Hey all,

I'm happy to hit send on this public beta newsletter #9 describing the latest round of Readwise Reader updates 🙂

I write this newsletter every one to two months covering features we've just shipped, bugs we've recently fixed, and what we intend to work on next. I also share tips and tricks to help you get the most out of Reader. If you prefer to read these in-app, you can subscribe to the private RSS feed linked here.

I don't have any housekeeping matters to share this newsletter, so let's get straight into the product updates:

• 👻 Ghostreader v2​​​​​​ – You can now customize the default Ghostreader ​prompts as well as craft your own ​creating the potential for all kinds of clever AI use cases while you read. As part of this, Ghostreader is now much easier to access on mobile and we've refreshed the default prompts.
• 📂 Mobile Folders – You can now browse your RSS feed folders and individual feeds on mobile using a left sidebar in the Feed section. As part of this, the Library section also received a left sidebar making it more intuitive to navigate your document tags and types of content.
• 🟡 Highlighting Upgrades – You should now notice several subtle upgrades to the highlighting experience ​including better grabbing of leading and trailing punctuation, smoother cross-page highlighting in Paged Scroll mode on mobile, and, most importantly, the ability to resize existing highlights!
• 📖 Ebook Upgrades​​​​​ – You'll now find the ebook reading experience to be more pleasant as a result of preserving the original EPUB styles, a full-width justification option, and faster load times featuring the book cover.

👻 Ghostreader v2

As a reminder, Ghostreader is Reader's AI-powered assistant ​that helps you get the most out of whatever you're reading without context shifting and breaking flow.

For most users, the headline of this version 2.0 is that Ghostreader is now much easier to access in the mobile app and the de​fault "prompts" have been reimagined from scratch ​based on everything we've learned after 18 months.

For example, you can now ask a question after highlighting anywhere in the document and receive much higher quality answers than before.

Or you can x-ray any character, place, or term to get a description of how that entity is used in the particular document.

Or you can "pick up where you left off" to get a quick recap of your last reading session (think: "previously on" when you start the latest episode of House of the Dragon or The Bear).

Or you can have Ghostreader run through your notes and highlights after you're finished reading to extract personalized key takeaways and to dos from that document.

And several more.

For power users, the headline of this announcement is that you can now create your own prompts using our powerful templating system enabling all kinds of creative and highly personalized use cases. In addition, you can opt-in to use the GPT-4 models provided you've inputted your OpenAI API key.

Here's a short Loom where I demonstrate how to make an etymology prompt that gives me the origins of words and phrases.

To begin creating your own, head to the Ghostreader prompts section of the web app Preferences. If you want to get your hands dirty, you'll probably want to read up on some the documentation in our new but still under construction Documentation Center (props to Cayla).

Now for a little product discussion (you can skip ahead to Mobile Folders if not interested in some of the thought process behind this feature).

There's a whole lot of handwringing in the product and design communities about the right and wrong ways to develop AI features. I hope to someday write a proper essay sharing our findings on this matter, particularly as it relates to our domain of reading software, but there are three core beliefs, loosely held, which I can briefly summarize here.

1. We're still in the command line era of AI. Many commentators like to point out that "prompting" is not a great user interface, implicitly suggesting that anything short of a novel interaction isn't worth shipping. We have no doubt that some innovative UI/UX patterns for large language models will be discovered, but at the same time, we don't want to sit on the sidelines waiting for that to happen. Instead, we think the best way to invent the future is to tinker in the present. This means embracing the metaphorical command line, as high bar as it is, until opportunities to innovate purpose-built AI features reveal themselves. As an example of this tinkering in practice, we've found that the "pick up where I left off" prompt is such a delightful use case of LLMs while reading that we'll surely elevate this into a first class feature with its own user interface. We know this is worth doing, and how to do it, thanks to getting our hands dirty with the messy command line.
2. Chatting while reading is too distracting. There might be some ways to complement the reading experience before or after with an AI chatbot, but as far as we're concerned, chatting while reading is a clash. To use a crude analogy, it's like having a conversation while watching a movie or sitting in a lecture. You'd be better off just paying attention in the moment and saving your conversation for after. What's with ​all these "chat with PDF" apps attracting millions of users then? As far as we can gather, none of our users regularly use these products. If there is a use case that retains, it's probably people using those apps as a substitute for reading rather than a reading companion. For example, a kid trying to quickly answer a homework question without reading the textbook or an office worker trying to avoid reading a boring 200-page research report. Those are valid use cases, but not what we're building software for. Ultimately, we want to maximize quality time spent reading long-form documents, not substitute for it.
3. The killer feature of LLMs are their flexibility. Every time I talk to a user who's excited about AI and reading, they open with something like, "Well, here's what I personally do with ChatGPT, but it's really weird and no one else would ever want this so I doubt you'll care or build it..." On the contrary! This boundless flexibility is, in our opinion, the killer feature of large language models. The fact that you can get so creative or so personalized is what makes ChatGPT such a hit. This means that to build an AI feature that prematurely abstracts away this flexibility in favor of "user friendliness" would actually be denying the user the best part of AI. Like serving french onion soup without the gooey melted cheese on top. (No artificial intelligence was consulted in crafting that A+ metaphor.)

Hopefully you can see all these beliefs reflected in this current iteration of Ghostreader. If you have a use case in mind and need help crafting a prompt, feel free to reach out and Cayla and/or I will try to help.

Mad props to Hannes on building everything Ghostreader related. To ship a feature like custom prompts, which is basically a low code interface, requires all kind of infrastructure and miscellaneous fixes.

📂 Mobile Folders

Last update we shared a new left sidebar on web and desktop that made the app easier to navigate and, more specifically, enabled you to create folders of RSS feeds for a more traditional feed reading experience. We've now adapted these same concepts to the mobile app.

You'll notice that you now navigate to your specific types of content (eg Articles, Books, PDFs, etc) and to document tags from the Library section; you browse your RSS feeds and folders from the Feed section; and the Views section has been removed to make way for a first class Account button on the bottom tray.

Props to Mati for seeing this delicate navigation project through to the finish line across both web and mobile.

🟡 Upgraded Highlighting

If there's one feature Readwise users care about, it's highlighting. Highlighting in Reader received several subtle upgrades over the past two months which I'll describe here.

First, there's been a chronic paper cut, particularly on mobile, where it'd be hard to grab leading or trailing punctuation when that punctuation was close to either edge of the screen. This is now fixed across iOS, Android, and web.

Second, highlighting across pages when reading using pages instead of continuous scroll is now smoother.

Last but certainly not least, you can now resize highlights on web and mobile!

While highlight resizing has been thoroughly battle tested on web, we only shipped the the feature to mobile this week, so we still consider resizable highlights to be in beta, particularly on Android. We are also still putting the finishing touches on making it work on e-ink devices like Boox and Daylight. If you run into any issues, let us know. We'll be continuously improving this feature over time.

Props to Adam for figuring out how to resize highlights – it's surprisingly hard from a technical perspective. Fun fact: Adam first demonstrated resizable highlights at our last company hackathon in Norway. Props to Artem on the other highlighting upgrades.

📖 Ebook Upgrades

Now that Reader has pagination and is increasingly being installed on all these great e-ink devices such as Boox and Daylight, the ebook use case has become more prominent in Reader. This has motivated us to start taking the ebook reading experience to the next level.

Most notably, ebooks will now default to the original styles contained within the EPUB file rather than Reader's parsed typography. In other words, most well-crafted ebook files contain instructions on how to nicely format their text, which Reader is now using. In addition, there's now an option to fully justify text in the Appearance panel.

Forgive my hideous photography, but I wanted to demonstrate on e-ink.

These changes might appear subtle in the example above, but they're surprisingly important in evoking that elusive feeling of "bookishness". Relatedly, whenever you open an ebook, the cover image will animate in as part of the loading screen. There's still much to do on the ebook front, particularly around performance, but the reading experience is maturing nicely.

Props to Artem for eating his own dogfood.

🔜 Coming Up

• EPUB Performance Refactor – In order to develop the ebook reading experience even further, we first need to improve the performance of long documents, particularly on slow devices. This will unlock refinements such as displaying the chapter title and chapter progress, better navigation, and better device handoffs.
• Send to Kindle – We've long resisted adding a Send to Kindle feature because highlights you make on sideloaded Kindle documents don't automatically sync to Readwise. (You need to manually import them.) That said, user demand has ultimately won us over.
• YouTube and PDF Repair – We're still experimenting with various techniques here, but it appears promising that we'll be able to clean up the YouTube transcripts we get from Google to use proper capitalization and punctuation without breaking the rest of the feature. In addition, we'll be able to apply this technique to the parsed text layer of PDFs to further improve that experience.
• Paywall Update – Last update I shared that we were working on the so-called "nuclear approach" to make saving your paywalled subscriptions more convenient. I regret to share that we've decided to shelf this for now due to extreme technical difficulties. Suffice it to say that publications like The New York Times and Wall Street Journal have entire teams dedicated to ensuring you read their content in their apps. To get articles from these sources into Reader, you'll need to continue to use the browser extension on web or Safari on iOS, both while logged in.

🦐 Minor Improvements

• Our text-to-speech provider Unreal Speech has released an upgraded set of voices based on a newly trained model. This model is still in beta so its voices are not set by default yet, but you can try them by clicking the waveform icon.
• You can now copy your highlights from any document to your clipboard on mobile or share them into another app using the share sheet.
• Muscle Memory Alert Update: Last update, we shared that we made the reading experience feel "firmer" by removing the single tap anywhere on the screen to reintroduce the user interface. Instead, you must either scroll backwards or tap the bottom bar. While experienced users loved this, we found that new users struggled. Accordingly, we've reverted that change and instead added an option for advanced users.
• Documents should now open faster, especially newly saved ones, and text-to-speech should start playing faster, especially on long documents.
• You can now set a toggle for specific email newsletters to always default to the original email formatting rather than parsed text. In addition, Reader now automatically sets these original styles as default for many popular newsletters where this formatting is superior.
• You can now use a keyboard shortcut to expand and collapse all the sections in the left sidebar: Shift + X.
• You can now include published dates as a field of metadata with CSV imports.
• The folders of RSS feeds were improved to exclude documents moved to the Library, consistent with the rest of Feed behavior.
• A “pull to refresh” interaction was added to the Home screen on mobile, consistent with other sections of the app.
• You can now update the desktop app via an option in the native top menu.
• Several of the animations inside the iOS app were upgraded to be smoother.

🐛 Bug Fixes

Since the last update, we've spent a lot of time fixing the bugs you all have reported, and overall making Reader more reliable:

• Fixed issue where search would freeze and/or erroneously show no results
• Fixed issue where mobile would show different search results from web/desktop
• Fixed bug where reading position would not be saved sometimes on first opening of a document
• Fixed bug where scrolling past the first 20 documents in a list on mobile wouldn’t work
• Fixed bug where there would be no back button showing while a document was loading
• Fixed bug where highlighting on web would accidentally scroll you away from the highlight
• Fixed bug where the ios app would sit too low on the screen, being overlapped by the iOS bottom navigation bar
• Fixed bug where mobile app would sometimes crash when transitioning from offline to online
• Fixed bug where app would crash on e-ink and other android devices
• Fixed bug where “paged scroll” default settings would sometimes be forgotten
• Fixed bug where highlight text in the Readwise 1.0 mobile app would overflow
• Fixed bug where Notion export would sometimes duplicate highlights
• Fixed syncing bug where app would not receive new changes from other devices
• Fixed bug where youtube playback speed button wouldn’t show on tablets
• Fixed bugs with autosummarization
• Fixed bug where Suggested Feeds section would load forever for some users
• Fixed bug where long highlights could not be deleted from the Notebook on web
• Fixed bug where white screen would show when opening a youtube video
• Fixed bug with iOS app crashing intermittently
• Fixed bug where tweets would show as the wrong category
• Fixed various breakages with Twitter List Digests due to Twitter’s site/api changes
• Fixed bug where Arxiv PDFs were not saving properly
• Fixed bug where document loading counter would show higher numerator than denominator
• Fixed bug where Android app would crash when scrolling through list of documents with broken images
• Fixed broken formatting on “Download Annotations” button
• Fixed bug where long tweets wouldn’t parse fully
• Fixed bug where highlight action buttons on web would be covered by a youtube video
• Fixed bug where iOS share sheet would crash if you have too many tags
• Fixed bug where adding urls the don’t start with https://
• Fixed bug where Folder dropdown on Manage Feeds page might not show all Folders
• Fixed bug where the top padding for youtube videos on web/desktop would be too small
• Fixed bug where emojis in Folder names and Filtered Views would sometimes format wrong
• Fixed bug with ghostreader keyboard shortcut sometimes adding the letter g to the text input box
• Fixed bug where the readwise.io domain would show in metadata for epubs
• Fixed bug with missing books from Libby highlight imports
• Fixed bug where user preference for “clean view” of emails not saving consistently
• Fixed bug where built in category views would not show when added to Home screen
• Fixed bug where tags would not sort correctly in the tag dropdown
• Fixed bug where tags imported from Instapaper were unable to be deleted in Reader
• Fixed various bugs with Ghostreader autotagging
• Fixed bug where unsaved filtered views on mobile could not be split
• Fixed lagginess with mobile app navigation
• Fixed bug where PDF reading progress could be lost too easily
• Fixed bug in one mobile build where part of the sidebar would be always visible in the reading view
• Fixed bug where empty Views would not show on mobile
• Fixed bug where opening documents directly via url on web would show a loading screen forever
• Fixed issue with PDFs where the note icon wouldn’t appear in the correct position after changing the zoom level
• Fixed bug with the __exact filter query
• Fixed issues with expired ghostreader api keys
• Fixed bug with pinning then unpinning feed folders
• Fixed issues with adding notes to specific ePubs
• Fixed bug where refreshing on web would scroll you to the second item in a document list
• Fixed bug with cover images of exported notion highlights
• Fixed bug where top icons would sometimes overlap with the iOS native status bar
• Fixed bug where relative urls for images/links would not work inside of emails
• Fixed bug with PDFs in the Daily Digest
• Fixed bug where the sidebar on web would show weird extra spacing in some browsers
• Fixed mobile list swipe actions causing list items to get stuck
• Fixed confusing toggle text for tap-to-open
• Fixed bug where Focus Mode would not hide all unfocused elememts
• Fixed bug where documents on web could flash “could not parse document” when loading for the first time
• Fixed bug where advancing to the next document could cause a blank screen
• Fixed bugs in the update flow from the old to new Notion integration
• Fixed a regression in the styling of twitter list digests
• Fixed bug with opening images wrapped in links on mobile
• Fixed bug where sometimes images on mobile wouldn’t load properly inside documents
• Fixed bug where the Table of Contents would randomly collapse as you navigated
• Fixed bug where the mobile side panels could get stuck part-way open sometimes
• Fixed bug where mobile document lists would only show 20 elements when scrolling down
• Text to Speech (TTS)
  
  • Fixed bug where TTS would fail on Android for articles with no cover image
  • Fixed bug where TTS would crash the app in some files with specific newline characters
  • Fixed rare case where TTS would not play the correct audio
  • Fixed bug where TTS button would show on youtube videos
  • Fixed issue with TTS where app would sometimes crash while listening
• Desktop
  
  • Fixed bug in desktop app where links to external urls would not work
  • Fixed bug where the desktop app would crash in a few situations
  • Fixed bug where TTS control bar did not show in the desktop app
  • Fixed issue with the Profile page buttons not working in the desktop app
  • Fixed bug where desktop app would show incorrect cmd/control keys for keyboard shortcuts
  • Fixed the keyboard shortcut o to open the original document in the desktop app
  • Fixed bug where fullscreen videos would not work in Reader desktop app
  • Fixed bug with the Help menu in the mac desktop app being unresponsive

📼 Creator Content

Craig Mod

When we started Readwise in 2017, Craig Mod was the authority on the future of reading digitally. He's written less on this front in the past few years, but we were excited to see his latest Roden newsletter enthusiastically cover his experience with the Boox Palma and Readwise Reader. (Fun fact: One of the best days in Readwise history was when Craig onboarded as user #50, tweeted about it, and drove 40% growth in a single day – a whopping 20 users!)

Steven Johnson

Bestselling author and longtime friend of Readwise, Steven Johnson, has been working with Google to launch its own AI-native note-taking app called NotebookLM. In this interview with Dan Shipper of Every, Steven demonstrates how NotebookLM can unlock surprising insights from your Readwise highlights for creative projects.

👋 Farewell

In my last update, I posted a role for a Principal Product Designer. After chatting with dozens of impressive folks, we're delighted to officially welcome Kris Niles to the team! Kris is a long-time user of Readwise and even built Podhighlighter in his spare time. We're excited to start cooking together.

In addition, we're recruiting for two engineering roles: a Senior/Staff Engineer and an Integrations Engineer. If you or someone you know might be a good fit, please don't hesitate to reach out!

Thank you again for your continued support and please reach out any time 🙏
– Dan, Tristan, & the Readwise team