# OCR for Paralegals: A Daily Workflow That Holds Up

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2026-10-10T02:36:23Z
- Updated: 2026-10-11T19:22:20Z
- Category: Choosing legal OCR
- Canonical: https://www.schemalaw.com/blog/ocr-for-paralegals

> A daily OCR workflow for paralegals: an intake routine for scans, which export fits each task, what to verify every time, and how to hand work to an attorney.

OCR for paralegals works best as a fixed routine: convert every scan at intake, pick the export by the task it feeds, verify names, dates and amounts against the page image, and tell the attorney what was checked. Done that way, converted text can be searched, quoted and tabulated without anyone retyping it or trusting it blindly.

The paralegal, legal assistant or legal secretary is usually the first person to touch a scanned document and the last to check it before an attorney relies on it. This post is written for that seat. It gives a routine you can run every day, a format guide, a handoff note and the words to use when someone expects more from OCR than it can deliver.

Build the routine around the work you hand an attorney. At client intake, convert the new file and log it in matter management before you draft. For discovery, export tables you can total. For a deposition transcript, keep the page and line numbers you will cite. Add the minutes to your timekeeping entry so the checking work is visible on the bill.

## Where does OCR save a paralegal the most time?

OCR saves a paralegal the most time on four jobs: finding things in large sets of scanned pages, lifting language out of documents nobody in the office wrote, moving figures from scanned tables into spreadsheets, and building chronologies. Each one replaces reading or typing with searching and checking.

**Finding.** A records set or a production that arrives as images can only be read page by page. Once it has text, you can search every page for a name, a date or a claim number in one pass. The gain grows with the size of the set.

**Lifting language.** Discovery responses restate the requests. A proposed order tracks the motion. An amendment quotes the agreement. When the source is a scan, each of those begins with someone typing out another person's words. A Word export gives you the words to start from.

**Tables.** Billing ledgers, bank statements and payment histories are grids. Keying them into a spreadsheet by hand is slow and invites transposed digits. Exporting the table and then reconciling the totals is quicker and easier to audit. The guide to [getting a scanned PDF into Excel](https://www.schemalaw.com/blog/scanned-pdf-to-excel) covers how tables come across.

**Chronologies.** A chronology is built from dated events scattered through hundreds of pages. Searchable text lets you jump to each date and each provider instead of paging through the whole file.

### Where it does not save time

OCR does not help much with a short, clean document you only need to read once. It helps less than people hope with handwriting and with heavily degraded faxes, where the checking can take as long as typing. Know those cases so you can say so up front. The post on [OCR for handwritten notes](https://www.schemalaw.com/blog/ocr-handwritten-notes) explains what to expect from handwriting.

## A daily intake routine for scanned documents

A daily intake routine for scanned documents has six steps: save the original, test for text, convert, verify, file both versions, and log what you did. Run it when the document arrives, not when someone asks for it, so that every file in the matter folder is already searchable on the day it is needed.

1. **Save the original untouched.** Name it by date received, source and description, such as `2026-10-08 Def production vol 2 (as received).pdf`. This file is never edited, renamed or replaced.
2. **Test for a text layer.** Try to select a word and search for one you can see. Check a page from the middle and one from the end as well, because assembled files often mix typed and scanned pages.
3. **Convert what needs it.** Upload the file and wait for the pages to be read. Note the page count the tool reports and compare it with the original.
4. **Verify the parts that carry weight.** Use the checklist later in this post. For a routine letter this takes a minute. For a contract or a ledger it takes longer and is worth it.
5. **File the export beside the original.** Use the same name with a suffix such as `(OCR)` so the pair sorts together and nobody opens the wrong one.
6. **Log it.** One line in the matter's document log: date, file, pages, export made, what you verified, anything unreadable.

### Sizing the routine to the day

On a normal day the routine adds a few minutes per document. On the day a large production lands, do steps 1 and 2 for everything first, so the originals are safe and you know the scale. Then convert in order of need. The guide to [batch OCR for discovery documents](https://www.schemalaw.com/blog/batch-ocr-discovery-documents) covers splitting and tracking a set that size.

### Where the routine usually breaks

It breaks at step 6. The log feels optional until an attorney asks whether page 212 was checked. Keep the log short enough that you will actually fill it in. A spreadsheet with six columns is enough.

It also breaks when the "original" gets overwritten by a searchable copy with the same name. Keep the suffix rule strict.

### If the set carries Bates numbers

Check that the numbers survive conversion and still match their pages before anyone cites them. The post on [how to OCR Bates numbered documents](https://www.schemalaw.com/blog/how-to-ocr-bates-numbered-documents) explains the order of operations.

## Which export should a paralegal pick for each task?

A paralegal should pick the export by what happens to the document next. Searchable PDF is the default for anything you will keep and search. Word is for language you will edit. Excel or CSV is for tables. Plain text is for quoting and comparing. JSON, Markdown and HTML serve other systems.

| Task | Export | Why |
| --- | --- | --- |
| File a received document so it can be searched later | Searchable PDF | The page image stays as it arrived, with text behind it |
| Draft responses, an amendment or a proposed order from a scan | Word | You get editable paragraphs to work from |
| Total a billing ledger or bank statement | Excel | Each table lands on its own sheet as rows and columns |
| Feed a table to another program | CSV | Nearly every database and review tool can import it |
| Paste a passage into a brief or an email | Plain text | No formatting to strip out |
| Compare two versions of a document | Plain text | Comparison tools work best without layout noise |
| Hand structured content to a developer or an internal system | JSON | Pages, text and tables arrive in a predictable shape |
| Put content on an intranet or in a notes system | Markdown or HTML | Headings and tables carry over as structure |

Schema Law offers those eight formats, and the site's [export formats section](https://www.schemalaw.com/#formats) describes the main ones. All eight are available on every plan, the free one included, so the choice depends on the task and not on what the office pays for.

### Two rules that prevent rework

Make the searchable PDF every time, even when the task calls for Word or Excel. It is the copy that shows what the page looked like, and it is the one someone will want six months later.

Do not choose Word for a document that is mostly a table. You will spend your time fixing column widths. Export the table as a spreadsheet and the surrounding text as Word if you need both. You can export the same upload more than once.

### When you are not sure

Ask what the attorney will do with it, not what format they want. "I need to find every time the adjuster is mentioned" means searchable PDF. "I need to respond to these" means Word. People name the format they know, which is not always the one that fits.

## Handing converted documents to an attorney

Handing converted documents to an attorney means giving three things together: the original scan, the converted file, and a short note saying what you verified and what you could not read. The note is what lets the attorney decide how far to rely on the text without redoing your work.

A handoff note can be five lines. Copy this and fill it in.

```text
Document:   Smith lease, 14 pages, received 2026-10-06 by email
Original:   2026-10-06 Smith lease (as received).pdf
Converted:  2026-10-06 Smith lease (OCR).docx
Verified:   party names, dates, rent amounts, section numbers on pp. 1-14
Not legible: handwritten initials p. 9; fax header covers first line of p. 12
```

### What to say plainly

State the scope of your check. "Verified all dollar amounts" and "spot-checked" are different statements, and the attorney needs to know which one applies.

Flag anything you changed. If you corrected a misread name in the Word file, say so and say where. A silent correction is still an edit to someone else's document.

Point to the pages that were hard to read. A stamp across a paragraph, a faint fax line, a margin note. The attorney may decide to request a better copy.

### What not to do

Do not send the converted file alone. Without the scan beside it, nobody can settle a doubt about a word.

Do not describe a converted document as "the" lease or "the" order. It is a working copy. Anything quoted in a filing should be read against the page image by whoever signs it, or by you with that instruction recorded in the note.

### Keeping the pair together

Use one folder per document set, with originals and conversions side by side, or a single folder with the suffix rule. Either works. What fails is a separate "OCR" folder that drifts out of step with the originals.

## What should a paralegal double-check every time?

A paralegal should double-check six things on every conversion: page count, names, dates, numbers, section references, and anything near a stamp, signature or handwriting. Those are the places where one wrong character changes the meaning and where recognition fails most often.

- **Page count.** The export should have as many pages as the scan. A missing page is the most serious error and the fastest to catch.
- **Names.** Parties, witnesses, providers, judges. Unusual spellings are where software guesses.
- **Dates.** Check day, month and year. A 3 read as an 8 moves a deadline.
- **Amounts and other numbers.** Dollar figures, account numbers, case numbers, Bates numbers. Check decimal points and commas.
- **Section and paragraph references.** "Section 12.1(b)" and "Section 12.l(b)" look alike on screen.
- **Edges and overlaps.** Text under a stamp, beside a signature, in a header or footer, or written by hand.

### Add a check for text that reads too well

Newer tools produce smooth, well-formed output, and an error in it can look like a correct sentence. Read amounts and names against the image even when the paragraph around them is perfect. The post on [AI OCR for lawyers](https://www.schemalaw.com/blog/ai-ocr-for-lawyers) explains why this kind of error occurs.

### How much to check

Match the effort to the use. A document you only need to search can tolerate a few wrong characters. A passage going into a filing cannot tolerate one. For tables, add up a column and compare it with the printed total. If they match, the column is very likely right.

The full method, including how to compare page and text efficiently, 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).

### A habit worth forming

Verify before you file, not when asked. A check made at intake takes a minute because the document is in front of you. The same check made three months later means finding the scan, reopening the export and working out which version is current.

## Explaining OCR limits to the rest of the team

Explaining OCR limits to the rest of the team comes down to three statements: converted text is a working copy, a search that finds nothing does not prove the word is absent, and some pages will not convert well. Say them early and the same way each time, so nobody is surprised near a deadline.

### "Searchable" does not mean "complete"

If a name was misread on page 80, searching for the name will not find page 80. When an attorney asks whether a term appears in a set, answer with what you searched and how. "No hits for 'Hendricks' or 'Hendrix' across 1,400 pages, converted from fax-quality scans" is an honest answer. "It's not in there" is not.

For important searches, try likely misreadings as well: a lowercase L for a one, "rn" for "m", a zero for the letter O.

### Some pages need a person

Handwritten notes, pages copied many times over and text under stamps may come out partly or wholly wrong. Say which pages those are instead of passing along output you do not trust. The post on [why OCR output is full of errors](https://www.schemalaw.com/blog/why-is-ocr-output-full-of-errors) is a useful link to send a colleague who wants the reasons.

### The converted file is not the record

The scan is what was received. The export is a tool for working with it. Cite, produce and file from the original unless the attorney directs otherwise.

### A short script for common requests

| Someone says | You can say |
| --- | --- |
| "Just OCR it and send me the Word file." | "Will do. I'll send the scan with it and note what I checked." |
| "Search the production for X." | "I'll search for X and the likely misreadings, and tell you which volumes were poor scans." |
| "Is this table right?" | "The column totals match the printed totals on every page except 14, which I've flagged." |
| "Can it read the doctor's notes?" | "Typed sections, yes. The handwritten ones I'll need to read against the page." |

If you work for a single lawyer and handle intake alone, much of this handoff happens in one head. The post on [OCR for solo attorney practices](https://www.schemalaw.com/blog/ocr-for-solo-attorney) has a shorter version of the routine for that case.

## 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)