# How to Produce Scanned Documents in Discovery

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2022-10-14T07:07:56Z
- Updated: 2026-10-11T19:22:20Z
- Category: Litigation support
- Canonical: https://www.schemalaw.com/blog/how-to-produce-scanned-documents-in-discovery

> How to produce scanned documents in discovery: the form to send, the text that goes with it, the privilege screen, and a quality check before delivery.

To produce scanned documents in discovery, scan the paper in the order it was kept, split it into documents, review it for privilege, number the pages, and run OCR on the final images so the text matches what you send. Then deliver the form the request or protocol asks for, with a cover letter stating the Bates range.

The risk sits in the middle of that list, between scanning and delivery. This post follows a paper set from the client's filing cabinet to the other side's inbox. It explains document handling and is not legal advice. What a production must contain depends on the requests, any agreed protocol, and your court's rules.

## What form should a scanned paper production take?

A scanned paper production can take one of three forms: searchable PDFs, one per document; single-page images with a text file per document and load files; or plain image PDFs with no text. The request, protocol or agreement decides which. Where none speaks, a searchable PDF per document is a sensible form to propose.

| Form | What the recipient gets | Suits |
| --- | --- | --- |
| Searchable PDF per document | One file per document, page images with a text layer, Bates number on each page | Small and mid-sized sets, recipients who read in a PDF viewer |
| Images, text files and load files | A folder of page images, a folder of text files, and files that tie them together | Sets going into a review platform, agency productions, anything under a written specification |
| Image-only PDF | Pictures of pages | Only where nothing requires text and the lawyer has decided to send it that way |

How the image side of those options compares, including why specifications ask for one image per page, is the subject of the post on [TIFF vs PDF for document production](https://www.schemalaw.com/blog/tiff-vs-pdf-for-document-production).

Paper has its own rule on organization. Under [Rule 34(b)(2)(E)(i) of the Federal Rules of Civil Procedure](https://www.law.cornell.edu/rules/frcp/rule_34), unless otherwise stipulated or ordered, "a party must produce documents as they are kept in the usual course of business or must organize and label them to correspond to the categories in the request." For a scanning job that sentence has a plain consequence: record where every stack came from before it goes on the scanner. Box number, folder label, and whose office it sat in. Once the pages are a PDF, that information exists only if someone wrote it down.

Whether the other side can insist on a searchable form for paper is a narrower question than it looks, because the form-of-production provisions in Rule 34 address electronically stored information. The reading of those provisions is in the section on [whether you can require a searchable production](https://www.schemalaw.com/blog/make-document-production-searchable#can-you-require-a-searchable-production). State rules differ from the federal ones.

### Unitize before anything else

A scanner gives you one file per stack. A production is counted in documents. Splitting the scan at the right places, called unitizing, comes first because every later step depends on it: privilege is decided per document, text files are made per document, and a log lists documents. Mark the first page of each letter, memo, contract or report, and keep attachments behind the thing they were stapled to.

## What does the request or protocol ask you to send with a paper set?

Read the instructions in the document requests, any ESI protocol or stipulated order, and your court's local rules for four things: image format, text, numbering and a load file. A paper set may be asked for document-level text files, a PDF text layer, or nothing beyond the images. Do not guess. Send what the governing paper says.

Where to look, in order:

1. **A court order or stipulated protocol.** It governs, and may have a paragraph headed "hard copy documents" or "paper documents".
2. **The instructions section of the requests.** Production format may be stated there in a sentence or two.
3. **Local rules and the judge's standing orders.** Some set defaults for discovery format.
4. **Correspondence.** An email exchange agreeing "searchable PDFs are fine" is worth keeping with the production file.

Agency specifications show how detailed the requirements can be. The [SEC's Data Delivery Standards](https://www.sec.gov/divisions/enforce/datadeliverystandards.pdf) state a preference that "all scanned paper and electronic file collections be produced in a structured format including industry standard load files, Bates numbered image files, native files and searchable document-level text files." They call for text "as separate document-level text files", named after the first Bates number of the document. Where the SEC approves a PDF production in place of that, every PDF "must contain embedded text that includes all discernible words within the document". Those are one agency's requirements, given here as an illustration. A private civil case is governed by its own requests and orders.

Reduce what you find to a short specification sheet before work starts:

| Question | Your answer |
| --- | --- |
| Image format | PDF per document, or single-page images |
| Text | Text layer, separate text files, or none |
| Bates prefix and starting number | |
| Confidentiality legend required on pages | |
| Load file | None, or which type |
| Color | Scan in color only where it carries meaning, or throughout |
| Delivery | Link, portal or physical media; encryption |

If the specification names fields a scan cannot supply, such as email metadata, or you cannot tell what it means, the post on [matching a custom load file specification from OCR output](https://www.schemalaw.com/blog/custom-load-file-ocr) covers reading one field by field. Raise open points with the other side before you produce, in writing.

Say in the cover letter what form this production takes. If the request for production asks for images and text, your export is the electronically stored information you are sending. Name the Bates range, the load file, and the folder of page images. The receiving party can then import the set without guessing which file begins a document.

On tools: Schema Law reads uploaded PDFs, PNGs and JPEGs and exports searchable PDF and plain text on every plan, with Word alongside. Bates stamping and redaction are not part of it, so those are done in your PDF software. It does not produce load files on any plan. The [export formats](https://www.schemalaw.com/#formats) section of the site lists everything it produces.

## Screening scanned documents for privilege before they go out

Screen a paper set for privilege after it is unitized and before it is numbered for production. Use OCR text to find likely candidates, then have a lawyer look at every candidate and decide. The search narrows where people look first. It does not clear the documents it misses, so the rest of the set still needs eyes.

A screen in two passes suits a paper set.

**Pass one: search.** Run OCR on the unitized working copies and search for:

- Names of every lawyer and law firm that acted for the client, inside and outside, including former ones.
- Firm letterhead terms and email domains.
- Words such as "privileged", "confidential", "legal advice", "attorney", "counsel", "draft".
- Names of consultants and experts retained for the dispute.

**Pass two: look.** Some privileged paper has no searchable words at all. Page through for handwritten notes of calls and meetings, annotations in a lawyer's hand on ordinary business records, unsigned draft agreements, and fax cover sheets that name a law firm in a logo the OCR could not read. Recognition is weakest on exactly these pages.

Sort each candidate into one of three outcomes: produce in full, produce with redactions, or withhold. The decision belongs to the lawyer. Software that converts pages to text makes no privilege determinations, and a search hit is only a reason to read the page.

Record the screen as you go: who reviewed which box, which terms were searched, and what was pulled. That record matters if something slips through. In federal proceedings, [Rule 502(b) of the Federal Rules of Evidence](https://www.law.cornell.edu/rules/fre/rule_502) provides that an inadvertent disclosure does not operate as a waiver if "the holder of the privilege or protection took reasonable steps to prevent disclosure" and promptly took reasonable steps to rectify the error. Rule 502(d) also allows a federal court to order "that the privilege or protection is not waived by disclosure connected with the litigation pending before the court". Whether to seek such an order, and what counts as reasonable in your case, are questions for the lawyers.

Withheld and redacted documents then need to be listed. How to draft those entries from paper is covered in the guide to building a [privilege log from scanned documents](https://www.schemalaw.com/blog/privilege-log-from-scanned-documents).

One more screen runs alongside privilege: material that is not privileged but is restricted. Personal identifiers, medical details and anything covered by a protective order may need redaction or a confidentiality legend. Check the order's terms before a vendor of any kind handles those pages.

## Re-running OCR on redacted pages before they go out

Text made before a redaction still contains the redacted words. After redactions are applied and the pages are flattened to images, discard the old text for those documents and run OCR again on the redacted images. The text you deliver must be a reading of the pages as produced, not of the pages as they were.

This mistake hands over the very words the redaction was meant to withhold, and it is easy to make because every file looks right on screen. The black box covers the paragraph. The text layer behind it, or the text file beside it, was created last week from the clean scan and still holds every word.

The SEC's standards put the requirement in one line: "For redacted documents, provide the full text for the redacted version." The order of work that satisfies it:

1. Apply redactions in your PDF software and use its function that removes the underlying content, not a drawn rectangle.
2. Save the redacted document as page images, so nothing sits beneath the boxes.
3. Delete every earlier text file and searchable copy of that document from the outgoing folder.
4. Run OCR on the redacted images.
5. Open the new text and search it for a word you know was under a box. It should not be found.

Step five is the test, and it takes seconds. Run it on every redacted document where the count allows, and on a generous sample where it does not.

Why a drawn box leaves text behind, and the full safe sequence for combining the two operations, belongs to the post on [whether to OCR before or after redaction](https://www.schemalaw.com/blog/ocr-before-or-after-redaction). The point for an outgoing paper set is narrower: OCR runs last, on the final images, after stamping and redacting, so the Bates number and the redaction label are in the text and the redacted words are not.

Keep the unredacted originals and their text in a separate, clearly named location that is never the source for a delivery folder.

## A quality check for an outgoing scanned production

Check an outgoing scanned production against its own numbers and against the specification before anything is sent. The check is mechanical and should be done by someone other than the person who assembled the set. Allow time for it. An hour here is cheaper than a clawback letter.

**Completeness**

- Every box and folder on the collection list was scanned, and the source of each document is recorded.
- Page counts per document match between the scan, the stamped file and the text.
- No pages are missing where a two-sided original was scanned one-sided.

**Images**

- Each page is upright, complete to the edges and legible at normal zoom.
- Blank backs were handled the same way throughout the set.
- Color was kept where it carries meaning, such as highlighting or colored ink.

**Numbering**

- Bates numbers run in sequence with no duplicates, and each stamp is clear of the page's own text.
- Every gap is a withheld document that appears on the privilege log, or a slip sheet stands in its place.
- File names follow the specification, for example the first Bates number of the document.

**Text**

- Text exists for every document where text is required, and no text file is empty unless the page really is blank or all handwriting.
- Text was generated from the final stamped and redacted images.
- A sample of text files was opened and read against their pages.

**Privilege and restrictions**

- No withheld document is in the delivery folder in any form: image, text, file name or folder name.
- Every redacted document passed the search test for leftover words.
- Confidentiality legends are on the pages that need them.

**Delivery**

- The delivery folder was built fresh from final files, not by deleting things from a working folder.
- The set opens on a second computer from the delivery copy.
- An exact copy of what was sent is stored read-only with the transmittal.

The sample of text files is worth doing with some thought. Pick the worst-looking pages on purpose: a fax, a handwritten note, a form with small print. If the text for those is nonsense, decide with the lawyer whether to rescan, and remember that poor text is still a truthful reading of a poor page. Do not hand-correct produced text in ways that make it differ from the image. The routine for splitting large sets and tracking failed files during conversion is in the guide to [batch OCR for discovery documents](https://www.schemalaw.com/blog/batch-ocr-discovery-documents).

## The cover letter and production log for a paper set

The cover letter tells the recipient exactly what arrived, and the production log tells your own team exactly what left. Write both from the final delivery folder, after the quality check, so the numbers in the letter are the numbers in the files.

A cover letter for a paper production can state:

- The case name and number, and which requests the production responds to.
- The production volume name, such as VOL001, and the date.
- The Bates range, first and last number, and the total page and document counts.
- The form: for example, "searchable PDFs, one per document, named by beginning Bates number".
- That the documents were scanned from paper, and how they are organized, by source or by request category.
- Whether documents were withheld or redacted, and when the privilege log will follow or that it is enclosed.
- Any confidentiality designation under a protective order.
- How the set was delivered, and that the password is being sent separately.

Agency standards again show the pattern. The SEC's requires a cover letter with each production listing, among other things, each piece of media with its volume number and "a list of custodians, identifying the Bates range for each custodian", with passwords "provided separately either via email or in a cover letter apart from the media."

A sample core paragraph, for a fictional production by Tidewater Marine Supply:

> Enclosed is Defendant's first production, volume TMS001, consisting of documents Bates numbered TMS0000001 through TMS0004212 (612 documents). The documents were scanned from paper files kept at Defendant's Norfolk office and are produced as searchable PDFs in the order in which they were kept. Forty-one documents have been withheld or redacted on the basis of privilege; a privilege log will follow by 14 November.

The production log is internal. Keep one row per volume:

| Field | Example |
| --- | --- |
| Volume | TMS001 |
| Date sent and method | 31 October, secure link |
| Sent to | Named counsel and email address |
| Bates range | TMS0000001 to TMS0004212 |
| Documents and pages | 612 and 4,212 |
| Form | Searchable PDF per document |
| Sources covered | Boxes 1 to 6, Norfolk office |
| Withheld and redacted | 29 withheld, 12 redacted |
| Checked by | Initials and date |
| Archive copy location | Read-only folder path |

When a later volume, a replacement page or a clawback comes up, this row is how you find out in a minute what went out, to whom and in what form. If a privileged page does get through, [Rule 26(b)(5)(B)](https://www.law.cornell.edu/rules/frcp/rule_26) lets the producing party notify the recipient of the claim and its basis, after which the recipient "must promptly return, sequester, or destroy the specified information and any copies it has." Your log tells you which volume and which Bates numbers to name in that notice.

## Try it on a real document

Upload a scanned PDF and export it as Word, a searchable PDF or plain text on the free plan. Your documents are never used to train an AI model. [Start free](https://www.schemalaw.com/api/auth/google)