Why your highlights don't move from one reading app to another
A highlight is stored as an address in one particular file, written in the notation of the app that made it. Another app has a different copy of the book and reads a different notation, so it can't follow the address. What you can carry over is the highlighted text, and the second app then has to find that passage again in its own copy.
Each app writes the address its own way
Amazon has used its own numbering from the start. The user guide for the first Kindle, from 2007, calls locations the digital answer to page numbers: pages change when you change the text size, but a location stays put. A bookmark there belongs to the first location on the page, and the list of notes and marks shows each one under its location number.
A Kobo keeps highlights as rows of the Bookmark table in its database, and the form of a row depends on the file. A MobileRead forum post from someone writing an importer shows both forms. For a plain EPUB, the start of the highlight is a file path inside the book followed by a step path such as point(/1/4/2/4/38/1:6). For a KEPUB it's span#kobo\.153\.1, the ID of a sentence span that exists only in the converted file. One highlight in one book would therefore be written two different ways, depending on whether the Kobo got the book as EPUB or as KEPUB (the difference between the two is covered in KEPUB or EPUB for a Kobo).
KOReader ties highlights to where the file lives. Its user guide says it keeps progress, highlights, and notes in a folder named after the book with .sdr on the end, next to the book by default. Move or rename the book with any other file manager and, by the guide's own account, those notes are gone without a warning. A third option names the folder by a partial MD5 hash of the book's contents instead, which survives a move but not a change to the file itself.
Why a position breaks
The EPUB standard has its own address for a point in a book. EPUB Canonical Fragment Identifiers 1.1, which the IDPF published on 5 January 2017, walks the document tree in numbered steps: an even number for each child element, an odd number for each run of text between them, and a colon with a character offset at the end. A CFI is precise, but only for that exact file. If a paragraph is added before the highlight or the book is converted again, the step numbers point somewhere else.
The specification allows for this: a CFI can carry an element's ID and a few characters of the text around the point in square brackets, and an app should use them to correct an address that no longer fits. If it can't derive a correction, the CFI counts as an invalid reference. The W3C's model for annotations is blunter about plain character offsets, describing them as "very brittle with regards to changes to the resource".
The same goes for two editions of a book, even in one app. A second edition is a different file, and so is a publisher's corrected copy, or the store's copy next to one you added yourself. Every stored position is relative to the file it was made in.
What the exports give you
An export is the usual way out, and most of the formats on offer are made for a person to read. Kobo's help page describes Export Annotations on Kobo.com, for books bought from Kobo or read through Kobo Plus, as a PDF, HTML, plain text, or Markdown file. It says the export may not work for every book because it depends on the author's approval, that markups can't be exported, and that a generated file is deleted after 30 days.
Google Play Books writes each book's notes to its own document in Google Drive, in a folder called "Play Books Notes" unless you rename it, once you turn on "Save notes, highlights, and bookmarks" in the app's settings. Google's help page warns that highlighted text may be missing from the document when the publisher has turned off copying or when you pass its copy limit; the page number in the document then takes you back to the passage. It also says your own edits to that document are wiped the next time you add a note to the book.
KOReader's export, under Export Highlights in the top menu, writes text, HTML, Markdown, JSON, or a file in the format of Kindle's My Clippings.txt, and can send the highlights to services such as Joplin and Readwise (user guide). On a Kindle, the clippings file and Amazon's online Notebook are the two ways out, and where your Kindle highlights are covers both.
None of these files is meant to be read back in by another app. The IDPF drafted a format for that in 2015, Open Annotation in EPUB, which asked reading systems to import a conformant annotations.json, and it stayed a draft. The W3C has taken the problem up again: EPUB Annotations 1.0 was a Working Draft as of 15 October 2026. It asks an app to give each note more than one selector, to read them from the most precise to the broadest, and to fall back on the others when the most precise no longer fits the book's text. Where that file lives, and what a note in it may contain, is in the post on marginalia.
Finding the passage by its words
An anchor that survives a change of file stores the words themselves. The W3C Web Annotation Data Model, a Recommendation since 23 February 2017, calls this a TextQuoteSelector: the exact highlighted text, plus a prefix and a suffix taken from just before and after it, so that a phrase which appears twice in a chapter can still be told apart.
Hypothesis, the web annotation tool, set out how it uses this in a 2013 post on fuzzy anchoring. Each annotation keeps three selectors, an element path, character offsets, and the quote with 32 characters of context on either side, and they're tried in that order. If the page has changed, it searches for the prefix and suffix near where they used to be and checks the text between them against the quote within a set tolerance. Only after that does it search for the quote alone. The matching uses a modified version of Google's diff-match-patch library.
Readwyn anchors its highlights on the same idea. Each one records the chapter file, the offsets, and 32 characters of text before and after the highlight. If the EPUB has changed, it searches the chapter for the three together and takes the match nearest the old offset. A highlight it can't find stays in the Notebook's list but doesn't show in the book.