Selling company documents and shared drives for AI training

Yes: a company can license its own documents, wikis and shared drives to AI buyers, and buyers ask for them by name. Scale wants wikis and project records carried over with their complete history and metadata, and documents with their authorship, timestamps and access permissions as well as their contents; Turing says communications, tickets, wikis and databases stand on their own and are often the most valuable part. All four systems where most companies keep their knowledge (Google Drive, SharePoint and OneDrive, Confluence and Notion) let the company export its own content, and their terms say the company owns it as between it and the vendor. What they restrict is a third-party app pulling the same content through their APIs: Google's developer policy bars using data obtained that way to create, train, or improve a machine learning or artificial intelligence model. So the export has to be your own administrator's, never a buyer's connector. What stands between a shared drive and a sale is rarely the vendor. It is what is inside the drive: reports and PDFs other people wrote, files your customers and partners sent you under confidentiality, personal files, and documents written by contractors who never assigned their work. Buyers price documents as part of the company's whole record: across every public offer for one company's records the middle half sits at $174K to $693K per company, and the median of the sales that actually happened is $28.3K per company. The steps are: count what you hold, export it yourself while the subscription is live, take out what is not yours to license, de-identify the rest, and put its description before every buyer that fits at once.

This guide is for a founder, officer or trustee of a company, operating or closed, whose knowledge lives in documents: decisions, specifications, processes and playbooks in Google Drive, SharePoint and OneDrive, Confluence or Notion. It covers which documents buyers value and which they leave out, what each system lets you export and what an export can lose, the vendors' API terms, who owns documents written by employees, contractors and outsiders, the confidentiality owed to customers and partners, de-identification, what documents fetch as part of a company's records, preparing and delivering them, the case of a closed company, and how a sale through AIData.Broker works. Every quote comes from the page it names, read on the day shown in its record.

This guide is research from public sources, not legal advice.

Contents

1. What a company's documents hold, and what buyers value in them
2. What buyers leave out
3. Getting documents out of each system
4. What an export can lose: history, links and permissions
5. The vendors' API terms: why the export must be your own
6. Who owns the documents: employees, contractors and outside authors
7. Confidentiality owed to customers and partners
8. De-identifying documents
9. What documents fetch as part of a company's records
10. Preparing and delivering documents
11. A closed company's documents
12. How a sale through AIData.Broker works
13. Questions people ask
14. What this means for you

1. What a company's documents hold, and what buyers value in them

Chat records how a company talks; documents record what it decided and how it does things. A product specification sets out what was going to be built and why. A design document weighs three approaches and picks one. A runbook says what an engineer on call does when a service fails. A sales playbook says how the company qualifies a prospect, prices a quote and handles an objection. An onboarding guide, a pricing memo, a postmortem, a quarterly plan, a board-approved policy: each is the company's own knowledge, written down by the people who held it, in the order they worked it out. Public text on the web has very little of this, because companies do not publish their internal documents.

That is why AI buyers ask for documents. Models that are meant to work inside companies (agents that answer questions about a company's own processes, draft the next specification, follow a runbook, or hand a task to a colleague) learn from examples of how real companies structure and record their work. The buyers say so in their own words. Scale's data partnership page lists Jira, Confluence, Asana, Notion and Linear and wants "Tickets, wikis, and project records carried over with their complete history and metadata", and from documents "File contents plus their metadata, authorship, timestamps, and access permissions". Turing, which buys whole company records through its Lazarus program, answers the question whether it buys code on its own: "No. Communications, tickets, wikis, and databases stand on their own and are often the most valuable part." Turing also pays a stated rate for what it calls org artifacts, "tickets, specs, docs, support threads", at $50 per artifact when bundled with a codebase. And DataVendor, a buyer of codebases, asks for "architecture or product notes" and "any docs that explain what the code does" to come with the code.

Documents are also the most common record a company holds. Of the startup workspaces SimpleClosure, a wind-down service, evaluated for sale, 87% used Google Drive, close behind the 95% that used Slack. The largest records sale on public record shows the scale a large company reaches: the asset schedule of Spirit Airlines' de-identified data sale marks as included 17,082,644 OneDrive items and 20,577,677 SharePoint items beside its email and Teams, and Google agreed to pay $10,000,000 with delivery in deidentified form.

Five kinds of document carry most of the value.

Decisions with their reasons. A memo that records a choice, the options rejected and why, is a complete example of judgment. A decision log, a postmortem that ends in action items, a design review with comments from the people who approved it: each shows a problem moving to a resolution. A folder of final slide decks with no reasoning behind them is worth much less.

Specifications and designs. Product requirements, technical designs, architecture notes and API descriptions connect directly to the work product. They are what DataVendor asks to accompany a codebase, and Troveo, a buyer, tells holders to export "design docs and ticket archives, before any subscription lapses" alongside the repository.

Processes and playbooks. Standard operating procedures, runbooks, support macros, onboarding guides and sales playbooks describe how work is done, step by step. Red Deer Data reports a completed sale of a logistics firm's "Dispatch, customer messages, SOPs, logistics workflows, and operating conversations" for $500K, paid in 7 days, and names SOPs among the business software records it buys.

History. A document's value grows with its versions and its context: who wrote it, who changed it, when, and who could see it. Scale's words, "complete history and metadata" and "authorship, timestamps, and access permissions", are a buyer telling you that a document stripped to its final text is a smaller asset. Section 4 covers what each export keeps.

Links to the rest of the company. A specification that links its Jira epic, a postmortem that links the Slack thread of the incident and the pull request that fixed it, a playbook that links the CRM fields it refers to: each ties the document to the work around it. Sunset, a wind-down firm that buys closing companies' records, told Forbes it prices on how traceable records are and how well they connect across systems, and that a ticket tied to a specific code commit is worth more than a standalone document. Collinear, a buyer, says the same from the other side: among what it does not need are "Loose documents with no company around them".

That last point is the one to hold on to. Buyers do not buy documents as a pile of files. They buy a company's record, in which the documents explain what the chat, the tickets and the code were for. The Slack guide and the codebase guide cover the systems your documents link to.

2. What buyers leave out

A shared drive holds far more than the company's own knowledge. Over the years it collects downloaded reports, vendor brochures, customers' files, signed agreements, scanned passports, personal photos and the HR folder. A buyer's counsel works through the same list of questions for every records sale, and several of them are about exactly this material. Our rights reference holds the full list; these are the ones that decide what leaves a drive.

The source counsel cite for that list, a Frankfurt Kurnit article, puts the advice in one line: "Exclude high-risk data entirely rather than relying on de-identification." It then names the categories: "privileged material, HR and personnel files, payroll and tax records, health and benefits information, union-related communications, customer and partner confidential information, licensed data, and data from regulated systems."

Two kinds of exclusion follow, and they are done differently. The first is by place: a folder, a shared drive, a SharePoint site, a Confluence space or a Notion teamspace that is out of scope as a whole (Legal, People, Finance, Customers, a client's project space). Most of the excluded material in a typical company sits in a few such places, and excluding them is cheap and auditable. The second is by kind: third-party PDFs, scanned identity documents and personal files scattered through otherwise useful folders. Those need a pass over file types and contents. Do the first before the second; it shrinks what has to be read.

What remains after both is the company's own writing: its specifications, decisions, processes, playbooks, postmortems and plans. That is what buyers are asking for, and it is also the material with the clearest title.

3. Getting documents out of each system

All four vendors' contracts say the customer owns its content as between it and the vendor, and all four document an export. The differences are in who can run it, what it covers, and how long you have after the subscription ends. This table is built only from what our records of their terms hold, as read on the dates in each record.

Sources: Google Workspace Admin Help and the Google Workspace API User Data and Developer Policy, Microsoft's Products and Services Data Protection Addendum and API terms, the Atlassian Customer Agreement and data contribution pages, Notion's Master Subscription Agreement and export help page, each held in its terms record: Google Workspace, Microsoft 365, Atlassian, Notion.

What each row means in practice:

Google Drive. The organization export is the route, and it is open on every edition. The constraints are practical. The person who runs it must be a super administrator whose account is at least 30 days old, so a company that has lost its only super administrator, or creates a new one in a hurry, waits a month. The export is not instant: it is "available no earlier than 48 hours after you start the export". Exports for specific users or filtered exports are available "if available" on the edition, and advanced editions such as Enterprise Plus use a separate flow to choose what to export. Organizations with more than 1,000 users, or with FedRAMP authorization, must contact support first. Google Vault, on editions that include it, is the eDiscovery route. Our deletion clock records No recovery window stated for how long Google keeps a cancelled account's data, which is the strongest reason to export while the subscription is live.

SharePoint and OneDrive. Microsoft's data protection addendum is unusually clear on both points that matter: the customer "retains all right, title and interest in and to Customer Data", and "At all times during the term of Customer's subscription ... Customer will have the ability to access, extract and delete Customer Data stored in each Online Service". After the subscription ends, data is kept "in a limited function account for 90 days after expiration or termination of Customer's subscription so that Customer may extract the data", and our deletion clock puts the outer limit at 180 days at most. Bulk export for a whole organization is usually done through Microsoft Purview eDiscovery; Microsoft Learn says its premium features require Office 365 E5 or Microsoft 365 E5, or E5 add-ons. Our record notes that a vendor guide puts standard content search and export in E3 and above, which we have not verified against Microsoft's licensing documents, so check your own licences before you plan the export.

Confluence. Atlassian's customer agreement, effective 1 October 2026, says "As between the parties, Customer owns all intellectual property and other rights in Customer Data", and that "The Documentation describes how Customer may retrieve its Customer Data from the Cloud Products." Confluence offers space and site exports. The clock is the risk: "Following expiration or termination, unless prohibited by Law, Atlassian will delete Customer Data", and the deletion clock shows 60 days for Jira and Confluence Cloud. One more point belongs in your rights file. Since 17 August 2026, Atlassian uses de-identified, aggregated metadata and in-app data to improve its own AI features, by default on its lower plans: "If your Atlassian organization's highest active plan is Free, Standard, or Premium, metadata contribution is always on", while "every customer has the option to opt out of contributing in-app data". Atlassian holds no exclusive right over your content, but a buyer asking for exclusivity will ask whether any vendor has already derived value from it ( vendor already trained). Look at your data contribution settings and write down what they were.

Notion. Notion's agreement says the customer "retains all right, title, interest and ownership of the Customer Data" and "may export Customer Data from the Services using Notion's self-service export functionality in an industry-standard format" at any time during the subscription. Its help page names the formats: "You can export all your pages as HTML, Markdown, or CSV (for databases)". Three limits matter. First, "Pages that the exporter doesn't have access to, such as private pages of other users, will not be included in the export", so one administrator's export is not the whole workspace, and you must say what it leaves out. Second, the workspace's top administrators can switch export off in the security settings, so check it is on before the day you need it. Third, the download link "will expire after 7 days". A workspace-level export through Notion's admin API is in beta on Enterprise. After termination, Notion "will make all Customer Data available to Customer for electronic retrieval for a period of thirty (30) days".

Across the four, the pattern is the same as for every other business system we have read: of 18 systems, 13 let the holder export freely. The vendors are not what stops a document sale. The calendar is, together with what sits inside the files.

4. What an export can lose: history, links and permissions

Buyers price a document with its context. Scale names the context it wants: "complete history and metadata" for wikis and project records, and for files "metadata, authorship, timestamps, and access permissions". An export can lose each of them, and the vendors' pages in our records do not promise that any of them survive.

Version history. None of the four vendors' export pages in our records says that its export carries a document's earlier versions. Notion's help page names HTML, Markdown and CSV, which are formats for a page's content. Treat version history as something to test, not to assume: export one document you know has many revisions, and look at what arrived. If the history matters to a buyer (a specification that changed over a year is a better example than its final text), say so in the description and agree in the licence how it is delivered. Do not promise in the description what the export cannot deliver; Scale's own warning on its terms is that history and metadata are part of what is bought.

Links. A link from a document to another document, a ticket, a chat thread or a code change carries value only if the other end comes too. Two records show how links break. Notion's export leaves out other users' private pages, so links into them lead nowhere. And when documents are shared in chat, the chat export may not carry the file: Slack's export holds links to files, not the files themselves. The fix is the same in both cases: export every system the links point to, while all of them are live, and keep a note of which system each link belongs to. Troveo asks for exactly that: "package the exports with a short note on what each system was used for and over what dates".

Authorship and permissions. Who wrote a document, when, and who could see it are part of what Scale asks for. They are also personal data (an author's name) and evidence (a permission setting shows a document was confidential). Keep them in the raw export, then decide in de-identification how authors are represented (section 8). For confidentiality review, the permissions are useful: a document shared with an outside domain, or restricted to a few people, is a document to look at before it is licensed.

Comments and suggestions. The discussion on a document (comments, suggestions, resolved threads) is often where the reasoning is. Our records do not say whether each export keeps it. Check it in the same test export.

The preparation guide sets out the same test for every system a company runs, with what buyers ask from each. For documents the rule is simple: run one export early, open it, and write down what it kept and what it lost. That note is what lets you describe your documents honestly to a buyer, and it is part of the record of how each export was made.

5. The vendors' API terms: why the export must be your own

Some buyers and brokers offer to connect an app to your Google Workspace, Microsoft 365 or Notion and pull the documents for you. That is the route to refuse, because it is the route the vendors' terms restrict.

Google. Google's Workspace API User Data and Developer Policy, last updated 3 September 2026, prohibits an app that obtains Workspace data through Google's APIs from "Transferring, selling, or using user data to create, train, or improve a machine learning or artificial intelligence model beyond that specific user's personalized model for the appropriate use case or user-facing feature", and from "Transferring or selling user data to third parties like advertising platforms, data brokers, or any information resellers". Our record's reading: export with the native Data Export tool or Vault and transfer the files under your own licence, rather than authorizing a third-party app to pull the data.

Microsoft. Microsoft's API terms, which cover Graph, SharePoint and OneDrive, bar an app from "Scrape, build databases or otherwise create copies of any data accessed or obtained using the Microsoft APIs, except as necessary to enable an intended usage scenario for your Application", and from "Redistribute or resell, or sublicense access to, the Microsoft APIs, any data obtained using the Microsoft APIs". A change to Microsoft's Product Terms on 1 May 2026 adds, for the Office apps, that "outputs generated through Robotic Process Automation, bots, or other similar technologies may not be used to develop, train, evaluate, improve AI or machine learning models". Our record reads that clause as aimed at bot or automation outputs of the Office apps; it does not on its face bar a customer from licensing its own documents. The same record's advice is to prefer customer-run eDiscovery or admin export over a third-party Graph app.

Atlassian and Notion. Neither customer agreement, as we read it, bars the customer from using its own data to train AI. Atlassian's bars using "the Products to develop a similar or competing product or service", and Notion's bars accessing the services "for the purpose of building a competitive product or service or copying its features or user interface". Atlassian's developer terms were not retrievable when we read them, and Notion's were not reviewed, so what they say about apps is not established. Notion's April 2026 agreement update adds a point to note: enabling integrations may require sharing Customer Data with third parties, so a buyer's connector on your workspace is a disclosure you authorize.

What it adds up to. Of the 18 business systems whose terms we have read, 6 bar training AI on what an app pulls through their API, and 2 bar copying it. None of the four document systems here bars a company from licensing what its own administrator exported. So the route that works is the one buyers' counsel treat as the standard: "Each system was exported by its own admin with the vendor's native export tool, the method is recorded per system, and no third-party connector touched it." The red flag on the same question names data pulled through a connector or scraper from ... Google Workspace APIs. A buyer who asks you to install its app is asking you to deliver data its counsel will later call tainted.

6. Who owns the documents: employees, contractors and outside authors

The vendors' contracts settle ownership between you and them. They do not settle it between you and the people who wrote the documents. The IP assignment question says why: "SaaS terms say the customer owns its content only as between it and the vendor, not against its own contractors." The words "as between the parties" in Microsoft's and Atlassian's terms mean exactly that. Three groups of authors are in a typical drive.

Employees. Under US copyright law, a work "prepared by an employee within the scope of his or her employment" is a work made for hire, and for a work made for hire "the employer or other person for whom the work was prepared is considered the author", unless the parties have expressly agreed otherwise in a signed written instrument. A specification an engineer wrote as part of the job, a playbook a sales lead wrote for the team, a postmortem written after an incident: these belong to the company. The limit is in the words "within the scope of employment". A side project an employee kept in the company drive, a personal essay, a document written before the person joined are not covered by the definition, and they belong in the personal-files pass of section 2.

Contractors and agencies. A contractor is not an employee, and the second branch of the same definition covers only works "specially ordered or commissioned" for a narrow list of uses, such as a contribution to a collective work. Most of what a freelancer writes for a company (a strategy deck, a brand guide, a technical design) belongs to the company only if the freelancer signed an assignment. The good answer to the IP assignment question is "Signed invention and IP assignment agreements for all contributors, including contractors and agencies; gaps listed and excluded." The red flags are "Freelancers or agencies without written assignment" and "founders who never assigned pre-incorporation work". In a drive, contractors' work is usually easy to find: it sits in folders named for the agency or the project, or it was shared from an outside domain. List it, check the agreements, and exclude what has no assignment.

Outside authors. The PDFs, reports, articles, slide decks from conferences and pages copied from vendors' documentation that fill every company drive belong to the people who wrote them. The company has a copy, not the copyright. Buyers' counsel treat this as a blocker, and the reason is a recent ruling: in September 2026 the Third Circuit held, in Thomson Reuters v. ROSS Intelligence, "that Thomson Reuters's materials are copyrightable and that ROSS's use was not fair", where ROSS had used Westlaw's headnotes to train a competing legal research tool. The court found the use "minimally transformative, at best", and that the evidence showed "the market for licensing headnotes as text to train AI is rapidly developing". For a holder, the lesson is not about fair use at all: you cannot license what you do not own, and a buyer will ask you to warrant that what you deliver does not infringe. The source counsel cite for that warranty puts it plainly: buyers want a promise "That the use of the data for training as specified in the agreement does not infringe a third party's copyright". The fair use guide sets out the rulings in full.

Bought material is a separate case. Research reports, market data, stock photos, templates and fonts the company paid for come with a licence, and those licences "usually forbid resale or sublicensing, so it usually cannot be passed on to a lab without the licensor's consent". The good answer is that "Bought and licensed material is identified and removed before packaging, and the provenance file shows what remains is the company's own."

The authority to sell is the last ownership question, and the simplest to get wrong. The licensor must be the company, not a founder personally: buyers ask that "The legal entity that collected the data is the licensor; the board has approved the licence in writing", and the red flag is "A founder selling personally after a shutdown, a former employee holding an export, or no board resolution." Where the documents are all or most of what the company has left, Delaware law requires shareholder approval for a sale of "all or substantially all of its property and assets".

7. Confidentiality owed to customers and partners

A shared drive is where other companies' information ends up. A customer sends a requirements document, a partner shares a roadmap under an NDA, an acquirer opens a data room, a supplier sends its price list. Each arrived under a promise about how it would be used, and that promise travels with it into any sale.

Partners and counterparties. The counterparty question is direct: does the data contain "other companies' confidential information, such as partner documents, pricing, roadmaps or anything shared under an NDA?" Its reasoning: "NDAs usually limit use to a stated purpose and survive termination, so passing counterparties' material to an AI lab invites breach of contract and trade secret claims." The red flags it names are drive material: "vendor negotiation threads, M&A data rooms, or client deliverables in the archive." The Spirit Airlines sale shows the risk is real. Five third parties objected to the sale over their own data inside the records, one of them in a cross-motion "for Enforcement of Confidentiality Agreements", and the employee data inside the records repeatedly delayed the hearing. The good answer is that "partner folders and attachments from third parties are excluded or reviewed, and you can list which counterparties appear."

Customers. Where your documents hold your customers' material (their uploaded files, the reports you produced from their data, their account plans and correspondence), your contracts with them decide what you may do with it. The customer contracts question explains why: "Where you hold customer data as their processor you may use it only on their instructions, and a processor that decides its own purposes becomes a controller in breach." Its source, GDPR Article 28, says so in terms: "if a processor infringes this Regulation by determining the purposes and means of processing, the processor shall be considered to be a controller in respect of that processing." The good answer is that "Customer content ... is excluded, or each customer contract expressly permits this use, or the data is anonymised" to a proper standard. A further point comes from Venable's analysis of AI training contracts, cited in the same record: "training data can be "deidentified" and still contain confidential or valuable information". De-identifying a customer's strategy document removes the names; it does not remove the customer's strategy.

What you promised in your own policies. The privacy policy and terms in force when the documents were collected matter too. The FTC's position, recorded in the prior promises question, is that "A business that collects user data based on one set of privacy commitments cannot then unilaterally renege". For most internal documents this is a short check, but for drives that hold customer or user material it can decide the scope.

Your own trade secrets. The same review is a chance to decide what the company itself wants to keep. A pricing model, a proprietary method, the source of a competitive edge: licensing it to a buyer that also works for your competitors may cost more than it earns. The good answer to the privilege and trade secrets question includes that "trade secrets the company wants to keep are excluded or the buyer's use is restricted contractually".

In practice, the confidentiality review of a drive runs by place first. Folders and spaces named for customers, partners, deals and legal matters are usually out of scope as a whole. Files shared from or with outside domains are the next pass, and the sharing permissions an export keeps (section 4) make them easy to find. What remains is mostly internal writing, which is the material buyers want.

8. De-identifying documents

Every buyer that takes company records expects personal data to be removed before or as part of delivery. Who does it, and when, differs by buyer. Scale says it "anonymizes everything inside your environment, before any data is extracted". Turing says "Personal data and secrets are removed before anything reaches a model" and that "you approve scope before we begin". Polyshares says "You also do not need to clean or scrub an export. Polyshares handles anonymization before anything moves." Troveo says names, contact details, credentials and client identities "are scrubbed to our documented standard before your data is put to work". Corpus Bridge, a broker, says the standard is negotiated with the buyer and you approve it before anything is delivered. In most of these models, someone other than you reads your raw documents first, so decide which you accept before you choose a buyer.

Documents have their own shapes that a de-identifier must handle:

• Authors and editors. The metadata Scale asks to keep (authorship, timestamps, permissions) names people. Replace names with stable pseudonyms so that the buyer still sees that the same person wrote the specification and approved the design, without seeing who it was.
• Names and contact details in the text. Documents name staff, customers, candidates and partners in their body: an org chart, a meeting note, an account plan, a hiring scorecard. These need text-level detection, not just a swap of the author field.
• File and folder names. A folder called with a customer's name, a file named for an employee's review, a path that holds an account number: these leak identity even when the contents are clean.
• Comments and suggestions. The discussion on a document is often more personal than the document: who disagreed with whom, and in what words.
• Secrets. Runbooks and engineering wikis hold credentials, connection strings and API keys. They must come out whatever the buyer's standard.
• Sensitive topics. Health, family matters and beliefs appear in HR folders and in casual notes. A content screen removes records about health, children, beliefs, union membership and other special categories rather than trying to mask them line by line.

California's law defines deidentified information with commitments attached: the business must "Publicly commit to maintain and use the information in deidentified form and not to attempt to reidentify the information" and bind recipients by contract ( CCPA deidentification). Our de-identification standard sets out the method we advise: removal before anything leaves you, by you or an independent provider you choose, with a group-size floor of 11 for any delivery and 300 random records read by a person before it goes. Our free de-identification tool runs in your own browser, so the documents never leave your computer, and it reads Word, PDF, Excel and other office files. On company Slack, email and documents it measured an F1 of 86.8%, finding 95.8% of people's identifiers. No tool is perfect, which is why the standard pairs it with a person's review before transfer.

Exclusion and de-identification work together, and in that order. Exclusion removes whole folders and spaces that cannot be licensed (legal, HR, customers, partners); de-identification makes what remains safe to deliver. For documents the order matters even more than for chat, because the excluded folders hold the densest personal material: an HR folder de-identified is still an HR folder.

9. What documents fetch as part of a company's records

No buyer we have read publishes a price per document for one company's internal files. They price the company's whole working record, of which documents are one part, and several name documents or drives among what the price covers. Here is what the public record shows.

Across every public offer for one company's records, the middle half sits at $174K to $693K per company. The middle of the buyers' offers is $316K per company, and the median of the sales that actually happened is $28.3K per company. Published tiers sit above what closes, partly because sales are disclosed more often for small closed companies; the prices guide explains the gap.

Three patterns in those rows matter for a holder of documents.

Documents raise the price of the whole package more than they fetch alone. Ooak's estimator values a 51 to 100 person operating company with four or more years at $85k to $121k with six tools, against $16k to $22k with only Jira, GitHub and HubSpot; the lever drawn from it reads "Bring every connected system, especially documents and email". The difference is not all documents, but it shows how buyers treat them: as the explanation that makes tickets, code and chat worth more. Corpus Bridge's illustration puts a separate figure on Drive files, $24,000 for 86k files out of $152,700, which is a useful sense of proportion: a real share of the total, not most of it.

A per-document price from a catalog is a different market. Text sold as a ready-made dataset lists at $4.88 to $49 per document or page off the shelf. That is the asking price for public or licensed documents packaged for anyone to buy, not for one company's internal record, and it is not the right yardstick for yours. Turing's $50 per artifact is the closest per-document figure for company material, and it applies only when bundled with a codebase; ask any buyer that quotes a lump sum for its per-unit build-up ( the lever).

Closed companies get less, and the first offer is rarely the best. The market pays for a closed or closing company's records about 0.65 times what it pays for the same company operating. Business Insider reported in October 2026 that small businesses receive six-figure offers that expire in 48 hours for their internal records, and a buyer told MIT Technology Review that buyers quote very differently because they buy on different theses, which is why refusing a short deadline and collecting several offers is the most useful single step a holder can take. Our valuation puts a range on your own company from these records before you talk to any buyer, with no upload.

10. Preparing and delivering documents

The preparation guide sets out the full order for any company's records. For documents it comes down to these steps, most of which need no buyer.

1. Name the systems and their administrators. Write down each system that holds documents (Google Drive, SharePoint and OneDrive, Confluence, Notion, any others), the plan or edition, and who holds the top administrator role. On Google, check the super administrator account is at least 30 days old; on Notion, check that export has not been switched off.
2. Count before you export. Count files and pages per system, the years they span, and how many people wrote them. Turing's intake starts from a rough count per system and says estimates are fine; it accepts an intake only from someone with admin authority.
3. Map the places. List the shared drives, SharePoint sites, Confluence spaces and Notion teamspaces, and mark each one in scope, out of scope (legal, HR, finance, customers, partners, deals) or to review.
4. Run one test export and open it. Choose a document with a long history and links, export it, and write down what arrived: versions, comments, authorship, permissions, links (section 4).
5. Run the full export yourself. Use each vendor's own export, run by your own administrator: Google's organization export or Vault, Microsoft's eDiscovery or admin export, Confluence's space or site export, Notion's workspace export. Never install a buyer's app to pull the files. Download Notion's export before its link expires after 7 days.
6. Export what the documents link to. Tickets, chat and code, so that links still lead somewhere. Keep a note of what each system was used for and over what dates.
7. Exclude, then de-identify. Remove out-of-scope places from the copy, then third-party, licensed, customer and personal files by kind, then de-identify what remains to the standard you chose (section 8). Draw any sample for a buyer from the de-identified copy, never from the raw export.
8. Write down how you did it. Who exported what, when, from which system, what was excluded and why, which contractors' work had no assignment, what the Atlassian data contribution settings were, and who de-identified. This becomes your rights file, and it is your proof of title: buyers have been offered stolen archives, and one buyer described a group that "did not have the rights to sell this data".
9. Agree delivery and acceptance in the licence. The format (as exported, or converted), what history and metadata are included, who de-identifies and to which standard, objective acceptance criteria and a review window.

Keep two things apart: what you describe to buyers and what you deliver. Buyers price from the description first: counts, years, systems, the kinds of document (specifications, decisions, runbooks, playbooks), what links to what, and what stays out. Polyshares says "Someone on the team reads the anonymized sample before Polyshares prices the record." The full export moves only after a licence is signed, and only the scope the licence names.

11. A closed company's documents

A closed company holds its documents on a clock it does not control. Three questions decide whether they can still be sold: do the files still exist, who can sign, and who can run the export.

Do the files still exist? Each vendor deletes on its own schedule after the subscription ends. Atlassian will delete Customer Data after expiration or termination, and our deletion clock shows 60 days for Confluence Cloud. Notion makes data available for 30 days after termination. Microsoft keeps it in a limited function account for 90 days so it can be extracted. For Google Workspace the deletion clock records No recovery window stated. So the first act of any wind-down that may sell its records is to keep the document systems paid, or to export them in full before the subscriptions lapse. Troveo's advice is the same: export "before any subscription lapses". The deletion clock shows the dates for each of your systems.

Who can sign? The company sells, not its founders personally. Buyers' counsel ask for the legal entity that collected the data to be the licensor, with the board's approval, and for a wind-down, board approval and any investor consents. In Delaware, where many US startups are incorporated, a dissolved corporation continues for three years to wind up its affairs, and it must adopt a plan of distribution that provides for its claims first. In a chapter 7 bankruptcy the trustee signs, because the trustee's duty is to "collect and reduce to money the property of the estate", and a sale that includes personally identifiable information runs through section 363(b)(1) and its privacy conditions. Buyers ask that the right officeholder signs. Our estates page sets out the path for trustees, assignees and dissolved companies, and the guide to SimpleClosure and its alternatives compares the wind-down services that also sell records.

Who can run the export? The administrator may have left. On Google, a new super administrator account must be at least 30 days old before it can run the organization export, so create it, or transfer the role to an officer or the trustee, well before you need it. On Notion, one administrator's export misses other users' private pages, which matters more when those users are gone.

A closed company's documents carry the same questions as an operating one's, with one difference: the authors and the counterparties are harder to reach. That makes the contractor and confidentiality passes more important, not less, because nobody remains to ask. The value is still real: the cielo24 records, built over more than a decade and including Google Drive files, sold for what SimpleClosure describes as mid-six figures during the wind-down, while the market pays a closed company about 0.65 times what it pays an operating one.

12. How a sale through AIData.Broker works

AIData.Broker represents companies selling their data to AI buyers. We act only for the holder, as its agent in the sale, for holders based in the United States, the United Kingdom and Canada. For a holder of documents and shared drives, a sale through us runs like this.

The valuation comes first, with no upload. The valuation asks about your systems (which document tools, how many years, how many people, what else the company ran) and returns a range from the public records in section 9. If you add files, they are measured on your own computer. The valuation binds no one.

You keep the documents and deliver them yourself. We do not hold your export, show any part of it to a buyer, or license it in our name. You, or a de-identification provider you choose, deliver directly to the buyer, and the buyer pays you directly. That keeps your documents inside the route the vendors' terms allow: your own administrator's export, never a connector.

Every buyer that fits sees it at once. Every buyer whose published requirements fit your records receives a description that does not identify you; each that gives its confidentiality undertaking receives the same datasheet, with one deadline for offers. Offers come back on the same terms, so you can compare price, de-identification model, payment timing and exclusivity side by side. Our free offer check reads any offer you already have against the published terms of the buyers we track.

The fee is due only out of money you receive. It is a success fee on what you are actually paid, band by band: 15% of the first $250,000, 10% from $250,000 to $1 million, and 6% above $1 million. On a $300,000 sale the fee is $42,500. Mandates are exclusive for 2 months, with a tail of 12 months for buyers we introduced, and the smallest sale we take on is $26,000. Several buyers publish an introduction fee; in a sale we run we decline it and ask the buyer to add it to your price instead, so we earn the same whichever buyer you choose. How it works sets out every step and term.

Questions people ask

Can I sell my company's documents for AI training? Yes, if the company exports them itself with each vendor's own export, takes out what it does not own or owes to others (third-party reports, licensed material, customer and partner files, legal and HR folders), de-identifies the rest, and the right person signs for the company. Buyers ask for documents and wikis by name: Scale wants them with their complete history and metadata, and Turing says wikis stand on their own.

Which documents are buyers most interested in? Decisions with their reasons, specifications and designs, processes and playbooks, and documents linked to the tickets, chat and code around them. Turing prices specs and docs as artifacts, Red Deer Data names SOPs in a completed sale, and Collinear does not need loose documents with no company around them.

Do Google's and Microsoft's terms allow it? Their terms say the company owns its content as between it and the vendor and let it export it: Google's organization export is available in all editions, and Microsoft's customers can access, extract and delete Customer Data. What both restrict is an app pulling data through their APIs: Google bars using such data to train a machine learning or AI model, and Microsoft bars building copies or reselling it ( API terms). Export it yourself.

Can we export Confluence and Notion? Yes. Confluence offers space and site exports, and Notion exports HTML, Markdown and CSV, though an export leaves out other users' private pages.

Will the export keep version history? None of the vendors' export pages in our records promises it. Test one document with a long history before you describe your records to a buyer, and agree in the licence what history is delivered.

Who owns documents employees wrote? In the US, a work prepared by an employee within the scope of his or her employment is a work made for hire, and the employer is considered the author. Contractors' work belongs to the company only with a written assignment ( IP assignment).

Can we include reports and PDFs we downloaded? No. Their copyright belongs to their authors, and buyers' counsel list shared drives full of downloaded PDFs and reports as a red flag. Bought reports and stock material usually carry licences that forbid resale or sublicensing.

How much are a company's documents worth? Buyers price the whole working record. Public offers for one company's records have a middle half of $174K to $693K per company, and the median of disclosed sales is $28.3K per company. Corpus Bridge's illustration values 86k Google Drive files at $24,000 within a larger record. The valuation gives a range for yours.

Our company has closed. Can we still sell its documents? Often yes, if the files still exist and the right person signs: in Delaware a dissolved corporation continues for three years to wind up, and in bankruptcy the trustee acts. Export before the subscriptions lapse: Notion keeps data for 30 days after termination and Microsoft for 90 days. See the estates page.

What this means for you

Your documents are where your company wrote down what it knew: why it built what it built, how it does its work, what it learned from what went wrong. Buyers ask for them by name and pay more for records that include them, because they explain everything else the company kept. They are also full of material that is not yours to license, and they sit in systems that delete them on a schedule once the subscription ends.

So do four things now, before any buyer conversation. Keep the document systems paid, and check who can run each export. Count what you hold, system by system, in files, years and authors. Map your drives, sites, spaces and teamspaces, and mark what stays out: other people's reports, licensed material, customers' and partners' files, legal and HR folders, and contractors' work with no assignment. And get a range for the whole record from the valuation, so the first offer you see is measured against the market rather than against nothing. Then export it yourself, de-identify it to a standard you chose, and put it before every buyer that fits at once. The rest of the What sells section shows the other records buyers pay for beside your documents.