# How to Scan a Document and Change the Text

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2026-10-09T01:52:53Z
- Updated: 2026-10-11T19:22:20Z
- Category: Converting scans
- Canonical: https://www.schemalaw.com/blog/how-to-scan-a-document-and-change-the-text

> How to scan a document and change the text when only paper survives: convert the scan to Word, proof it against the page, then edit a clean copy.

To scan a document and change the text, scan the paper to a PDF, run OCR to convert that scan into a Word file, and proof the Word file against the page before you alter a word. The scan is a picture and stays as it is. Your changes go into the rebuilt copy.

This is the situation where the paper is all there is. The firm's standard letter was typed on a computer that was retired years ago. A client brings in a printed policy and wants it updated. A precedent survives as a photocopy in a ring binder. In each case the goal is a clean, current file that can be edited from now on.

## Can you scan a document and change the text on it?

Yes, but not on the scan itself. A scan is a photograph of the page, with no letters in it to retype. To change the text you convert the scan into a document made of real characters, with OCR, and edit that. The result is a new file that reproduces the paper's wording.

You are not editing the old document. You are building a replacement for it and then changing the replacement. Three consequences follow.

- **The scan remains the record of the paper.** Keep it untouched and named as a scan, so there is always something to compare against.
- **The rebuilt file starts out unverified.** Software typed it. It needs proofing before it can stand in for the paper.
- **Appearance is rebuilt, not copied.** Fonts, margins and letterhead have to be set up again in Word. The scan does not hand them over.

There is another route for small changes, which is to patch the picture inside a PDF editor. It suits a single word or date on a page that must keep its look. For how to choose between the two, see the comparison of [editing in the PDF versus editing a Word conversion of the scan](https://www.schemalaw.com/blog/editing-scanned-pdf#editing-in-the-pdf-versus-editing-a-word-conversion-of-the-scan). When the aim is a clean new version, conversion is the route.

### Which scans convert well enough to rebuild from

Typed or printed pages in good condition give the best starting point. Faint photocopies, pages with handwriting between the lines and faxed copies give a rougher one, and the proofing takes longer. Schema Law's homepage puts it this way for its own service: recognition is strongest on clean print and weaker on faint faxes, stamps over text and handwriting. Whatever tool you use, run two representative pages before you commit to the whole document.

## Rebuilding a paper document as a clean Word file when the original file is lost

Rebuild in seven stages: scan to PDF, convert to Word, save the conversion untouched, make a working copy, apply your own styles, proof the working copy against the scan, and freeze the proofed version as the new master. Changes to the wording start only after the master exists.

A worked example. Garrick & Lowe, a fictional two-partner firm, has a six-page client care letter that exists only as a printed copy in the office manual. The file it was typed in is gone. Ines, the office manager, has been asked to produce an editable version so the complaints paragraph can be updated.

1. **Scan to one PDF.** Ines scans the six pages on the office copier using the text preset, checks that all six are present and upright, and saves the file as `Client care letter 2017 (scan of paper master).pdf`.
2. **Convert to Word.** She runs the PDF through the firm's OCR tool and exports Word. With Schema Law that means uploading the PDF in the browser, reading the parsed pages on screen and choosing the Word export, which the [formats list](https://www.schemalaw.com/#formats) describes as editable text with its headings, lists and tables.
3. **Save the conversion exactly as it arrived.** `Client care letter 2017 (OCR conversion, unchecked).docx`. She does not open and tidy this file. It is the evidence of what the software produced.
4. **Make a working copy.** `Client care letter 2017 (rebuild in progress).docx`. All further work happens here.
5. **Apply the firm's styles.** She selects all, sets the firm's body style, then marks each heading with a heading style and each numbered paragraph with the firm's numbering. This removes whatever formatting the conversion guessed and replaces it with formatting the firm controls.
6. **Proof against the scan.** With the PDF beside the Word file, she reads the rebuild against the paper, page by page, and corrects recognition errors only. No updating yet.
7. **Freeze the master.** She saves the result as `Client care letter 2017 (rebuilt from paper, proofed IR 2026-10-06).docx` and marks it read-only. This file says what the paper said.

Now the complaints paragraph can be rewritten, in a further copy named for the new version, with tracked changes switched on so the difference from the 2017 wording is visible.

### Why the order matters

Styling before proofing looks backwards, since it means formatting text that still has errors in it. It is deliberate. Proofing is easier when headings look like headings and numbered paragraphs are numbered, because a missing paragraph 7 shows up as a gap in the sequence. And freezing the master before any change is made keeps two kinds of difference apart: places where the software misread the paper, and places where you chose to change the wording.

### What the conversion will not carry over

Expect to supply these yourself.

| Element on the paper | In the converted Word file | What to do |
| --- | --- | --- |
| Letterhead and logo | Absent, or recognized as stray text | Use the firm's current letterhead template |
| Signature | Absent, or stray characters | Leave a signature block; the page is signed afresh when issued |
| Page numbers and running footers | May appear as text in the body | Delete them from the body and add real Word footers |
| Automatic numbering | Numbers arrive as typed characters | Re-apply automatic numbering so later edits renumber correctly |
| Cross-references | Plain text such as "see paragraph 9" | Leave as text, or rebuild as Word cross-references if the document will keep changing |
| Handwritten amendments | Unreliable | Read them from the scan and decide whether they belong in the rebuilt text |

If the rebuilt document needs a signature image or a logo placed in it, the post on [scanning a signature into Word](https://www.schemalaw.com/blog/scan-a-signature-into-word) covers how those travel as pictures and how to place them.

## Matching the fonts, margins and letterhead of the paper original

Match the paper only as far as the document's purpose requires. Measure the margins with a ruler, choose a typeface from the same family at the size that gives the same number of words per line, and use your current letterhead. Aim for an exact facsimile only when the job is a reconstruction.

Decide first which of two aims you have, because they pull in different directions.

- **A current version.** The rebuilt document will be issued under today's branding. Apply the firm's present template and stop. Matching the old look would be wasted effort.
- **A faithful reconstruction.** The rebuilt document should look like the paper, for example a form that people recognize by its layout, or a reprint of a historical document for an internal archive. Then margins, typeface and spacing are worth measuring.

### Margins and page size

Take them from the paper, not from the scan on screen, because a scan can be scaled or cropped.

1. Check the sheet size. Letter and A4 differ by enough to change every line break.
2. Measure from the paper's edge to the first line of body text, the left edge of the text, the right edge of the longest line and the last line on a full page.
3. Enter those four values in Word's page setup.
4. Measure the paragraph indent and the gap between paragraphs, and set them in the body style.

### Typeface and size

You do not need to identify the original font by name. You need one that behaves alike.

1. Decide whether the original is serif or sans serif, and whether every character has the same width, as typewriter-style faces do.
2. Pick a candidate from the fonts your firm already uses.
3. Type out one full line from the middle of a paragraph and compare its length with the same line on paper. If your line is shorter, increase the size by half a point and try again.
4. When one line matches, check a whole paragraph. If the lines break at the same words as on the paper, the page count will follow.

Matching line breaks has a practical benefit beyond appearance. If people cite the document by page, a rebuild that keeps the same text on the same pages keeps those references valid.

### Letterhead

Do not crop the old letterhead out of the scan and paste it in as a picture unless a reconstruction truly calls for it. A scanned letterhead is soft at the edges, its colours have shifted, and its details may be out of date: a former partner's name, an old address, a superseded regulatory statement. Use the current letterhead file, and get it from whoever looks after the firm's templates.

### How close is close enough

Print one page of the rebuild and lay it over the paper original against a window or a light box. Where lines of text coincide, you have matched. Where they drift apart down the page, adjust line spacing. Stop when a reader would take the two for the same document at a glance.

## Turning a scanned form letter or precedent into a reusable template

To turn a scanned form letter or precedent into a template, rebuild and proof it first, then replace every detail that belonged to one client or one matter with a clearly marked placeholder, add drafting notes where a choice has to be made, and save it as a template file that opens as a fresh copy.

### Strip the specifics

A form letter that survives on paper is a letter that was sent to someone. Before it becomes a template, every trace of that someone has to go: names in every form they take, addresses, dates, amounts and reference numbers. The post on converting a signed agreement to Word sets out that search, and the check that follows it, under [turning a converted agreement into a reusable template](https://www.schemalaw.com/blog/convert-signed-agreement-to-word#turning-a-converted-agreement-into-a-reusable-template).

A letter hides its specifics in places a contract does not, and a search for names and figures will not find them.

- **Facts embedded in sentences**, such as "your purchase of the flat" or "following our meeting on Tuesday".
- **Dates written in words**, such as "the first of March" or "last Thursday".
- **Gendered pronouns and singular or plural forms** that were chosen for the original recipient.
- **The salutation and the sign-off**, which were pitched at one reader: a first name, "Yours sincerely" or "Yours faithfully", a reference line.
- **Enclosure lists and copy lines** naming documents and people from the old matter.

Read the finished template once more, asking of each sentence whether it would be true for every future recipient.

### Add placeholders that cannot be overlooked

Use one searchable placeholder style throughout, such as square brackets with capitals: `[CLIENT NAME]`, `[DATE OF ENGAGEMENT]`, `[FEE ESTIMATE]`. Avoid blanks and underscores, which are easy to leave unfilled and hard to search for. Where the drafter must choose between alternatives, show both and label them:

```text
[OPTION A: fixed fee] Our fee for this work is [AMOUNT] plus VAT.
[OPTION B: hourly] Our fees are charged at the hourly rates set out in Schedule 1.
```

The final check on any letter produced from the template is then a single search for `[`. Zero results means nothing was left unfilled.

### Record where the template came from

Keep a short provenance note with the template: the paper source, the date it was rebuilt, who proofed it, and the date the wording was last reviewed by a lawyer. The review date is the item a paper precedent needs most. A precedent rebuilt from a ten-year-old photocopy carries ten-year-old law. Converting it does not update it, and whether its content is still right is a legal review that sits outside the conversion job.

The letter in the example is an engagement letter. Leave placeholders for the client name, the scope and the fee agreement, and keep a conflict check note outside the letter body. If the paper includes a client confidentiality paragraph, keep that wording and swap only the names.

### Save it as a template

Save the file in Word's template format so that opening it creates a new, unnamed document instead of opening the master for editing. Store it where the firm's other templates live, and withdraw the paper copy from the manual, or mark it as superseded, so nobody scans it again next year.

## How do you know the rebuilt document says what the paper said?

You know only by comparing the rebuilt text with the scan before any deliberate change is made. Proof the conversion against the page images, correct recognition errors alone, and save that proofed state. From then on, every difference between your document and the paper is one you chose to make.

The proofing method itself, including the order of checks and techniques such as reading digits aloud, is set out under [proofreading OCR output against the original](https://www.schemalaw.com/blog/why-is-ocr-output-full-of-errors#proofreading-ocr-output-against-the-original). What is specific to a rebuild is the discipline around it.

### Keep three states apart

| State | File | What it proves |
| --- | --- | --- |
| The paper | The scanned PDF | What the original said |
| The proofed rebuild | Word file, read-only, named with the checker and date | That a person compared the text with the scan |
| The new version | A later Word file with tracked changes | Exactly what was changed, by whom |

If recognition fixes and wording changes happen in one sitting in one file, the middle state never exists, and nobody can later say whether "30 days" in the new version was a decision or a misreading of "90".

### Places a rebuild goes wrong without looking wrong

The method for proofreading OCR output covers figures, dates and operative words. Three further faults belong to a rebuild, because restyling the text can hide them.

- **Dropped lines.** A line at the very top or bottom of a page, or one overlapped by a stamp or a punch hole, can be absent with no gap to show for it once the text has been restyled. Count paragraphs per page against the scan.
- **Numbering.** When you re-apply automatic numbering, check that Word's numbers agree with the paper's. One paragraph merged into its neighbour shifts every number after it.
- **Defined terms.** A capital letter read as lower case turns "the Property" into "the property", and applying a style will not put it back.

Do not count on conversion output to carry warning marks. Schema Law's homepage says nothing marks a passage as uncertain, and tells users to check names, dates and amounts against the original. A clean-looking page is not a checked page.

### A second reader

For a document that will be issued to clients or relied on as a precedent, give the comparison to someone other than the person who rebuilt the file. The rebuilder chose the styles, rejoined the paragraphs and re-applied the numbering, so a paragraph that went missing during that work looks correct to them.

## Changing one page and putting it back among the scanned pages

To change one page and return it to a scanned PDF, rebuild that page in Word so it ends at the same words as the original, save the page as a PDF, and use a PDF program's replace-pages function on a copy of the scan. First decide whether a mixed file is appropriate at all.

### When this makes sense

It suits internal and unsigned documents: an office procedures manual where one page of extension numbers has changed, a training bundle, a draft pack. It does not suit a document that has been signed, served, filed or produced, where the scan is a record of a specific piece of paper. For those, read the post on [whether you should ever alter the scan of a signed or filed document](https://www.schemalaw.com/blog/editing-scanned-pdf#should-you-ever-alter-the-scan-of-a-signed-or-filed-document), and ask the responsible lawyer.

### The procedure

1. **Copy the scan.** Work on `Procedures manual (scan, p 9 replaced 2026-10-06).pdf`, never on the original scan.
2. **Extract and convert the page.** Convert page 9 alone to Word and proof it against the scan.
3. **Fit the page.** Set page size and margins to match the paper, then make your change. Confirm that the rebuilt page still starts and ends at the same words as the original page 9, so that pages 8 and 10 still join it mid-sentence correctly. If your change adds lines, tighten spacing slightly or accept that text will run on to a new page 9A.
4. **Save the page as a PDF.** One page, same sheet size as the scan.
5. **Replace.** In your PDF program, replace page 9 of the copy with the new page. Check the page count is unchanged.
6. **Mark the page.** Add a small footer to the new page, such as "Page reissued 6 Oct 2026", so a reader knows why it looks different.

### What a mixed file looks like

The replaced page will be sharp, level and bright among pages that are soft, slightly tilted and grey. That contrast is honest, and it is better left visible than disguised. The file is also now mixed in another way: the new page has real text that can be searched, while the scanned pages around it have none unless the scan was made searchable. If people will search the document, run the whole file through OCR and export a searchable PDF, so every page answers a search.

For a short document, compare the effort with rebuilding the whole thing. Six pages converted and proofed once give you an editable master for every future change. For a first trial of that, Schema Law's free plan takes one document a day of up to three pages with Word export included, and the [pricing page](https://www.schemalaw.com/pricing) lists the limits for longer documents.

## 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)