuable because pages link to each other, but comments, @mentions and page history carry names.
Company policies	Expense, travel, security and acceptable-use policies	Scale AI (documents and files with metadata)	Low	Yes	Usually your own text. Check whether a policy was adapted from a paid template.
Employee handbook	Handbook with PTO, conduct and benefits sections	Not named as a type on any buyer page we read	Medium	Yes	Your own drafting is yours to license; benefit-plan summaries from a provider and named HR contacts are not.
Operations and product manuals	Field service manual; customer onboarding guide	Not named as such; nearest is micro1’s "operations records"	Low	Yes	Yes if you wrote it. A manufacturer’s manual you keep on file belongs to the manufacturer.
Sales and support playbooks	Objection-handling playbook; escalation guide	Not named as such; nearest are micro1’s sales and support records	Medium	Yes	Worked examples often quote real customers and deals. Those lines come out.
Project and meeting docs	Planning docs, retros, meeting notes	micro1 (product planning); Scale AI (documents and files)	Medium	Yes	Full of first names and opinions about people. Licensable once those are removed.
HR case files and reviews	Performance reviews, investigations, payroll exports	Scale AI lists HR records, with employee identities excluded	High	No	The content is about identifiable people. Treat as out unless counsel signs off on a de-identified set.
Client deliverables	Reports, decks or code written for a client	Buyers ask for rights you hold; these are often the client’s	Medium	No	The client contract usually assigns the work product to the client. Check the MSA before counting these.
Third-party material	Purchased courses, vendor manuals, copies of paid standards	Not yours to license	Low	No	Someone else holds the copyright. Leave it out of the inventory entirely.

Disclosure: a buyer named above can pay us once an introduction turns into a deal. The table is written without regard to it.

Why do AI buyers want procedures and manuals?

A procedure is a record of how a real business makes decisions in order: check this, then call that person, then log it here. Public web text has very little of that. Buyers selling to AI labs want material that shows real operational workflows and depth in a domain; Scale AI's page uses close to those words when it says what it looks for.

micro1 is the most explicit about documents. Its data page lists SOPs, process documentation and internal knowledge bases next to bug tracking, support, sales and legal records.

That does not make every document valuable. A two-page leave policy copied from an HR template tells a model little. A 300-page field service manual your engineers wrote and kept current for six years tells it a lot.

What makes one set of documents worth more than another?

Signals that help, and what they look like in practice
Signal	What it looks like
Years of history	Versions and edit dates going back several years, not a single snapshot.
Structure	Pages linked to each other, labels, owners, templates used consistently.
Domain depth	Specialist work (logistics, field service, compliance) described step by step.
Language	Mainly English; micro1 says this is what it prioritizes.
Clean rights	Written in-house, no client ownership, no third-party text.
Low personal data	Few names in the body, so redaction does not gut the text.

Sources, with read dates

micro1.ai/data (opened 11/7/2026)
scale.com/data-partnership (opened 11/7/2026)
nyne.ai/sell-company-data (opened 11/7/2026)

How should you count what you have?

Buyers scope from a description, so the useful first step is an inventory, not an export. One row per space, drive or folder:

Where it lives (tool and space name).
Rough page or file count, and the oldest edit date.
Which team writes it.
Whether it names employees or customers in the body text, often, rarely or never.
Whether any of it was written for a client or bought from someone.

That table answers most of a buyer's first questions, and it is what our form asks for. Then read what to redact before you license.

Questions on file

Are old or outdated SOPs still worth including?: Often, yes. Revision history shows how a process changed, and micro1’s page counts several years of records in a company’s favor. Mark superseded versions so a buyer can tell them apart.
Do we need a minimum number of documents?: No page we read sets a document count. Every published threshold is about the company instead. micro1 wants a full-time staff of 30 or more, Scale AI would rather start at 40, and Nyne asks for a US W-2 payroll of 20 plus three years of trading.
Can we license documents written in Spanish or French?: Possibly, but micro1 says data primarily in English is what it prioritizes. Other buyers do not publish a language rule. Note the language split when you describe the collection.
What about documents built from a paid template?: The parts you wrote are yours; the template text may still belong to its publisher. Check the template license or rewrite those sections before counting them.
Is a Confluence space better than a folder of Word files?: Usually. Linked pages, labels, owners and dates are metadata, and Scale AI lists documents and files with metadata and Confluence knowledge records among what it takes. Loose files can still be licensed; they simply carry less context.

Related memos

Get a quote