Selling a company's paper archive for AI training

Yes: a company's paper archive can be licensed to AI buyers, once it is scanned into searchable text, because what a buyer licenses is the words, never the paper. Scanning is the cheap part. One vendor puts the national average at $0.07 to $0.12 a page, a records company's quote filed with a city council prices it at $0.11 an image with full text recognition included, and the median of the scanning prices on record is $0.092 per document or page. A standard banker's box of about 2,500 documents costs about $225 to scan. What the text is worth depends on what it is: our valuation prices general writing at $1,000 to $5,000 per million words and edited, published writing at $5,000 to $50,000 per million words, and off-the-shelf text is listed at a median of $0.001 per word. The other question an old archive raises is who owns it. Records employees wrote as part of their jobs generally belong to the company, and under US law their copyright ends 95 years after publication or 120 years after they were written, so the oldest of them are out of copyright altogether. The steps are: estimate the words, value the archive before paying to scan it, set aside what is not the company's or is about identifiable people, scan the rest with its box and folder kept, and put its description before every buyer that fits at once.

This guide is for a founder, officer, trustee or adviser of a company, operating or closed, whose history sits on paper: boxes of correspondence, ledgers, reports, minutes, files and drawings in a warehouse or an archive room. It covers what buyers pay for text and where an old archive fits, what scanning costs and how to estimate it, what to scan and how, who owns a company's old records, the personal information inside them, shredding and the originals, closed companies and estates, and how a sale through AIData.Broker works. Every price comes from the page or document its record names, read on the day shown there.

Contents

1. What an AI buyer pays for in a paper archive
2. The prices on record for text
3. What scanning costs
4. Value it before you scan
5. Scanning for a buyer: text, order and provenance
6. Who owns a company's old records
7. Personal information in old files
8. Shredding, and keeping the originals until the scans are checked
9. Closed companies, estates and archives in storage
10. Where text archives sell today
11. How a sale through AIData.Broker works
12. Questions people ask
13. What this means for you

1. What an AI buyer pays for in a paper archive

Language models learn from text, and what buyers look for is writing the web never saw: a company's own record of how it worked, decided and dealt with the world, and archives of edited writing in subjects and periods that are thin online. A paper archive can be both. A large corporation's boxes often hold board minutes, correspondence with customers and suppliers, engineering reports, procedures, catalogues and annual reports, spanning decades in which none of it was ever typed into a computer. What labs buy shows which kinds of data are in demand now.

None of that is sellable as paper. A buyer receives files, not boxes, so the archive has to become images and then searchable text, each page kept with the box and folder it came from. That conversion is a cost the holder carries before any buyer sees the data, and section 3 puts numbers on it.

Three things decide what the text is worth once it exists.

• What kind of writing it is. Edited, published writing (reports, catalogues, house magazines, published histories) is priced well above everyday business correspondence and forms. Our valuation draws that line and prices everyday writing at $1,000 to $5,000 per million words and edited, published writing at $5,000 to $50,000 per million words.
• How much of it there is, and how clean the text comes out. Buyers count words. Typed pages come out of text recognition far cleaner than handwriting, faded carbon copies or ledgers in columns, and a page whose text is unreadable is worth nothing to a model.
• Whether the company can license it. Records employees wrote as part of their jobs are generally the company's; letters others sent it, and material it collected, belong to their authors (section 6). Personal information is taken out before anything is sold (section 7).

2. The prices on record for text

There is no published price yet for a company's historical paper records as such: no company has disclosed one, and no buyer publishes a rate for them. The nearest prices on record are for archives of published writing and for business documents listed as datasets, and they bracket the range.

Three readings follow from that table.

Scale and edited quality are what move the price. News Corp's licence to OpenAI covered current and archived journalism, and News Corp also licensed its US and UK content, archives included, to Meta. Those are the largest licences of edited writing on record. A company's internal files are not journalism, but a large corporation's records over many decades are a large body of professional writing, and professional writing with a clear rights holder is what raises a text archive's price in our valuation.

Business documents are listed as datasets by the document. The contracts batch above came to about $4.88 a document at its price on 11 October, and the off-the-shelf price per document or page across listings runs $4.88 to $49 per document or page. Those are commercial listings, which a holder's own records would compete with only after their personal and confidential content is out.

Research-only licences are a different product and do not apply. Corpus libraries sell text for research use only, from old newspaper corpora such as one million words of 1965 Swedish newspapers to newswire collections. A licence to train commercial models is a different product, and our valuation prices only commercial use.

What a buyer would pay for a particular archive is decided by competition among the buyers that want it, which is why the range in the valuation is wide and why a round of offers matters more for text than a quoted rate (section 11).

3. What scanning costs

Scanning is priced three ways: by the page or image, by the box, or as a project quote. All three are on record, from vendors' own pages and from a quote a records company filed with a public body.

The median of the per-page prices on record is $0.092 per document or page.

A box is priced as if it were full. SecureScan's example assumes about 2,500 documents in a standard banker's box, and Corodata's quote says a standard records box of single-sided documents usually runs 2,000 to 2,200 images, for which it budgets about $255. A flat price per box, such as $250, costs more per page when boxes are half full; an archive of partly filled boxes is cheaper priced by the image.

Drawings and film cost far more than letters. A blueprint scanned at $1.65 an image costs many times a letter-size page at $0.11 in the same quote. If the archive holds plans, maps or microfilm, count them separately.

Indexing is priced on top. Corodata charges $0.30 per index field for indexing beyond the box-and-file organisation it delivers. A buyer needs to know where each page came from, which the box and folder give, more than it needs a typed index of every document, so keep indexing to what the box labels already say.

4. Value it before you scan

Scanning a large archive is a real outlay, and the order of steps protects it: estimate what the text is worth first, then decide what to scan.

Estimate the words. Our valuation asks how many words an archive holds, in bands from the smallest archives to the largest. To estimate it without scanning, count the boxes, estimate how full they are, take the pages per box from the scanning quotes above ( 2,000 to 2,200 images or about 2,500 documents for a full standard box), and count the words on ten typical pages from different boxes. Pages times words per page gives the total. A warehouse of full boxes of typed correspondence is far past the top band.

Say what kind of writing it is. The valuation distinguishes general writing (correspondence, forms, documentation) from edited and published writing (reports, catalogues, magazines, published histories). Most company archives are mostly the first with some of the second; if the published material is substantial and separable, it is worth valuing on its own.

Compare the two numbers. The scanning cost is pages times $0.092 per document or page, or the box count times the box price. The valuation's range is what the text could fetch. Where the scan costs a small share of the range, scanning everything is sound; where it does not, scan the boxes richest in typed professional writing first and leave the rest.

Value the archive as text: it takes a few questions, no upload, and returns a range from the public prices on the board.

5. Scanning for a buyer: text, order and provenance

A scan made for a records-management project and a scan made for a buyer of training data differ in what matters.

• Searchable text, not only images. A buyer trains on text. Ask for full-text recognition (OCR) on every page; two of the vendors above include it in their price ( Store More Store, Corodata). Keep the page images too: a buyer may want to run its own recognition on them.
• One file per document, kept in order. Documents scanned in their original order, one file each, keep the context that gives company records their value: a letter and its reply, a report and the decision that followed.
• Box and folder on every file. Corodata delivers images organised by box, each file individually indexed. That box-and-folder trail is the provenance a buyer's diligence asks for: where each page came from, and that it came from the company's own archive.
• Resolution and colour. Both vendors scan at 300 dpi. Corodata charges nothing extra for a few colour pages in a job, and colour is worth taking where pages carry stamps, annotations or coloured forms.
• Handwriting, carbons and ledgers. Recognition reads print well and handwriting poorly. Pages that will not yield usable text still have value as images to a buyer that does its own recognition, but they should be counted apart in the description, so no buyer is promised text that is not there.

6. Who owns a company's old records

A buyer pays for text the company can license, and in an old archive three kinds of material sit side by side.

What the company's employees wrote is the company's. US copyright law calls work an employee prepared within the scope of employment a " work made for hire", and its author is the employer. Correspondence, reports, minutes and procedures written by staff are the company's to license.

Very old records are out of copyright. The copyright in a work made for hire endures for a term of 95 years from the year of its first publication, or a term of 120 years from the year of its creation, whichever expires first, and a work created before 1978 that was never published endures for the term provided by section 302, the same rule. Every term runs to the end of the calendar year in which it would otherwise expire. So under US law, in 2026 the company's own unpublished records from 1905 or earlier are in the public domain unless they were first published between 1978 and 2002 (those stay protected through 2047), and so is anything published in the US in 1930 or earlier, under the older rules for published works. UK and Canadian terms differ. Nobody needs a licence to use public-domain text. What a buyer pays for then is the text itself, which exists only in the company's boxes and its scans: access and the work of digitising, not a copyright licence. That still has a price, but no buyer can be promised that nobody else will ever offer the same text.

What others wrote belongs to them. Letters the company received, reports by outside consultants, contracts drafted by the other side and publications it collected were written by someone else, and their copyright is the writer's, or the writer's heirs', for the life of the author plus 70 years for an individual. The company owns the paper but not the right to license those words. In a sale they are left out, or valued separately where they are old enough to be out of copyright too.

One court has called training on books "quintessentially transformative", and the same case shows why buyers still pay for clean sources: books taken from pirate libraries cost the buyer about $3,000 a work in the settlement that followed. That concerns the buyer's use, not the holder's right to sell, but it is why a buyer asks where every page came from.

7. Personal information in old files

Old does not mean anonymous. A company archive holds personnel files, payroll ledgers, customers' accounts and correspondence naming individuals, and the more recent the records, the more of those people are alive.

Our rule is the same for paper as for any records: records about people's health, about children, and people's bank, card or account details are cut out whole before anything is sold, never scrubbed and sold, and the rest is de-identified before a buyer receives it. For a paper archive that means setting aside whole boxes and folders by kind (personnel, medical, payroll, customer accounts) before scanning, which is cheaper than scanning them and removing pages afterwards. The de-identification standard describes the rest of the process, and the valuation asks whether the archive is about any of those categories.

8. Shredding, and keeping the originals until the scans are checked

Scanning firms commonly destroy the paper they scan, as one more line on the quote: Corodata prices shredding at $3.00 a box in the same quote as its scanning. Two cautions apply before any box goes.

• Check the scans first. Count the files against the box inventory and open a sample from every box. A missed or unreadable box cannot be rescanned after shredding.
• Know what must be kept. Some records carry legal retention periods (tax, employment, regulated industries), and a company in an insolvency proceeding should ask its fiduciary before destroying its books. That is a question for counsel or the fiduciary before the shredder, not after.

Destroying the originals does not change who owns the text: the scans are the company's records, and the rights questions in section 6 apply to them unchanged.

9. Closed companies, estates and archives in storage

Paper archives often outlive the companies that made them: boxes in a storage warehouse after a closure, a sale or a bankruptcy. Who can license them is whoever holds the company's rights now: its officers while it exists, or the fiduciary (a trustee, liquidator or administrator) of a closed or insolvent company. The valuation asks who is answering for the records, and the estates page explains how a closed company's data is sold and why it should be sold before the storage contract or the estate ends.

An adviser writing for a client can start the valuation for it; whoever can sign for the company signs the agreements later.

10. Where text archives sell today

Six buyers on our board take text archives, and they work in three ways. Which one fits depends on how much the holder wants to do itself and how much competition it wants for the price.

A marketplace suits an archive that is ready to list as it stands, with its rights settled and its personal files out: the holder writes the listing, answers buyers' questions and accepts the price offered or set. A round suits a large archive whose value is uncertain, because competing offers find the price, and a listing at a set price can leave money on the table or never sell. The two are not exclusive: our mandate is non-exclusive, so a holder may list the archive on a marketplace and run a round at the same time.

11. How a sale through AIData.Broker works

AIData.Broker represents holders selling or licensing their data to AI companies, an archive's holder or fiduciary among them. We work for the holder only: we value the archive against public prices, settle its rights with the holder, and put it before every buyer that fits at once, so the holder sees competing offers rather than the first one. The holder itself licenses or sells to the buyer, and we do not take ownership of the records.

1. A valuation first. The valuation asks a few questions about the archive (how many words, what kind of writing, who can sign) and returns a range from the public prices on the board, with no upload.
2. The rights file and the datasheet. We go through the rights questions with the holder (what the company wrote, what others wrote, what is out of copyright, what is personal), record the answers, and write the datasheet that describes the archive; the holder approves it before any buyer sees it.
3. The round. A blind profile of the archive goes to every buyer that fits, without the holder's name, less any buyer the holder leaves out. Buyers who want to go further sign an undertaking and see the datasheet and, where the holder agrees, a sample.
4. Offers and the licence. Buyers make offers, which we set side by side; the offer check reads any written offer against market terms. The holder chooses, and signs the licence with the buyer directly.

The steps and documents are on how it works. Our fee is a success fee on what the holder is actually paid, band by band: 12% of the first $250,000, 10% up to $1 million, 8% up to $3 million and 6% above $3 million, and nothing if nothing sells; on a $300,000 sale it comes to $35,000. The mandate is not exclusive: the fee is due only on a sale to a buyer we introduced, during the mandate and, if that buyer is on the list we send in writing when it ends, within 12 months after it. We run sales only for holders based in the United States, the United Kingdom and Canada.

Which archives fit. The smallest sale we take on is $23,000. Archives where a round makes the most difference are large, typed, professional and mostly the company's own writing: decades of a corporation's correspondence, reports and procedures, with the personal files set aside.

Questions people ask

Can I sell old company documents to AI companies? Yes, once they are scanned into searchable text and the company holds the rights to them. AI buyers license text; the paper itself has no buyer. The company's own records written by its employees are the company's to license, records old enough are out of copyright, and letters and reports others wrote are not the company's to license.

How much does it cost to scan documents? One vendor puts the national average at $0.07 to $0.12 a page, and the median of the scanning prices on record is $0.092 per document or page. By the box, prices on record run from about $225 to $250 for 10 boxes or more, the second with text recognition included, and Corodata's budget allows more a box (its quote above); fewer boxes cost more at one vendor. Drawings, film and handwritten ledgers cost more.

How many pages are in a banker's box? About 2,500 documents in a standard banker's box by one vendor's count, and 2,000 to 2,200 images for a standard records box of single-sided documents in another's quote. A box half full holds half that.

What is a paper archive worth to AI companies? No one has published a price for a company's historical paper records yet. Our valuation prices general writing at $1,000 to $5,000 per million words and edited, published writing at $5,000 to $50,000 per million words, and off-the-shelf text is listed at a median of $0.001 per word. A large, typed, professional archive the company wrote itself is at the upper end of what such writing fetches; correspondence that is mostly forms and receipts is at the lower.

Are a company's century-old records in the public domain? The company's own records written by employees are protected for 95 years from publication or 120 years from creation, whichever expires first, to the end of that calendar year. Under US law, in 2026 that puts unpublished records from 1905 or earlier (unless first published between 1978 and 2002), and anything published in the US in 1930 or earlier, in the public domain; UK and Canadian terms differ. Letters and documents others wrote follow their authors' terms.

Can the scanning company shred the boxes afterwards? Yes, scanning firms quote it as one more line, $3.00 a box in one quote on record. Check the scans against the box inventory first, and confirm with counsel or the fiduciary that no record must be kept on paper.

What this means for you

A company's paper archive is one of the few sources of professional writing the open web never saw. The cost of turning it into text is known and modest per page; what the text fetches depends on how much of it the company wrote, how clean it reads, and how many buyers see it.

1. Count before you scan. Boxes, how full, pages per box, and words on a sample of pages. Value the archive as text on that estimate.
2. Separate what is not the company's or is about people. Letters and reports others wrote, personnel and medical files, customer accounts: by box and folder, before scanning.
3. Scan for a buyer. Searchable text and the page images, one file per document, box and folder on every file. The datasheet builder shows what a buyer reads first.
4. Keep the paper until the scans are checked, and until counsel or the fiduciary confirms nothing must be kept.
5. Put it before every buyer at once. The price board, the text and archives page and the guide to selling company documents show what buyers pay for writing; competing offers decide what this archive fetches.

AIData.Broker acts as the holder's agent: the holder itself licenses or sells its records to the buyer, and we do not take ownership of them. The holder pays us a fee fixed in the mandate and paid only if a sale closes, and we run sales only for holders based in the United States, the United Kingdom and Canada.