Scanned PDF accessibility: why OCR alone isn't enough, and what a scan needs

Scanned PDF accessibility starts with one fact: a scan is a photograph of a page. It looks like text, but a screen reader finds no words, and nobody can search it, reflow it or change its colors. Optical character recognition (OCR) fixes part of that by adding a hidden text layer. It does not add structure such as headings, lists, tables and reading order, and its mistakes are read aloud exactly as written. For clerks and records staff, this covers a lot of the library: signed minutes, resolutions, ordinances, plats and the letters inside agenda packets. This guide shows how to spot a scan, what OCR does and doesn't do, the full repair path in Acrobat Pro, and when retyping, archiving or providing on request is the better choice.

How to tell if a PDF is scanned

Three quick tests settle it.

1. Select a sentence. If you can't highlight individual words, the page is an image.
2. Search for a word you can see. If Ctrl+F finds nothing, the page is an image.
3. Run Acrobat's Accessibility Checker. Its Image-only PDF rule reports content that isn't accessible text. In Adobe's words, "If the document appears to contain text, but doesn't contain fonts, it could be an image-only PDF file."

One trap: many copiers and scanners run OCR as they scan. Those files pass the search test, but their text layer has never been checked or tagged, so they still need the rest of this guide.

The W3C sums up what an untouched scan means for people: "Assistive technologies cannot read or extract the words; users cannot select, edit, resize, or reflow text nor can they change text and background colors." A screen reader user gets none of the content. A person with low vision can't reflow the page to read it without scrolling sideways, and a person who listens with text-to-speech hears nothing. DOJ has noticed. In its November 2023 letter of findings about Upton County, Texas, it listed PDFs on the county's election website that were "inaccessible scanned images" among the problems it found.

What OCR does and doesn't do

OCR reads the picture and writes the words it recognizes into an invisible layer behind the image. Adobe describes the result as "a searchable text layer." The page looks the same, and a screen reader can now reach the words.

That is real progress, and the W3C counts it: technique PDF7, "Performing OCR on a scanned PDF document to provide actual text," is a sufficient technique for WCAG 1.4.5 Images of Text. Even so, PDF7's own example doesn't stop at OCR. It goes on to correct the words OCR was unsure of, add tags and run an accessibility check.

Here is what OCR leaves for you.

• Structure (1.3.1). No headings, lists, paragraphs or table cells. A screen reader user can't jump between sections or hear a table's headers.
• Reading order (1.3.2). Two-column minutes, letterheads and stamped margins can come out in the wrong sequence.
• Correct words. Acrobat keeps showing the original image and places the recognized text behind it, so a sighted reviewer rarely sees the errors. A screen reader user hears every one of them: an "l" for a "1" in a resolution number, or a dropped decimal point in a dollar amount.
• Descriptions (1.1.1). Signatures, seals, stamps and photos have no text equivalent.
• Language (3.1.1). OCR needs to know the language to recognize it well, and the document must declare it so screen readers pronounce it correctly.

How to make a scanned PDF accessible, step by step

Menu names follow Adobe's Acrobat Pro help pages as of October 2026.

1. Scan well, or rescan. If you still have the paper, a better scan saves hours of correction. The South Dakota State Archives sets 300 pixels per inch for standard typed text "to support the use of optical character recognition (OCR)," and grayscale or color at 300 to 600 ppi for faint or handwritten pages.
2. Keep pages straight. Adobe warns that on a skewed scan "the characters can become unrecognizable." When you scan in Acrobat, its Deskew filter straightens tilted pages.
3. Choose the language first. Acrobat asks for it in the Recognize Text dialog, and Adobe lists incorrect language selection among the causes of poor results. Set the document's language too, on the Advanced tab of its Properties.
4. Run OCR. Choose All tools > Scan & OCR > In this file, select the pages and language, then select Recognize Text. In multiple files handles a folder of scans.
5. Correct the text. Choose Scan & OCR > Correct recognized text, turn on Review recognized text, and accept or fix each highlighted word. Adobe also advises you to "review the document to ensure the text is accurate and complete," so read names, dates, dollar amounts and ordinance numbers against the image.
6. Tag it. Choose All tools > Prepare for accessibility > Automatically tag PDF. Auto-tagging guesses headings from type size, and Illinois State University's guide (opens an external site) warns that larger, bolder text often becomes a heading "even if they are not supposed to be headings," so check every level.
7. Fix the reading order. Our guide to logical reading order shows how to check and repair it.
8. Rebuild tables. Tables on scanned pages often come out as loose lines of text and need real table tags with header cells.
9. Describe what OCR can't read. Section508.gov says a scanned signature "is simply an image. It should be marked as a 'figure' and provided with equivalent alternative text." Where a signature overlaps typed text, its advice is to "simply mark the entire signature block as a figure and add all of the text as alternative text." Alt text such as "Signed by the mayor; attested by the city clerk" or "Stamped: Received March 3, 2026" does the job. Marks that carry no meaning become artifacts that screen readers skip.
10. Decide about handwriting. Adobe notes that OCR "may struggle with" handwriting. Transcribe a short margin note into the alt text or a typed note. Pages of handwriting need a typed transcript, or one of the options in the next section.

Then run the Accessibility Checker and read the result with a screen reader.

Fix, retype, archive or provide on request: which fits each scan

Not every scan deserves the full repair. Sort by what OCR can recover and whether people still use the document.

The three alternatives each have a place.

• Retype or recreate. When OCR output needs more correction than retyping would take, type from the original. For recent records, look for the source file first. The adopted resolution usually started in a word processor.
• Archive. Title II's archived content exception can cover web content that "reproduces paper documents created before" your compliance date, if all four archive conditions are met. Old signed minutes and historical plats can fit. Our guide to archive, keep or fix walks through the conditions.
• Provide on request. Exempt scans may still need to reach someone who asks, in a format that works for them. Decide in advance who handles requests and how fast.

Scanned pages inside agenda packets

A 200-page agenda packet often mixes born-digital staff reports with scanned letters, signed contracts, petitions and plats. Current packets matter, because people use them to attend meetings and comment. One scanned insert leaves part of the packet unreadable, and Adobe warns that combining tagged and untagged files "results in a partially tagged PDF," adding that screen reader users may be "unaware of the pages that don't have tags."

• Ask applicants, consultants and vendors for the original digital file, not a printed-and-scanned copy.
• OCR, tag and check each scanned insert before it goes into the packet, or retag the combined packet after you assemble it.
• Give every signature page a short description instead of leaving it as an unexplained image.

How Ockem Proof handles scanned documents

Ockem Proof, our product, treats a scan as a page to rebuild, not just a text layer to add.

• Detected on arrival. Image-only documents are identified the moment they arrive, and OCR can run across a whole library as one program.
• Rebuilt as structured content. Each page becomes real headings, paragraphs, reading order and tab order, with the scan kept underneath as an artifact, a background layer screen readers skip. The page looks identical.
• Checked like any other document. The structured output feeds tagging and reading-order repair, and every rule re-verifies it on the same document.
• Estimates flagged. Heading levels recovered from a scan are estimates, so flagged pages deserve a person's review, and a person confirms reading order.
• Language followed. OCR follows the document's declared language across 37 language packs, and 31 are recognized automatically when nothing is declared. Right-to-left scripts still need a person's pass on reading order.
• Contrast measured from the image. On scanned pages, contrast is measured from the page image, and you are asked only where it falls short. A color change is never applied without a person.
• Honest dead ends. A document whose content can't be recovered gets a "needs re-authoring" advisory instead of endless retries, and "Have Ockem do this" asks our experts for a price in documents, which you accept or decline.

Pricing stays predictable. Proof checks every file for free first. A file that needs far more work than its page count suggests, which can include a scan, shows a fixed price in documents before anything is spent. Accepting is one click, and skipping is free.

A person still decides what handwriting says, what a signature block should announce, and whether a scan should be fixed, retyped or archived. See the scanned documents rule page.

Next steps

You can try one PDF (opens an external site) of up to 10 pages free, with no account. To find out how many PDFs your site holds, use the free PDF inventory tool. The scanned documents rule page shows how Proof handles OCR. For old scans kept for the record, read archive, keep or fix. For reading order after OCR, see logical reading order, and for maps and plats, see alt text for complex images.

Questions

This guide explains the rules in plain language. It isn’t legal advice. Check how they apply to your organization with counsel.