# What a screen reader actually does with a scanned PDF Canonical: https://paperlyft.com/articles/scanned-pdf-accessibility By Shubham Tiwari Published: 2026-10-14T18:37:11.861Z Updated: 2026-10-14T18:37:11.861Z A scanned PDF looks like a document, but to a screen reader it’s just an image. Here’s why OCR alone doesn’t fix it, and what a scanned PDF needs to be truly accessible. A scanned PDF looks like a normal document. You can see the heading, the table, the signature line. Open it with a screen reader and you often get nothing. Or you hear “image” and then silence. We first heard about this from county election officials at the ASU AI + Elections Hackathon. Many of their documents were scanned or handwritten, and they had no good way to make them accessible. It’s one of the most common accessibility problems in public records today. ## Key takeaways - A scanned PDF is an image. A screen reader has nothing to read. - OCR adds text, but not the headings, reading order and table structure a screen reader needs. - An accessible PDF needs accurate text, real structure, alt text and validation against WCAG 2.1 AA and PDF/UA. **Why a scanned PDF is just a picture** When you scan a paper document, the scanner takes a photo of each page and saves it inside a PDF. The file looks like a document, but there is no text in it. There are only pixels shaped like text. A screen reader like JAWS or NVDA doesn’t look at pixels. It reads the text and structure stored inside the file. If there is no text, there is nothing to read. So when a resident who is blind opens a scanned agenda or public notice, the screen reader has nothing to say. **OCR helps, but it isn’t the fix** Many tools run OCR (optical character recognition) on scanned files. OCR finds the letters on the page and adds a hidden text layer. Now the screen reader has words. But words alone don’t make an accessible document. A screen reader also needs to know: - What is a heading and what is body text - What order to read things in - Which column header goes with which table cell - Which label goes with which form field - What an image or stamp means, and which lines are just decoration Without that, the screen reader reads text in whatever order it landed. Two columns get read straight across. A table becomes a list of disconnected numbers. A form becomes a stream of labels and values that don’t line up. The text is there. The document still doesn’t make sense. **What an accessible PDF needs** An accessible PDF has a tag tree. Think of it as a script for a narrator. It tells assistive technology what each piece of content is and in what order to read it. **A simple test:** if someone read the document to you in that order with your eyes closed, would it make complete sense? For a scanned page, getting there means: - Recognizing the text accurately, including handwriting. - Identifying structure: headings, paragraphs, lists, tables and form fields. - Setting a logical reading order, which isn’t always the visual order. - Adding alt text for images, logos and stamps that carry meaning, and marking decorative elements so they are skipped. - Setting the document title and language. - Validating the result against WCAG 2.1 AA and PDF/UA. Each step depends on the one before it. If the text is recognized wrong, everything built on top of it is wrong too. **Why this matters now** Under the DOJ’s ADA Title II rule, state and local governments need their web content, including PDFs, to meet WCAG 2.1 AA. The deadlines are April 2027 for governments serving 50,000 people or more and April 2028 for smaller ones. People can also ask for an accessible version of a document at any time. **Where Paperlyft fits** We built Paperlyft for exactly these documents. It takes scanned and handwritten pages, recognizes the content, rebuilds the structure and produces an [accessible PDF]() that keeps the original layout. Before anything goes out, you review every page side by side with the original and edit anything that needs it. Have a scanned document? Send us one and we’ll show you what it looks like after. [Request access]()