# OCR for Law Clerks: Scanned Records and Old Opinions

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2023-10-30T10:01:46Z
- Updated: 2026-10-11T19:22:20Z
- Category: Choosing legal OCR
- Canonical: https://www.schemalaw.com/blog/ocr-for-law-clerks

> How law clerks can use OCR on scanned trial records and old opinions: whose permission to get, how to search a record, and what to verify for the judge.

OCR for law clerks turns the scanned trial records, exhibits and older opinions on your desk into text you can search and quote. Get permission from your judge or supervising attorney before you upload anything, use the converted text to find passages, and read every quotation against the page image before it reaches a bench memo.

This post covers two kinds of clerk: the judicial clerk in chambers and the law clerk or summer associate at a firm. The documents overlap, the permission rules do not, and in both seats you are the junior reader whose work someone senior will rely on without rereading the source.

In chambers, ask the judge before a trial record or a deposition transcript leaves the court's systems. At a firm, ask the assigning attorney the same question before you upload pleadings from a closed file. Whether you are a judicial clerk or a firm law clerk, quote from the image and keep your notes inside the systems your supervisor already approved.

## Which documents on a law clerk's desk arrive as scans?

The documents on a law clerk's desk that arrive as scans are mostly things nobody typed recently: paper exhibits admitted at trial, older transcripts, agency records, handwritten or self-represented filings, and opinions from before courts published electronically. Briefs and recent orders usually carry real text. The record behind them often does not.

### In chambers

A judicial clerk reads the parties' briefs as text and then follows their citations into a record that is a mix. In a federal appeal, the [record on appeal under Appellate Rule 10](https://www.law.cornell.edu/rules/frap/rule_10) is made up of the original papers and exhibits filed in the district court, the transcript of proceedings, if any, and a certified copy of the docket entries. Each of those can be a scan:

- Trial exhibits that were paper in the courtroom and were imaged afterward, sticker and all.
- Transcripts from older proceedings, or transcripts that reached the file as a printed copy that someone scanned.
- An administrative record from an agency, often thousands of pages copied from a paper file.
- Filings by people without lawyers, written by hand or typed and then photographed.
- State court records lodged in a federal habeas matter, copied more than once before they arrived.

### At a firm

A law clerk or summer associate at a firm sees the same paper from the other side. The assigning attorney hands over a closed file to mine for a brief the firm wrote years ago, a set of records the client sent as phone photos, an opposing party's production, or a stack of unpublished orders collected from a courthouse file room. The research memo is due Friday and none of it can be searched.

### A quick way to sort the pile

Before you plan your week, open each file and try to select one word on a page near the middle. If you cannot, the page is an image, for the reasons in the post on [why you can't copy text from a PDF](https://www.schemalaw.com/blog/why-cant-i-copy-text-from-a-pdf). Make a two-column list: files with text, files without. The second column is the part of the assignment where OCR changes how long the work takes.

## Is a law clerk allowed to upload court or firm documents to an outside tool?

A law clerk is allowed to upload court or firm documents to an outside tool only when the person the clerk works for has said so. In chambers that is the judge. At a firm it is the supervising attorney, acting under the firm's policy. Your own judgment that a tool looks safe is not the permission.

### Judicial clerks

Chambers information is confidential in ways a new clerk can underestimate. The federal [Code of Conduct for Judicial Employees](https://www.uscourts.gov/sites/default/files/guide-vol02a-ch03-2.pdf) says in Canon 3D that a judicial employee should never disclose any confidential information received in the course of official duties except as required in the performance of such duties, and it applies the same restriction to former employees. State courts have their own codes, and a judge may add chambers rules on top.

For a clerk, the practical questions are narrower than the canon:

- Is the document public? A filed, unsealed exhibit is a different thing from a draft opinion, a bench memo or a sealed filing. Drafts and internal memos should not leave court systems at all unless your judge says otherwise.
- Does the court provide a tool? A court's IT office may have decided which software staff may use. Ask the career clerk or the court's IT contact before you ask the internet.
- Does the upload itself reveal something? Even with a public record, which pages chambers is studying can hint at what the judge is thinking about.

Check with your judge or the chambers' senior staff before the first upload, and take the answer as covering only the kind of document you asked about.

### Firm clerks and summer associates

At a firm, client files are the firm's to protect, and the lawyers answer for what you do with them. Ask the assigning attorney which tools are approved, and read the firm's written policy if it has one. The post on [writing a law firm AI policy for document uploads](https://www.schemalaw.com/blog/law-firm-ai-policy-document-upload) shows what such a policy usually covers, which helps you know what to ask. The underlying duties are explained in the post on [attorney-client privilege and AI tools](https://www.schemalaw.com/blog/attorney-client-privilege-and-ai-tools); that analysis belongs to the lawyers, not to you, but you should know it exists.

Do not sign up for a personal account on your own and run client documents through it to save time. If the firm has no approved tool, say so to your supervisor and let them decide.

### What to bring when you ask

A supervisor can answer faster if you bring facts. For any tool, find its published statements on three points: whether uploads are used to train a model, who else can see them, and how deletion works. Schema Law, for example, lists the same [security commitments](https://www.schemalaw.com/#security) for every plan, among them "Your documents are never used to train an AI model", "Documents stay inside your own account", "Encrypted in transit and at rest" and "Delete any document, at any time". Hand over the vendor's own page and let the judge or attorney make the call.

## Searching a scanned trial record for a bench memo

Searching a scanned trial record for a bench memo means converting the volumes the briefs cite, running searches for the witnesses, exhibits and phrases in dispute, and then reading the hits on the page image. Search tells you where to read. It does not tell you what the record says, and it does not prove a thing is absent.

### A worked example

Suppose you are assigned *Marlow v. Tessier Hauling*, a fictional appeal from a jury verdict. The appellant argues that the trial court admitted a maintenance log without foundation. The record is nine PDF volumes, about 2,100 pages, and volumes 4 through 7 are scanned trial transcript.

1. **List what the briefs cite.** Pull every record citation from both briefs into a table: volume, page, the proposition it is cited for.
2. **Convert the volumes that are images.** Export each as a searchable PDF so the page images, and the record page numbers stamped on them, stay exactly as filed.
3. **Search around the issue, not only at the citations.** Search "maintenance log", "Exhibit 22", "foundation", "objection", and the name of the witness who sponsored the exhibit. Parties cite the pages that help them. The ruling you need may sit forty pages from either side's citation.
4. **Search for likely misreadings.** If "Tessier" returns twelve hits in a four-day trial about Tessier's trucks, something was misread. Try "Tessler" and "Tessicr". Try "Ex. 22" and "Exhibit 2 2".
5. **Read each hit on the image.** Open the page, read the surrounding testimony, and note the record page from the stamp on the image, not from the PDF viewer's page counter.
6. **Record what you searched.** Keep the terms and the volumes in your notes so the memo can say how you looked.

### Reporting a negative

The hardest sentence in a bench memo is "the record contains no objection." When part of the record was scanned, write it so the judge knows its basis: which volumes you searched, with which terms, and which stretches you read page by page because the scan was poor. If the issue turns on the absence, read the relevant testimony in full. A misread word on a faint page is invisible to search.

### Page numbers

Records carry several numbering systems at once: the transcript's own pages, a stamped record or appendix number, and the position in the PDF. Decide at the start which one your court's citation practice uses, and take it from the image every time. For transcripts in particular, the post on [making deposition transcripts searchable](https://www.schemalaw.com/blog/how-to-make-deposition-transcripts-searchable) explains how page and line numbers behave after conversion. For scanned pleadings and orders, see the guide to [OCR for court filings](https://www.schemalaw.com/blog/ocr-court-filings).

## Pulling language from scanned opinions and older reporter volumes

Pulling language from scanned opinions and older reporter volumes works when you treat the converted text as a draft to correct. Old volumes were typeset in narrow columns with footnotes, running heads and worn type, and each of those leaves marks in the output that you must clear before the passage is quotable.

Most of your research will come from a database with clean text. OCR matters for what the databases lack or carry only as page images: slip opinions and unpublished orders from a courthouse file, older state reports, session laws, legislative journals, treatises long out of print, and the briefs filed in an old case.

### What goes wrong in old print

| Feature of the page | What it does to converted text | What to do |
| --- | --- | --- |
| Two columns | Lines from the left and right columns can interleave | Compare the first and last line of each column against the image |
| Headnotes and syllabus | Editor's summary runs straight into the court's words | Mark where the opinion itself begins before you copy anything |
| Running heads and page numbers | "SMITH v. JONES 417" lands mid-sentence | Delete them, and keep the page break location for your pin cite |
| Words hyphenated at line ends | "judg- ment" stays split | Rejoin, then check whether the hyphen was part of the word |
| Footnotes | Note text merges into the paragraph above it | Separate notes from body before quoting either |
| Worn or broken type | "c" for "e", "h" for "b", "1" for "l" | Read names, numbers and citations letter by letter |

The post on [whether OCR can handle two-column documents](https://www.schemalaw.com/blog/can-ocr-handle-two-column-documents) goes further into reading order.

### Keep the image with the text

Save the scanned pages you relied on next to your notes. When a judge or partner asks where a quotation came from, you want the picture of the page, with the volume and page number visible, and not a text file that says what software thought the page said.

Quotation mechanics, including pin cites and how to mark an alteration, are covered in the guide to [quoting text from a scanned PDF](https://www.schemalaw.com/blog/how-to-quote-text-from-a-scanned-pdf).

## What should a law clerk verify before converted text reaches a judge or partner?

A law clerk should verify every quotation word for word against the page image, every page citation against the number printed on the page, and every statement that something is missing from the record. The reader above you will assume those three things were checked, because checking them is the job.

Use this before a memo or draft leaves your hands:

- **Quotations.** Read each one aloud against the image, including punctuation and capitalization. Converted text that reads smoothly can still be wrong.
- **Negations.** Look hard at "not", "no", "never" and "un-". A dropped "not" reverses a holding and still reads as good English.
- **Names and numbers.** Parties, witnesses and judges, spelled as the source spells them, and every date, dollar amount, statute section, sentence length and docket number. The general method is in the post on [proofreading OCR output against the original](https://www.schemalaw.com/blog/why-is-ocr-output-full-of-errors#proofreading-ocr-output-against-the-original).
- **Citations inside the passage.** Volume and page numbers in an old opinion are exactly the characters worn type garbles.
- **Page breaks.** Know where the quoted passage crosses a page, so the pin cite is right.
- **Whose words they are.** Majority, dissent, headnote, or a party's argument the court was summarizing.
- **Negative findings.** State what you searched and what you read.

### Say what you did

Add a line to the memo or the cover email: "Volumes 4 to 7 of the record are scans. I located passages by search and confirmed each quotation against the page." That sentence tells your reader how much to trust and where to look if something seems off. It also protects you.

### Do not tidy the source

If the opinion has a typo, or the witness said something ungrammatical, the quotation keeps it. Fixing a source without marking the change is an alteration, even when the fix came from software.

Students who learn these habits on a journal or in a clinic arrive at the clerkship ahead. The post on [OCR for law schools](https://www.schemalaw.com/blog/ocr-for-law-schools) covers that earlier stage.

## Choosing an OCR plan for a one-year clerkship or a summer position

Choosing an OCR plan for a one-year clerkship or a summer position starts with a question that is not about price: whose tool are you supposed to use? A court or firm that provides one may also have decided you may not use another. Buy your own only for work you are cleared to put in it.

### If the court or firm provides a tool

Use it, even if it is clumsier than something you tried in school. The documents belong to the court or the client, and so does the choice. If the provided tool fails on a particular scan, report that and ask what to do next.

### If you are choosing for yourself

A clerk's volume is uneven: nothing for two weeks, then a 2,000-page record. Count pages before you pick. Using the Schema Law plans as the example:

| Your situation | Plan that fits | Limits to know |
| --- | --- | --- |
| Occasional short public documents: a three-page order, a few pages of an old opinion | Free | 1 document a day, up to 3 pages each, every export format |
| Regular record work in a busy month, cleared by your supervisor | Solo, listed at $49 | 1,000 pages a month, up to 250 pages per document, every export format |
| Several clerks sharing records | A decision for chambers or the firm, not for you | Each account is for one person, so each clerk needs their own; the Team plan allows 10,000 pages a month |

The [pricing page](https://www.schemalaw.com/pricing) has the current figures. Three practical points follow from those limits.

A transcript volume often runs past 250 pages, and uploads are capped at 20 MB per file. Split a long volume at natural breaks, such as a day of trial, and name the parts so the record page ranges are obvious.

Sign-in is through a Google account. If your court or firm does not issue one, ask which account you are permitted to use before you create anything.

No plan deletes documents for you. A document stays in your account until you delete it, so export what you need and delete the upload when the assignment is finished, without waiting for the end of the term.

### When the term ends

A clerkship or summer has an end date, and your copies should not outlast it. Before you leave, delete every document you uploaded, hand converted files that belong in the case file to whoever keeps it, and close any account you opened for the job. Tell your supervisor it is done.

## 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)