How to edit a scanned PDF: what you can change after OCR

OCR adds an invisible text layer to a scan for search and copying, but the letters in the picture stay a picture. Add notes on top of the page, replace a value with a redaction plus new text, and do larger rewrites in Word.

Need to fix a date in a scanned contract or add a reference number to a certificate? Open the scan in Edit PDF: the same editor can recognize the text, add new wording and cover the old one. It helps to know up front what OCR does not do. It adds an invisible layer of recognized words to the page, but the printed letters in the image stay part of the image and do not change by themselves.

This article is about scans and photos of documents. If the PDF was created in Word or another program and contains real text, editing works differently: see how to edit PDF text without converting to Word.

Find out what is on the page

Keep a copy of the original file first. Then open the document in the editor and click "OCR" on the toolbar. Hovering over it shows the full name, "Recognise text (OCR)", and on a phone the toolbar opens with the "Tools" button. Before anything runs, the panel checks the pages. The "Pages without text" option shows how many were found, and in "Choose pages" mode every page number carries a tag: "scan", "has text", "drawing" or "empty".

A "has text" tag on a scanned page means the file already has a text layer, for example one added by the scanner or by the app the document was saved from. Check its quality. Open the file in an ordinary PDF viewer, copy a line with a date or number into any text editor and compare it with what you see on the page. If the digits match, there is no need to recognize the page again.

Recognize text right in the editor

The main route keeps you inside the editor:

1. Click "OCR". 2. Choose the pages: "This page", "Pages without text", "All pages" or "Choose pages". 3. Under "Language", English is preselected; switch to "Automatic" only if the document also contains Russian text. 4. Under "Quality", choose "Accurate (slower)" if the scan has small print, tables or stamps. 5. Click "Recognise".

You can close the window and keep editing: the layer is added when processing finishes. After that, search finds the words on the scan, and "Undo" removes the recognized layer in one step. Recognition uses operations from your limit, and the panel shows how many before you start.

The separate OCR PDF tool does the same for a whole file. It is handy when you do not need to change anything and only want search and copying. Quality modes and deskewing are covered in detail in the OCR guide, so we will not repeat them here.

Which languages are recognized

Recognition currently knows two languages: English and Russian. The "Automatic" choice in the editor and the automatic option of "OCR language" in the OCR PDF tool do not guess the language; they read the page with both at once. That suits documents where Russian text is mixed with English names and codes.

Scans in other Latin-script languages, such as German, French or Spanish, are read with the English model when "English" is selected. Words are mostly readable, but accented letters like ä, é or ñ get lost or distorted. In "Automatic" mode, Latin letters may be recognized as look-alike Cyrillic ones. Scripts other than Latin and Cyrillic are not recognized at all. OCR will not help with such documents for now, while adding text and redaction from the sections below work without it.

If the scan already has a poor OCR layer

Running OCR again does not fix someone else's mistakes. A page that already has a text layer of at least 24 characters is skipped. In the panel it lands in the line "already have text, left alone", and the OCR PDF tool returns it unchanged. This protects real text, but for a scan with an old, faulty recognition it means a new run will not correct anything.

The workaround is to turn the pages back into plain pictures:

1. Run the file through PDF to JPG. The result arrives as a ZIP archive with one picture per page. 2. Unzip it and upload the pictures in order to JPG to PDF. To avoid a white frame, set "Page margin (pt)" to 0. If the original was A4, choose A4 under "Output page size". 3. Recognize the new file.

In one run, JPG to PDF accepts a limited number of pictures (the limit is shown on the tool page). Process a long document in parts and join them with Merge PDF. The workaround takes three operations from your daily limit, and more when you work in parts: one JPG to PDF run per part plus one for the merge.

Keep in mind that real text objects, links and bookmarks disappear together with the old layer. Use this route only for a document that consists entirely of scans.

Add a note on top of the page

When the original content must stay and you only need to add something next to it, such as an incoming number, a comment or a clarification like "Starts at 2 pm", click "Text" (its tooltip reads "Add text"). Place the box in an empty area, set "Font size" and color, then zoom in and make sure the note does not overlap the lines of the scan.

The "Background" field puts a fill under the note, white for example. This is only a visual cover: the old letters under it stay in the file, and search still finds their recognized copies. It is not a way to remove data.

Replace a fragment with a redaction

If the old value really has to go, click "Mask" (its tooltip reads "Add redaction"). A white area appears on the page: drag it over the old date, resize it with the handles on its edges, then place the new date on top with "Text". After "Apply changes" the finished PDF appears below, ready to download. Your original file is not changed.

This method has a price. A page with a redaction is exported entirely as a picture, so nothing is left under the covered area. But the text layer disappears from the whole page: its text can no longer be selected or found by search. The text you added on that page is also burned into the image and stops being text. Links outside the area stay, while the document's properties, attachments and bookmarks are not cleaned. More on hiding data in how to redact sensitive data in a PDF.

To bring search back, recognize the result once more: open the downloaded file in the editor or in OCR PDF. Pages that still have text are skipped, while the redacted pages are now plain pictures and get recognized again, this time with the new date.

For larger edits, move to Word

Rewriting paragraphs and tables on top of a scan is awkward. In that case, recognize the document first, then open PDF to Word and keep "Editable text" under "Result type". The "Exact copy (not editable)" option inserts pages as pictures, so there is nothing to edit in it.

Untick "Include images". In a scan the page itself is a picture inside the PDF, so with this box ticked the DOCX gets both the page image and the recognized paragraphs, and the text ends up in the file twice. Without images you get clean text. Stamps and signatures will not come along either: if you need them, insert them into Word separately.

Recognition does not restore the original fonts, line breaks or exact table layout. After editing, save the document from Word as a PDF and compare it with the original page by page: amounts, dates, numbers and names. A short note is easier to add directly in the editor, while rebuilding in Word pays off when a noticeable part of the text changes.