Yomitan only reads text that a browser page actually contains. It does nothing on a built-in PDF viewer, in Safari, in apps and games, or on pictures of text. The fix is either to get the text into a page it can read, or to read the screen instead.
Why Yomitan does nothing
Yomitan is a browser extension. Its GitHub page lists Google Chrome, Mozilla Firefox and Microsoft Edge as supported, and does not mention Safari. When you hover a word, it reads the characters in the page under your pointer. If there are no characters there, or the browser will not let an extension see them, nothing pops up.
That covers most "not working" reports. A PDF in the browser's own viewer is not a normal page. Google Docs does not give extensions ordinary page text to read, and a YouTube video shows subtitles that may be burned into the picture rather than text on the page. Before you reinstall anything, ask if the pointer is over real text on a supported browser. If you are, and the extension still does nothing, check that it is switched on for that site and that a dictionary is installed.
PDFs
A WaniKani forum thread on Yomichan for PDF (Yomichan is the older name of the same tool) has the poster saying it "does not work on pdf files for browser security related reasons". The workarounds in that thread are real ones. One reply says the easiest way is a clipboard watcher: copy the text out of the PDF and it appears on a web page, where the extension can read it. Another converted the pages to images and ran Mokuro.
A guide on Substack gives the third: open the file in Mozilla's pdf.js viewer at mozilla.github.io/pdf.js/web/viewer.html instead of the browser's own, then look up words as on any page. The author notes the PDF must be searchable, meaning its text can be selected. A scan is just pictures, so none of these three help there.
Safari, YouTube and Google Docs
Safari is the one with a clear answer. Yomitan's page does not list it, but a similar tool, 10ten Japanese Reader has a Safari version, and its App Store listing shows Mac support from macOS 11. It is a Japanese dictionary pop-up with pitch accent information, so it is a decent stand-in if you read in Safari. It is not Yomitan, and its Anki side is thinner, so check it does what you need.
For YouTube and Google Docs the workaround is the same as for PDFs: get real text in front of the extension. A caption track you can see as text, or a transcript, works. For a Google Doc, export it or paste the passage into a plain page. Not every combination has been tested, and these sites change.
Manga and other images
Text inside a picture is not text, however clear it looks. For manga the established fix is Mokuro. It runs OCR on each page offline, using manga-ocr, and builds a reader in the browser where the text is selectable, so Yomitan can look words up again. You install it with pip, which needs Python 3.10 or newer, and run it from the terminal. It is free and it works well, with the one cost being that setup is terminal work and you read inside that reader.
When there is no text to hook
Some things will never become page text: a scanned book, a game, a desktop app, a video with burned-in subtitles, a screenshot a friend sent. This is where LeafyApp helps. Press ⌥A, drag a box over the line, and macOS text recognition reads that patch of screen. An AI then explains the word for the sense that sentence uses, and the word is saved with its sentence, source and date.
It works in any app because it never asks the app for its text. That is the whole trick, and also the limit: it reads pixels, so a stylised font can fail, and it will not do the dictionary-grade lookups Yomitan does. For games specifically see Textractor for Mac, and for manga, PDFs and screenshots see Japanese OCR on a Mac.
What Yomitan still does better
Quite a lot. You install dictionaries yourself, including EPWING ones, so you can have several definitions, and pitch accent data side by side. Its Anki integration, through AnkiConnect, lets you map each part of an entry to a field on your note type. If you read Japanese on the web every day, use it and keep using it.
LeafyApp does not try to match that. It gives you one explanation of the word as your sentence used it, saves the sentence, and exports to Anki with a card preview, or to CSV, Excel, JSON or PDF. The picture is read on your Mac and thrown away, and only the word and its sentence go out, as the privacy page explains. For the bigger picture of mining from everywhere, see sentence mining on a Mac.
Questions people ask
Why is Yomitan not working on PDFs? The browser's built-in PDF viewer does not expose the page text to extensions. Open the PDF in the pdf.js viewer, copy the text onto a normal page, or convert the pages with Mokuro. A scanned PDF has no text at all.
Why is Yomitan not working on YouTube? Subtitles that are part of the video picture are not text on the page, so there is nothing to hover. Use a caption track the page can display as text, or read the screen instead with a tool like LeafyApp (protected streaming video often captures as a black frame on macOS, so test it first).
Why is Yomitan not working on Google Docs? Google Docs does not give extensions ordinary page text to read, and no official fix turns up. Paste the passage into a plain page or an exported copy and look it up there.
Is there a Yomitan for Safari or Mac? Yomitan's page lists Chrome, Firefox and Edge, which all run on a Mac. For Safari, 10ten Japanese Reader has an App Store version, and for non-web content on a Mac, see sentence mining on a Mac.
What do I use when Yomitan cannot read the text? Anything that is a picture needs OCR first. Mokuro handles manga in a browser, and Japanese OCR on a Mac covers screenshots, PDFs and apps.
When it is not text, read the screen.
Free while LeafyApp is in beta. One download, macOS 14 and up.
Yomitan, Yomichan, 10ten Japanese Reader, Mokuro, Safari, YouTube and Google Docs are trademarks or names of their respective owners. LeafyApp is an independent app, not affiliated with or endorsed by any of them, and the names are used here only to describe compatibility and for comparison.