PDF annotations are a rich and formally specified layer of interactive content that sits on top of the static page content. The PDF specification defines over 25 distinct annotation types, each with its own set of properties and behaviors. Understanding how annotations are structured internally explains what information can be extracted from them and why some annotations behave differently across PDF readers.

Annotations are stored as annotation dictionaries in the page's Annots array — a list of references to annotation objects associated with that page. Each annotation dictionary contains a Type entry (always /Annot), a Subtype entry identifying the specific annotation type, a Rect entry defining the annotation's bounding rectangle on the page, and various type-specific entries for content, appearance, and behavior.

The most common annotation types fall into several categories. Markup annotations include Text (sticky note icons that expand to show comments), Highlight, Underline, StrikeOut, and Squiggly (wavy underline). These are used during document review and typically carry a Contents string with the reviewer's comment, a T (title) entry with the author's name, and an M (modification date) entry. They may also have popup annotations associated with them — separate annotation objects that define the popup window displaying the comment text. The InReplyTo entry allows annotations to form threaded conversations, where replies reference their parent annotation.

Widget annotations are used for interactive form fields — text inputs, checkboxes, radio buttons, dropdown menus, and signature fields. Unlike markup annotations, widgets are closely tied to the document's AcroForm structure, with each widget linked to a field dictionary that defines the field's name, type, default value, and validation rules. Extracting widget information requires traversing both the annotation layer and the form field hierarchy.

Appearance streams are a critical concept for annotation rendering. Each annotation can have an AP (appearance) dictionary containing one or more appearance streams — these are miniature content streams (like tiny PDF pages) that define exactly how the annotation looks when rendered. The normal appearance (N) is displayed under standard viewing conditions, the rollover appearance (R) when the mouse hovers over the annotation, and the down appearance (D) when the annotation is clicked. PDF readers use these appearance streams to render annotations consistently, and annotations without appearance streams may look different across different readers as each applies its own default rendering.

FDF (Forms Data Format) is a companion format to PDF designed specifically for exporting and importing form data and annotations. An FDF file contains annotation data in a compact format that can be applied to a PDF document to add or update annotations. This mechanism enables workflows where reviewers export their annotations as FDF files, which are then merged into the master document. Some PDF review systems use XFDF (an XML-based variant of FDF) for easier programmatic processing of annotation data.

Annotation extraction involves parsing each annotation dictionary, interpreting its type-specific fields, resolving references to popup annotations and reply chains, and presenting the information in a structured format. The Contents field provides the text of comments and notes, but it may be plain text or rich text (formatted using a subset of XHTML). Highlighted text must be reconstructed by mapping the annotation's QuadPoints (coordinates defining the highlighted region) back to the underlying text content, which requires the same kind of text extraction and positioning analysis used by text extraction tools.