When a PDF looks correct but copies as gibberish, diagnose the text representation before adding tags or running OCR over the whole document. The visible shapes and the characters exposed to software are not necessarily equivalent. A useful repair restores the intended words and symbols while preserving their structure, then verifies both the content and the reading experience.
Reproduce the failure with a small text sample
Choose a short passage whose intended wording is known. In an illustrative report, that might include 'office', a negative temperature, an accented name, and a section symbol. Copy each sample into a plain-text editor, search for it in the PDF, and read it with the intended assistive setup. Record the exact mismatch without including confidential text in a public issue report.
Distinguish missing text from wrong text. A scanned page may have no text representation; an OCR layer may contain recognition errors; a born-digital PDF may have character mapping problems; and a viewer may interpret a file differently from another viewer. These observations narrow the investigation but do not establish a cause by themselves. Compare more than one representative passage before deciding on a repair.
Separate appearance from character meaning
Adobe's font-handling documentation describes embedding font data to preserve the intended appearance. An embedded font is therefore useful evidence about rendering, but it is not a complete test of the extracted text. A screenshot and a font-properties panel cannot show whether copying a word produces its intended characters.
The PDF Association's tagged-PDF syntax guide discusses Unicode mappings and ActualText for replacement text. A ToUnicode mapping helps associate encoded glyphs with Unicode characters. ActualText can supply a replacement sequence where needed, including cases such as a ligature representing multiple characters. The guide recommends limiting its use on structure elements to Span, or using Span marked-content sequences.
| Observation | What to investigate next |
|---|---|
| Text looks right but copies as unrelated letters | Character mapping, replacement text, and extraction behavior |
| Text cannot be selected on a scanned page | Whether an accurate OCR text layer exists |
| Only a few mathematical symbols fail | The source symbols, font/export handling, and semantic alternatives |
| One viewer differs from another | Viewer versions and interpretation of the same exact file |
Choose the repair closest to the source
When the original document is available, correct the source or export route and generate a new PDF first. Keep a small reproduction so a different font or exporter setting can be evaluated without rebuilding the entire publication repeatedly. Compare the corrected sample with the approved wording, then check other occurrences of the same character pattern across the document.
If a specialist applies replacement text directly, keep it precise. Replacing a whole table or paragraph with a broad plain-text string may change how its structure is exposed. Alternative text that summarizes a figure and replacement text that stands in for actual content serve different purposes; choose the property that matches the intended semantics. Validate the edited file against its chosen PDF profile.
The PDF Association's accessibility-technique overview treats correct textual representation and meaningful structure as related requirements. Treat the practical implication carefully: repairing characters does not also repair reading order, headings, or tables. Keep those checks in the review record rather than calling a successful copy-and-paste test an accessibility pass.
Avoid introducing a second, conflicting text layer
Do not assume that OCR is the best response to every extraction problem. For a born-digital source, a new recognition layer can introduce different spelling or symbol errors and may leave duplicated content. If OCR is necessary, inspect its result against the page, resolve mismatches, and check whether pre-existing text needs controlled treatment by the chosen tool. Preserve an untouched original for comparison.
Use the OCR text review guide for image-based material. For mathematics, consult the formula guide, because a readable linear string may still fail to communicate mathematical relationships. Select a repair based on the content type instead of applying one global conversion to every problem.
Verify content and user behavior separately
- Compare repaired samples with the approved source, including punctuation, signs, accents, and spacing.
- Repeat copying and searching in the same environment that revealed the problem.
- Test reading with the intended viewer and assistive technology, recording their versions.
- Inspect related content elsewhere, including repeated symbols and the same font in other sections.
- Recheck structure and validation findings after the repair, and keep unresolved differences visible.
A useful result states which text was corrected and which environments were tested. Avoid claiming universal compatibility from one successful viewer. Save the final PDF together with the reproduction notes so a future export can be checked for the same regression.
Frequently asked questions
Does embedding fonts fix garbled copied text?
Font embedding supports consistent appearance, but text extraction also depends on the PDF's character representation and the reading software. Check the actual copied characters and source wording rather than treating an embedded-font label as proof of correct text.
Is ActualText the same as image alternative text?
ActualText supplies replacement text for content, while alternative text can describe content such as a figure. Use the property that matches the semantic purpose and verify the resulting structure and reading behavior in the target environment.
Should I OCR a PDF that already contains selectable text?
Diagnose the existing representation first. A born-digital export problem may be better repaired in its source, while indiscriminate OCR can introduce recognition mistakes or duplicate text. If OCR is needed, compare the result with the approved content and inspect the resulting layers.
Sources
The statements above rest on the primary sources below. Where a source is a law or standard, the version and date named in the text are the ones checked on the review date. This guide is general information, not legal advice.
- PDF Association: Tagged PDF Best Practice Guide, Syntax . Unicode mapping and the use of ActualText.
- Adobe: Font embedding and substitution
- PDF Association: Understanding accessibility techniques
Keep reading
How to make a PDF accessible
A repeatable process: tag structure, add alt text, fix reading order, export as PDF/UA and HTML.
Read the guide →Accessible math in PDF: Formula tags and MathML
Plan accessible equations with the right PDF profile, preserve mathematical relationships, and test Formula and MathML output with real reading tasks.
Read the guide →How to test a PDF with a screen reader
A repeatable screen reader test for PDFs with NVDA, JAWS, or VoiceOver: which viewer to open the file in, the keystrokes to use, what to listen for, and independent data on which viewers expose tags.
Read the guide →