# Litigation OCR: What to Convert at Each Stage of a Case

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2023-03-25T20:33:33Z
- Updated: 2026-10-11T19:22:20Z
- Category: Litigation support
- Canonical: https://www.schemalaw.com/blog/litigation-ocr

> Litigation OCR by stage: which pleadings, productions, transcripts, exhibits and record pages need converting, which do not, and who on the team does it.

Litigation OCR is the conversion of scanned case papers into text at each point where paper enters the file: pleadings and motions served as scans, documents produced in discovery, third-party records, deposition transcripts and exhibits, trial exhibits and the record on appeal. Papers that already carry text need none. Someone on the team has to own the step.

The page counts and the purpose change as a case moves. Early on you convert a handful of filings so you can answer them. In discovery you convert thousands of pages so you can search them. Before trial you convert the few hundred that matter so you can find a passage on your feet. This post maps the whole run for one matter. It explains document handling and is not legal advice.

## What does litigation OCR cover from complaint to appeal?

Litigation OCR covers any document in the case that reaches you as a picture of a page and that someone will need to search, quote, edit, tabulate or file. That set changes with the stage: the opening papers, then the discovery record, then testimony, then the trial exhibits, then the appellate record.

The purpose decides the output. The same scanned page may be converted for different reasons at different points in the case.

| Stage | Scanned paper that turns up | What the text is for | Useful output |
| --- | --- | --- | --- |
| Pleadings | A complaint served on paper; an old order; papers from an unrepresented party | Drafting a response; calendaring; quoting | Word, searchable PDF |
| Motions | Opposing exhibits; declarations signed in ink and scanned | Checking what an exhibit says; quoting it accurately | Searchable PDF, plain text |
| Written discovery | Requests and responses served as scans | Drafting responses under each numbered request | Word |
| Document discovery | The client's paper files; the other side's image-only production | Search, review, coding | Searchable PDF, text files |
| Third-party records | Subpoena returns: bank, medical, employment, phone | Finding entries; building tables | Searchable PDF, Excel or CSV |
| Depositions | Older or photocopied transcripts; marked exhibits | Searching testimony; summarizing | Searchable PDF, plain text |
| Trial | The exhibit set; demonstrative source documents | Finding a passage fast; exhibit lists | Searchable PDF |
| Appeal | Trial court filings and transcripts assembled into an appendix | A searchable, citable record | Searchable PDF |

### Five purposes, five different standards

- **Search.** The text has to be good enough that a name or term is found. An error in one word costs one hit.
- **Quote.** The words go into a brief, so each one is read against the page image before it is used.
- **Edit.** You want a Word file to draft from, and layout matters as well as words.
- **Tabulate.** You want rows and columns out of a statement or ledger, and every figure is checked.
- **File.** The converted PDF itself goes to a court, and the court's format rules apply. The survey of [searchable PDFs for court filing](https://www.schemalaw.com/blog/searchable-pdf-for-court-filing) covers those.

Knowing the purpose before you convert tells you how much checking a document needs and which export to make. A production you only need to search can be converted in bulk and sampled. A paragraph you will quote to a judge gets read letter by letter.

Match the proofreading to the paper in front of you. A pleading you quote in a motion for summary judgment needs every date and name checked against the scan. Discovery you are only searching can be sampled. A trial exhibit needs the page image kept with the text. For an appeal, keep the printed page numbers so a record citation still lands on the right page.

## Pleadings and motions: the scanned papers that open a case

The first scans in a case are the opening papers themselves: a complaint handed to the client by a process server, a summons, an order from a prior related case, a motion mailed by a party without counsel. There are few of them, each matters, and you convert them to draft against and to quote.

OCR of pleadings and motions has a narrower aim than discovery OCR. You are not looking for a document. You have it. You want its words in a form you can work with.

### What to convert, and into what

| Paper | Why convert | Output |
| --- | --- | --- |
| Complaint served as paper or a scan | To draft an answer that responds to each numbered paragraph | Word |
| Opposing motion and memorandum received as a scan | To quote and respond to specific passages | Searchable PDF for reading; plain text for quoting |
| An order you must comply with | To pull deadlines and operative language | Searchable PDF; Word if you will draft a proposed order from it |
| Scanned exhibits to the other side's motion | To check whether the exhibit says what the brief claims | Searchable PDF |
| Prior-case filings the client kept in a box | To learn the history | Searchable PDF |

The conversion procedure for a single order or motion, including caption blocks and line-numbered paper, belongs to the guide on [converting an order or motion to editable text](https://www.schemalaw.com/blog/ocr-court-filings#converting-an-order-or-motion-to-editable-text). The point here is where it sits in the life of the case: at intake, on the day the paper arrives, before anyone starts drafting.

### A worked example

Fictional facts: Tidewater Marine Supply is served with a 34-page complaint by Delgado Boatworks. The client scans it on an office copier and emails the PDF. A search for "warranty" finds nothing, so the file is images.

1. The paralegal saves the scan untouched as the received copy.
2. She converts it and exports a searchable PDF and a Word file.
3. In the Word file she checks that the paragraph numbers 1 through 87 are all present and in order, comparing against the image. Two numbers have been read as "l1" and "7l" and she corrects them.
4. The associate drafts the answer in a new document, responding paragraph by paragraph, and pastes each allegation from the Word file only where the answer needs to quote it.
5. Every quoted allegation is read against the page image before the answer is filed.

The time saved is in step 4. The risk is in step 5, and it is the reason the received scan is kept beside the working copy.

### Do not file your own papers as scans

The reverse case needs no OCR at all. A motion you drafted exists as a word-processing file. Convert it straight to PDF. If you print it, sign it and scan it, you have turned text into a picture and created an OCR problem for the court and for yourself.

## Which litigation documents arrive with text and need no OCR?

Any document that was created on a computer and stayed electronic already has text: papers converted to PDF from a word processor, transcripts delivered as electronic files, native productions, and productions that come with text files. Test each file before converting, because running OCR on good text wastes pages and can replace exact text with a guess.

By stage, these are the sources to test first.

| Stage | Source | Test before you decide |
| --- | --- | --- |
| Pleadings and motions | A filing downloaded from the court's electronic docket | Search for a word in the body. A filing can have a typed brief and scanned exhibits in one PDF. |
| Written discovery | Requests served by email as PDF | Select a sentence. If it highlights word by word, it has text. |
| Document discovery | A production delivered with a load file and a folder of text files | Open three text files and compare with their images. The text is already there. |
| Document discovery | Native files: spreadsheets, emails, word-processing files | Open one. Native files are text. |
| Depositions | A transcript delivered electronically by the reporting firm | Search for the witness's name on a page of testimony. |
| Trial | Exhibits you created: summaries, charts, demonstratives | These came from your own software. |
| Appeal | Orders and opinions downloaded from the docket | Search for a word on the last page as well as the first. |

Federal document production is one reason text arrives at all. [Rule 34(b)(2)(E)(ii) of the Federal Rules of Civil Procedure](https://www.law.cornell.edu/rules/frcp/rule_34) says that if a request does not specify a form for electronically stored information, "a party must produce it in a form or forms in which it is ordinarily maintained or in a reasonably usable form or forms". What that means for a particular production is a legal question. The practical step is the same either way: look at what was delivered before you plan any conversion.

### Three results the test can give

**Text throughout.** Do nothing. File the document and move on.

**No text anywhere.** Convert it.

**Text in places.** This is the one to watch. A docket download, a production volume or an appellate record is assembled from many sources. Search for a word on an early page, a middle page and a late page. If any of the three searches fails on a page where you can see the word, the file is mixed. Convert the image-only parts, or convert the whole file if separating them costs more time than it saves.

### When existing text is not good enough

A file can pass the selection test and still have poor text. Paste a line into a blank document. If names and numbers come out wrong, the existing layer is unreliable, and a search of that file will miss things. For a document you need to rely on, convert it again from the images and compare. The discussion of extracted text and OCR text in the post on [eDiscovery OCR](https://www.schemalaw.com/blog/ediscovery-ocr) explains why text taken from a native file is exact and text recognized from an image is not.

## Discovery, depositions and trial: where the page volume peaks

In a document-heavy case, page volume peaks in discovery, when the client's paper files, the other side's production and third-party records come in. Depositions add transcripts and marked exhibits. Trial preparation then narrows the set to the exhibits and testimony that will be used, and the standard of checking rises as the count falls.

### Document discovery

Three streams of paper meet here.

- **The client's own files.** Paper kept in boxes or scanned years ago without text. You convert these to review them and, if they are produced, to deliver them in the form requested. Rule 34(b)(2)(E)(i) says 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", so keep the original order of each box or folder when you scan.
- **The other side's production.** If it arrived as images with no text, convert it before anyone reviews a page.
- **Third-party records.** Returns on subpoenas arrive in whatever form the custodian chose.

At this volume the work is batch work: splitting large files, naming them so order is kept, tracking failures and counting pages. The guide to [batch OCR for discovery documents](https://www.schemalaw.com/blog/batch-ocr-discovery-documents) covers that. Check the count against your tool's capacity before you start. In Schema Law the per-document and monthly page limits differ by plan, and every plan lets you upload several files at a time; the [pricing page](https://www.schemalaw.com/pricing) has the current figures.

### Depositions

Each deposition adds a transcript and a set of marked exhibits. Test the transcript for text first. Where a marked exhibit is a document from the production with a sticker added, the copy you already converted may serve for searching, and the marked copy is the one to cite. The post on [making deposition transcripts searchable](https://www.schemalaw.com/blog/how-to-make-deposition-transcripts-searchable) covers the transcript side.

### Trial

By trial the set is small and each page counts. [Rule 26(a)(3)(A)(iii)](https://www.law.cornell.edu/rules/frcp/rule_26) requires, among pretrial disclosures in federal cases, "an identification of each document or other exhibit". Every item on that list should be searchable in your own working set so a passage can be found while a witness is on the stand. Check your court's and judge's rules for the form of the exhibits you file or hand up.

### Converted text feeds the work product

The reason to convert at the peak is what the text lets you build afterwards: a document index, a privilege log, deposition summaries, an exhibit list and a timeline of events with a source page for each entry. The last of these is covered in the post on [how to build a case chronology](https://www.schemalaw.com/blog/how-to-build-a-case-chronology). None of them can be assembled from pictures of pages except by reading every one.

### Triage when you cannot convert everything at once

| Priority | What | Why first |
| --- | --- | --- |
| 1 | Anything with a response deadline running | The calendar decides |
| 2 | The documents the client says matter most | Early case assessment depends on them |
| 3 | The other side's production | Review cannot start without text |
| 4 | Third-party records needed for a scheduled deposition | The outline depends on them |
| 5 | Everything else, in the order it arrived | So nothing is skipped |

## Who on a litigation team runs OCR, and when?

Whoever receives a document should be responsible for seeing that it is tested and, if needed, converted before it is filed to the matter. Decide that by role at the start of the case and write it down. The timing rule is simple: convert at intake, once, before anyone works on the document.

Teams are built differently, so treat this table as a template to fill in for your matter, not a description of how any firm is organized.

| Task | Does it | Checks it | When |
| --- | --- | --- | --- |
| Test incoming files for text | The person who receives them | | On receipt |
| Convert served papers | Paralegal or assistant | Drafting attorney, against the image | Same day |
| Convert the client's paper files | Paralegal, litigation support or a scanning vendor | Paralegal, by sampling | Before review |
| Convert an incoming production | Litigation support or paralegal | Reviewer, by sampling | Before review |
| Convert third-party records | Paralegal | Attorney using them | On receipt |
| Convert transcripts and exhibits | Paralegal | Person writing the summary | On delivery |
| Verify quotations | The person who drafted the passage | A second reader | Before filing |
| Keep the log | One named person | Supervising attorney | Continuously |

### Rules that keep the step from being done twice or not at all

- **One owner per stream.** If two people can each assume the other converted the production, neither will.
- **Convert before filing to the matter.** Once an unconverted document is saved to the matter folder, nothing prompts anyone to go back to it until a search fails.
- **Keep the received copy.** Never overwrite the file as it arrived. Name the converted copy so the pair sort together.
- **Log it.** A sheet with the file name, source, date received, page count, whether it had text, who converted it and who sampled it answers every later question about a document's history.
- **Decide who may upload what.** Before a document goes to any online tool, someone must have checked whether a protective order, a client instruction or a confidentiality agreement restricts where it can be sent. Schema Law's [security commitments](https://www.schemalaw.com/#security) state how uploads are treated; whether that satisfies an order in your case is for the supervising attorney.
- **Do not let the tool have the last word.** Recognition is strongest on clean print and weaker on faint faxes, stamps over text and handwriting, and nothing marks a passage as uncertain. A person checks names, dates and amounts.

### A solo practitioner

With no team, the roles collapse into one person and the timing rule matters more. Convert when the document arrives. The alternative is discovering, the night before a response is due, that the thing you need to search is forty pages of pictures.

## A stage-by-stage litigation OCR checklist for one matter

Use this checklist as a matter template. At each stage it names what to test, what to convert, the output to make and the check to run. Copy it into the matter file when the case opens and mark each line as the papers arrive.

### Stage 1: case opening and pleadings

- Test the complaint, summons and any attached exhibits for text.
- Convert image-only served papers; export a searchable PDF and a Word file.
- Check paragraph numbering in the Word file against the image.
- Save the received copy, the searchable copy and the working copy together.
- Convert any orders from related or prior proceedings the client supplies.

### Stage 2: motions

- Test each opposing paper and each exhibit separately.
- Convert scanned exhibits before the response is drafted.
- Read every quotation against the page image before filing.
- Create your own filings directly from the word processor; scan only what exists only on paper.
- Check your court's format rules for the filed PDF.

### Stage 3: written discovery

- Convert scanned requests to Word so responses can be drafted under each request.
- Check request numbers and defined terms against the image.

### Stage 4: document discovery

- Scan the client's paper in its original order, with one file per document or per folder as the team has decided.
- Test every incoming production volume in three places.
- Estimate page counts and compare with your tool's limits.
- Convert, then sample: first, middle and last page of each file, plus any page with small print or tables.
- Log file name, page count, date and who checked.
- Note which documents are handwritten or too poor to convert, so reviewers know search will not find them.

### Stage 5: third-party records

- Number and log the set before anyone works on it.
- Convert statements and ledgers; export tables where figures will be used.
- Check totals and dates against the image.

### Stage 6: depositions

- Test each transcript for text on delivery.
- Convert scanned transcripts and exhibits.
- Note the difference between PDF page numbers and printed transcript page numbers.
- Cite from the image.

### Stage 7: trial preparation

- Confirm every exhibit on the list is searchable in the working set.
- Re-convert any exhibit whose text fails a spot check.
- Keep stickered and unstickered copies distinguishable by file name.

### Stage 8: appeal

- List what the appendix must contain. In federal appeals, [Rule 30(a)(1) of the Federal Rules of Appellate Procedure](https://www.law.cornell.edu/rules/frap/rule_30) names "the relevant docket entries in the proceeding below", "the relevant portions of the pleadings, charge, findings, or opinion", "the judgment, order, or decision in question" and "other parts of the record to which the parties wish to direct the court's attention". Your court's local rules add format requirements.
- Test each component. Trial court filings can be a mix of text and scanned pages.
- Convert image-only components before assembling.
- Follow the guide to building an [appellate appendix PDF](https://www.schemalaw.com/blog/appellate-appendix-pdf) for pagination, bookmarks and volumes.

### At every stage

| Question | Answer before moving on |
| --- | --- |
| Was the file tested for text? | Yes, in more than one place |
| Is the received copy preserved? | Yes, unaltered |
| Was the output chosen for the purpose? | Search, quote, edit, tabulate or file |
| Has a person checked names, dates and amounts? | Yes, against the image |
| Is the conversion logged? | Yes, with date and initials |
| May this document go to this tool? | Checked against orders and client instructions |

## 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)