How to Make PDF Readable to Search, Software and People

To make a PDF readable, first settle who or what has to read it. A person needs clear pages. Search and other software need real characters stored in the file, which a scan gets from OCR. A screen reader needs tags, a reading order and descriptions of images. Each kind of readable has its own fix and its own test.

The word causes trouble because the person asking rarely says which kind they mean. A clerk's office, a client's document system and a blind colleague can all send back the same complaint, "this PDF is not readable", about three different defects. The sections below separate them so you repair the right one.

What does it mean for a PDF to be readable?

A PDF is readable in three separate senses. A person can make out the page. A program can find and copy the words because the file stores them as characters. Assistive technology can announce the content in a sensible order because the file is tagged. A file can pass one of these and fail the other two.

Take a signed lease between two fictional parties, Corbin Mills LLC and Tessler Outfitters, Inc., scanned on the office copier and emailed to you. Open it and you can read every clause, so in the everyday sense it is readable. Press Ctrl+F and type "Tessler" and the reader reports no matches, because each page is a photograph and holds no letters. The post on what an image-only PDF is explains that kind of file. Hand the same lease to someone who uses a screen reader and the software announces an empty document.

Two points follow from the table. The senses stack: every row below the first needs text, so a scan must go through OCR before anything else can happen. And legibility sits apart from the rest. OCR does not sharpen a faint fax for the human eye, and a beautifully legible scan is still silent to software.

When someone asks you to convert a PDF to a readable PDF, the second row is usually what they mean. Ask anyway. The answer decides whether you spend two minutes or two hours.

Ask what the reader will look up. A clerk often wants a pleading searched for a citation and a statute, or an exhibit searched for a party name. Characters in the file make that search possible. Tags and reading order serve a person using a screen reader. Match the work to the request before you spend the afternoon.

Readable by search: giving a PDF text a computer can find

A PDF is readable by search when the file stores the words as characters and not only as a picture of them. PDFs saved from a word processor already do. A scan gets its characters from OCR, which reads each page image and writes what it found into the file as a hidden text layer.

Before you do anything, find out which kind of file you hold. Drag the cursor across one line. If single words highlight, text is present and the file may need nothing. If the whole page selects as one block, the page is a picture and needs OCR.

Three situations cover almost every file:

• The PDF came from Word or a case system. It has real text already. Do not run OCR on it. Recognition would add a second, less exact copy of text the file already holds perfectly.
• The PDF is a scan with no text. Run OCR and save a searchable copy. The guide to adding a text layer to a scan step by step walks through it.
• The PDF has text, but the text is wrong. A copier added a weak layer, or the fonts do not map to ordinary characters, so a paste gives you "Corbm Mi11s". The file looks searchable and behaves badly. It needs its text replaced, and the post on converting a PDF to readable text covers replacing a bad text layer by running OCR again.

To create a readable PDF from a scan in Schema Law, sign in, upload the PDF, PNG or JPEG (up to 20 MB), check the parsed pages against the original on screen, and export Searchable PDF. That export is your original file with a hidden text layer, so search finds the page. Every plan exports it, including the free plan, which takes one document a day of up to three pages.

Text in the file is necessary for search and does not guarantee a hit. Search compares characters, so a name the OCR misread stays hidden. The post on how a PDF reader scans a file for text covers how matching works and why a word you can see is sometimes not found.

What OCR will not do for a reader

A text layer changes what software can do with the page. It leaves the page picture as it was. If the complaint is that a person cannot make out the words, because the copy is faint, cropped or photographed at an angle, the remedy is a better image: rescan the paper, or ask the sender for a cleaner copy or the original electronic file.

Is a machine-readable PDF the same as a searchable one?

Not always. A searchable PDF gives a program the words. "Machine-readable" usually asks for more: content a program can process without a person interpreting it. Federal law defines the term for government data, and the definition turns on meaning as well as text.

The definition sits in the federal statute on information policy. Under 44 U.S.C. § 3502, the definitions section for federal information policy, machine-readable, when used with respect to data, means data "in a format that can be easily processed by a computer without human intervention while ensuring no semantic meaning is lost." That statute governs federal agencies' data. It does not regulate your exhibits. It is useful because it is a definition written into law, and it shows what the phrase is reaching for.

Hold a searchable scan up against that wording. A program can pull the words out easily. Whether "no semantic meaning is lost" depends on the document.

• A letter or a brief. The meaning is in the sentences. Text in reading order carries nearly all of it. A searchable PDF is, for practical purposes, machine-readable.
• A rent roll or a ledger. The meaning is in which number sits in which column. Flat hidden text gives a program a stream of figures with nothing to say which is the unit number and which is the arrears. A person can see the columns, and the program cannot.
• A completed form. The meaning is in which answer belongs to which question. A text layer rarely records that pairing.

So the honest answer to "convert this PDF to a machine-readable PDF" is a question back: what will the machine do with it?

Schema Law exports the same uploaded document in more than one format, so you can keep a searchable PDF for the file and send the tables as a spreadsheet. Every plan includes Excel, CSV and JSON. The list of export formats on the homepage says what each one holds.

If a native file exists, it beats all of this. A spreadsheet is more machine-readable than any PDF made from it, and asking for it costs one email.

Readable by a screen reader: tags, reading order and alternate text

A PDF is readable by a screen reader when it has text and a tag structure that tells the software what each piece of content is and in what order to announce it. OCR supplies the text. It does not, by itself, supply reliable tags, a checked reading order or descriptions of images.

The World Wide Web Consortium's accessibility guidance is blunt about scans. Its technique on using OCR to provide actual text in a scanned PDF says a document made of scanned images of text "is inherently inaccessible because the content of the document is images, not searchable text," and that assistive technologies "cannot read or extract the words." Recognition is the entry ticket.

Three more things sit on top of the text.

Tags. A tagged PDF carries a hidden outline that labels each element: this is a heading, this is a paragraph, this is a list, this is a table with these header cells. A screen reader user moves through a long document by jumping between headings, exactly as you skim one by eye. Without tags there is nothing to jump between.

Reading order. The same W3C guidance, in its technique on correct tab and reading order in PDF documents, states that "the reading order of a PDF document is determined primarily by the tag order of document elements." If the tags run in the wrong sequence, a two-column page can be announced straight across both columns, line by line, and a footer can interrupt a sentence.

Alternate text. A signature block, a site plan, a photograph of damage and a company logo are pictures inside the page. The W3C technique on text alternatives for images in PDF documents describes alternate descriptions as "human-readable text that can be vocalized by text-to-speech technology." Someone has to write them. Software can guess that a region is a picture, and it takes a person to say the picture is "Exhibit C, survey of Lot 14".

The practical route to an accessible PDF

How much work this is depends on where the document starts.

1. You have the source file. Fix it there. Use real heading styles, real lists and real tables in the word processor, add descriptions to images, then export to PDF with the option that keeps tags. This is far less work than repairing a PDF.
2. You have only a scan. Run OCR, then decide. For a long document, converting to Word, correcting the structure and exporting a tagged PDF is often quicker than tagging the scan page by page.
3. You must keep the scan as it looks. Run OCR, then add and correct tags in a PDF editor that has tagging tools. Check every page's order, mark page numbers and running headers as background so they are skipped, and describe each meaningful image.

Whichever route you take, test the result with the methods in the next section. Whether a document must be accessible at all, and to what standard, depends on who is publishing it and where. That is a legal question for the organization, and the rules differ between public bodies, courts and private firms.

How do you test whether a PDF is readable by software?

Test each sense separately, with the tool that will do the reading. Select and search for text. Paste a page into a plain editor to see what a program receives. Check for tags and listen to a page read aloud for accessibility. A file that passes one test tells you nothing about the others.

Work through these in order and stop at the level you need.

A few cautions make these tests worth the minutes they take.

• Test late pages. Bundles are assembled from several sources. Page 1 is often a typed cover sheet and page 40 a photograph of a receipt.
• Do not stop at "select". Selection proves characters exist. Only the paste test shows whether they are the right characters in a usable order.
• A "Tagged PDF: Yes" entry is a start. Automatic tagging can label everything as a plain paragraph. Listening to a page is the test that counts.
• The destination test settles arguments. If the e-filing portal, the client's document system or the review platform takes the file and finds your word, the file is readable in the sense that matters today.

Write down what you tested, for example "searchable copy, pp. 2, 19 and 44 checked by search and paste; not tagged". The next person then knows which claims the file can support.

Who asks for a readable PDF and what they usually mean

Courts usually mean text-searchable. Government agencies may mean accessible, or data they can load. Clients and opposing counsel mean they want to search and copy. AI tools need text they can take in. The same word carries four requirements, so read the rule or ask the person before you convert anything.

Courts. Where a court asks for text-searchable PDFs, the details differ from court to court. The post on what a court means by text-searchable covers the requirement and how to read your own court's version of it.

Agencies. A request for an accessible document is a request for tags, order and image descriptions, and OCR alone will not satisfy it. A request for machine-readable data usually means a spreadsheet or a data file, and a PDF of any kind may be the wrong answer.

Clients and counsel. Send a searchable copy and keep the scan as received. If they intend to edit the document, a Word file serves them better than any PDF.

AI tools. A chat assistant or summarizer can only work with text it can extract. Give it an image-only scan and it may report an empty file, read part of it, or run recognition of its own that you never get to check. Convert first, verify the text, then decide whether the tool should see the document at all. For client files, read the provider's terms before you upload, and compare them with the security commitments Schema Law publishes for its own service.

When the request is vague, one reply clears it up: "Do you need to search it, edit it, load the figures into something, or have it read aloud?" Each answer points to one row of the first table in this post.