Hey all,

I'm happy to publish public beta newsletter #8 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.

This is a fun update chockfull of crowd pleasing new features. Before I get into those, however, I want to share that we've begun recruiting for a Principal Product Designer. If you or someone you know might be a good fit, we'd love to hear from you!

On to the product updates:

• 📁 RSS Folders on Web – You can now organize your RSS feeds into folders and pin individual feeds to the left sidebar resulting in a more traditional feed reading experience.
• 🔊 Text-to-Speech on Web – You can now use text-to-speech in the web and desktop apps just like the mobile apps.
• 🔎 Improved Full-Text Search – You will now find full-text search to be vastly more accurate across all your library's content and metadata. (If you were ever frustrated by search, you should try again.)
• 🎛️ Custom Ghostreader Summaries – You can now customize the GPT prompt used to generate document summaries to suit your bespoke preferences.
• ♻️ Resurfacing in Summaries Emails – You will now find two randomly selected items from your backlog resurfaced alongside your recent saves in your summaries emails.
• 📧 Email Original Styles – You can now toggle email documents between a clean parsed view and an original styles view.
• 📑 Automatic PDF Metadata – You will now find that most of your research papers pull robust metadata from the internet automatically.

RSS Folders on Web 📁

The next time you open Reader on web or in the desktop app, you'll notice that the left sidebar has evolved into a left side panel. There are many advantages to this layout, but the primary motivation was to make way for a more traditional feed reading experience for RSS users. This includes folders, pinned feeds, and drag and drop interactions.

The fastest way to create these RSS folders is from the Manage feeds page.

Here's a quick Loom video (less than 1 min at 2x) demonstrating how to use the new left panel.

As long-time readers of this newsletter know, I can wax poetic about the behind-the-scenes process leading to certain product outcomes. I'll spare you such narcissism here other than to share that adding a new organizational concept to an all-in-one product like Reader is quite challenging. At least, it's challenging if you try to add something new without creating what feels like more complexity. This is why a seemingly straightforward feature for a reading app like "RSS folders" actually demanded a refactor of the entire left sidebar.

Whenever a user interface you're accustomed to is changed, as we've done here, there's often a brief adjustment period during which you might feel a tad uncomfortable. Once you've reacclimatized, however, we hope you'll agree that the advantages of this wider left panel outweigh the negatives. Here are a few:

Information Density. You can now fit several more items into the left sidebar using the same amount of vertical space without feeling overwhelmed.

After and before, for comparison.

Discoverability – You can now navigate the app with more confidence due to the addition of text labels beside the icons making it clearer what each item represents.

Collapsibility – You can now hide the left panel entirely ([) or collapse all items in the left panel when you want a cleaner view.

Symmetry – You'll now find the document list view to feel more balanced between left, right, and center. Opinions differ, but in my experience, many discerning users have expressed that the document list items were too wide for their tastes, so the widened left panel doesn't consume screen area at the expense of otherwise valuable real estate.

The left panel reduces the width of the document list item, but in a good way.

As part of this update, you can also drag and drop document tags and saved views to more easily pin them to the left panel.

RSS folders has been the 7th most requested feature in Reader and is now well underway. This is the first of several milestones of a larger effort to build a more intuitive organizational abstraction on top of filtering without eliminating the power and flexibility for more advanced users. The next milestone is to adapt these changes from web to the mobile user interface.

Props to Mati for grinding through all the details required to implement such a tightly coupled, intricate feature.

Text-to-Speech on Web 🔊

You can now use text-to-speech in the web and desktop apps just like you can in the mobile app.

Like all features on web, there are corresponding keyboard shortcuts. The primary shortcut to learn is P (for Play) which will stop and start the text-to-speech.

Text-to-speech in the web and desktop apps was the 3rd most requested feature. For the record, we obviously take user feedback into consideration when deciding what to work on, but it's not the only consideration. I'm just sharing this feature request data for fun because this particular update happens to contain several highly requested features.

Props to Adam for leading this seamless adaptation.

Improved Full-Text Search 🔎

If you've ever tried searching for a document in Reader and were frustrated by the results, you should try again. We've overhauled the full-text search algorithm to be vastly more accurate. In particular, multi-word queries across both document content and document metadata will generally return exactly the results you'd think they should. In addition, search is faster and consistent across web and mobile.

For some background, Reader's full-text search has to work offline on the client meaning we couldn't simply rely on a server-side search library such as Algolia or Elasticsearch.

Search v2 has been the 4th most requested feature. Technically, we internally reserve the term "Search 2.0" for when we merge filter queries into the search bar, but we suspect roughly 75% of the requests to date will be addressed by this Search v1.5 milestone.

Props to Mitch for writing the improved search code by hand.

Custom Ghostreader Summaries 🎛️

This is a fun one. Every time you save a new document to Reader, it's automatically summarized by GPT-3.5 (unless you toggle auto-summarization off in your account settings). You can now take control of the prompt used to generate this summary enabling all kinds of creative alternatives for your specific use case. These summaries then flow through to your document metadata, summaries email, and note-taking exports.

This interface uses the Jinja2 templating language and custom variables we've set enabling you to generate prompts with the underlying document content and metadata.

The most basic use case here is to rewrite the prompt in your native language. This is by far the most reliable way to coax the large language model – trained on massive amounts of English text – to not fallback to English in its response. But if you're comfortable with GPT and willing to learn some of Ghostreader's template variables, you can do so so much more.

Here are some examples from our team:

• My personal prompt checks if the document is a Twitter List digest. If so, GPT is prompted to tell me about the current thing or main character the very online people are talking about that day. Otherwise, the document is summarized normally.
• Eleanor's prompt checks if the document title includes a clickbaity question such as "Are people getting stupider?" (Apparently these really annoy her.) If so, GPT is prompted to write a no-BS summary answering the question. Otherwise, the document is summarized normally.
• Eleanor's prompt also checks if the document is tagged Shared from Dan via a domain-specific bundle I made for her and the team. If so, GPT is prompted to identify the key takeaways I'm likely to want her to focus on. (Praise.) Otherwise, the document is summarized normally.
• Erin's prompt analyzes the creative devices used by the author in their writing, such as dialogue, flashbacks, and historical examples to fuel her creative nonfiction practice.
• Abi's prompt attempts to generate a summary in the style of our Wisereads newsletter with a value-added, matter-of-fact introductory sentence followed by a choice excerpt pulled from the text. (Don't worry: No artificial intelligence is or will be used in the writing of Wisereads.)
• Tristan's prompt reduces each document to three sentences separated by new lines, each beginning with an emoji (a picture says a thousand words).

You get the idea. Here's a quick Loom video (less than 5 mins at 2x) demonstrating the new left panel.

To learn more and to see these underlying prompts, you can visit our Ghostreader guide from within the app (work in progress). If you'd like some help crafting a prompt for your use case, or to just be inspired by what others are trying, we'll be hosting a live workshop in our Discord server on Wednesday, April 10 at 11AM EST.

As part of this update, you can also add your OpenAI token to optionally use GPT-4 or GPT-4-Turbo to generate your summaries (using your own API credits). The results are indeed better, but be careful not to accidentally send an entire ebook to GPT-4 or something. That will get expensive, fast.

Bring your own GPT-4 token (5th most requested feature), Set default language for Ghostreader output, and Save custom Ghostreader prompts are partially done while Generate summaries from highlights and Produce Ghostreader summaries in the document's own language are achieved here.

The next step in the Ghostreader roadmap is to enable customization of the manual prompts used while reading such as Define, Translate, and Simplify. You'll be able to make these whatever you want. Also, if you play with custom summaries, you'll notice the preferences section includes an experimental option for tagging your documents with Ghostreader (2nd most requested feature). We're still iterating our way to the optimal user experience of automatic tagging, which is why I didn't get into that feature here. But if you're adventurous you can try it out.

Props to Hannes for singlehandedly leading all our AI features.

Resurfacing in Summaries Emails ♻️

Whenever you save at least 2 new documents to Reader, you'll receive a beautiful email containing their summaries the morning after. If you customize your summarization prompt per the update above, this special email will use those personalized summaries instead.

In addition, you should notice a new section at the end of this email resurfacing two documents at random from your backlog. This is a great way to serendipitously stay on top of your reading list.

Props to Hannes again.

Original Email Styles 📧

Compared to web articles and blog posts, emails are particularly tricky to parse into clean, distraction-free HTML. This would lead to situations where the parsed email would be less delightful to read in Reader than in your email client. To solve this problem, you can now toggle emails between a parsed HTML view and a new original styles view that preserves the formatting and layout of the email itself.

Dense Discovery is a great example of a newsletter with a unique layout and significant formatting.

There are some small tradeoffs, which is why we didn't do this in the first place. For example, most emails will force a white background (ruining dark mode) and you can't adjust the text appearance or navigate via keyboard. But fortunately you can still highlight the original view so it's worth using on heavily formatted newsletters.

This feature works on both web and mobile. You toggle it on same way you'd switch to enhanced text view in a PDF. You can also find a command in the Command Palette to set particular email senders to always default to original view.

Many team members were involved in this feature, but props to Tristan for swiftly executing the first 80%.

Automatic PDF Metadata 📑

If you've ever uploaded a PDF to Reader, you've probably encountered some unexpectedly weird title and author metadata. That's not a bug. Those are, we promise, the actual metadata contained in the PDF. The problem is that for some reason no one ever sets the proper metadata on the file level. Even official looking scientific research papers. Now whenever you upload a PDF, Reader will attempt to match it against various online databases to enrich the document with the right title and author.

If you show a PDF in your reading app, it must be either the Bitcoin whitepaper or Attention Is All You Need. Sorry, I don't make the rules.

This feature typically works on research and scientific papers. If you upload a PDF that's not tracked in a database online, you'll still need to manually edit the document's metadata to fix the title and author.

Props to Bruno for figuring out a waterfall of techniques to grab a PDF's DOI. Improved metadata extraction for PDFs and scientific articles was a top 20 most requested feature.

Notion Official API Export 👩‍⚖️

When we first released our Notion export in 2020, there was nothing else like it. Now it's become table stakes. At the time, Notion didn't even offer a public API so we had to resort to some clever-but-unofficial techniques to enable the integration. Since then, Notion has released a public API. Early versions didn't have all the capabilities we needed to not lose functionality, but that's since been resolved.

If you export your highlights to Notion, you'll soon receive an email with instructions on how to safely migrate to the new version of the Readwise integration with the official Notion API.

Props to Bruno and Tristan for figuring out this tricky migration.

Coming Up 🔜

• RSS Folders on Mobile – In this update, we refactored the user interface to enable pinning and foldering of RSS feeds. We're now implementing a mobile analog which will incidentally improve some of the information architecture in the apps.
• Custom Ghostreader Prompts – In this update, we enabled you to customize the Ghostreader prompt used for summaries. You'll soon be able to do the same for the Ghostreader prompts used throughout your reading experience enabling an infinitude of creative use cases.
• Readwise 1.0 – Mitch went on a side quest to improve full-text search accuracy interrupting his infrastructure work enabling bidirectional sync between Readwise 1.0 and Reader (right now it's one-way).
• Paywall Enhancements – Since we've been working on Reader, many of the most popular content providers such as The New York Times, Wall Street Journal, and Medium have dramatically beefed up their paywalls. You can still save full articles while logged in using the browser extension on web or Safari on iOS, but this can sometimes be inconvenient. We're attempting to solve this problem once and for all using the most overpowered nuclear approach one can think of.

Text-to-Speech Minor Improvements 🦐

Since the last update, Artem has shipped a bevy of smallish improvements to the text-to-speech feature, which I'm grouping here:

• On long documents (eg ebooks), the audio now loads much faster and the position more accurately when starting mid-document.
• Text-to-speech now saves your reading position even while listening in the background.
• Your playback speed should no longer mistakenly reset.
• The word indicator should fail less frequently when wrapping words at the end of a paragraph.
• The player bar now uses less vertical space in general and the (more intuitive) user interface doesn't cover the text when you scroll away from the tracked word.

Ebook Minor Improvements 🦐

Artem has also cooked up a batch of smallish improvements to the ebook and e-ink experience, which I'm grouping here:

• EPUBs are now set to paged scroll by default. You can of course undo this default if you prefer scroll mode.
• EPUBs should now load notably faster than before.
• While loading, EPUBs will display the book cover on mobile.
• You can optionally toggle on a "high contrast mode" in the mobile settings page to improve the visibility of highlights and other user interface on e-ink devices.

Other Minor Improvements 🦐

• MUSCLE MEMORY ALERT: Artem made some subtle improvements to taps and swipes on mobile to make the reading experience feel more solid. Notably, a single tap of the reading view will no longer reintroduce the user interface. Instead, you must either scroll backwards or tap the bottom bar. We've tested this thoroughly and we think it's quite an improvement once you retrain your muscle memory, but we do know some folks have gotten tripped up by this change. If you feel strongly this was a mistake, we're curious to hear from you.
• Artem improved the accuracy of saving your reading position between sessions when in paged scroll mode.
• Bruno added the ability to disconnect any export integration within Readwise 1.0.
• Arek added the ability to share your annotated emails from within the mobile apps.
• Hannes added a toggle to disable Ghostreader auto-summarization.
• Tadek added support for YouTube RSS feeds enabling you to subscribe to all new videos of a specific channel.
• Rasul added an Unsubscribe button for email newsletters on the mobile apps.
• Cayla cleaned up the copy across all automated emails and UX copy in a few places in the apps.
• Bruno added support for new Twitter Articles format (different from long tweets).

Bug Fixes 🐛

• Props to our QA Specialist Eleanor for 10xing our bug handling.
• Fixed bugs where clients would get stuck out-of-sync with other devices.
• Fixed bug where the status bar would cover the top of the app on newer Android OS versions.
• Fixed bug where book cover images were not loading.
• Fixed bug where clicking links would sometimes fail in paged scroll mode.
• Fixed issue where “Save URL via clipboard” would clutter the bottom of the screen.
• Fixed bug where sometimes PDFs sent via email would not import.
• Fixed bug with exporting annotations from the desktop apps.
• Fixed several random bugs with Ghostreader.
• Fixed bug where the counts of documents inside saved views wouldn’t update in the mobile apps.
• Fixed bug where the Listen button wouldn't tap in paged scroll mode.
• Fixed formatting of embedded tweets in paged scroll mode.
• Fixed bug where highlighting the beginning of a YouTube video transcript was prevented.
• Fixed bug where YouTube transcript position wasn’t saving on mobile.
• Fixed bug where the header wasn't formatting properly on shared documents.
• Fixed bug where YouTube videos would freeze after starting.
• Fixed bug where links from Google search results weren't saving.
• Fixed bug where the web highlighter wasn't working properly with Wikipedia links.
• Fixed bug where bulk deletions weren't deleting in full.
• Fixed bug where long Twitter threads were cutting off.
• Fixed multiple bugs with tags using spaces or special characters.
• Fixed bug where sorting by published date wasn't working properly on mobile.
• Fixed bug with Substack confirmation emails not appearing in Feed.
• Fixed bug where recently used tags were not rising to the top of the dialog.
• Fixed bug where tags wouldn't rename.
• Fixed bug where edited document titles were not exporting to note-taking apps.
• Fixed bug where highlights would not export to Reflect automatically.
• Fixed bug where the right sidebar would flicker in Safari.
• Fixed bug where bundles with spaces were not saving properly.
• Fixed bug where the transition between the first and second pages in paged scroll mode would glitch.
• Fixed bug where some tweets were being saved as articles.
• Fixed bug where domain__not and url__not queries were not working on web.
• Fixed bug where videos were not showing their length in the metadata panel.
• Fixed bug where the MacOS app would always show “Downloading update…” in the Profile context menu.
• Fixed bug where the Google Docs export integration would not complete.
• Fixed bug where deleted highlights would still appear in the API results.
• Fixed bug where inline links in paged scroll mode wouldn't navigate properly.
• Fixed bug where minutes filter key wasn't working.
• Fixed bug where podcast highlights wouldn’t export to Logseq.
• Fixed bug where the Daily Digest UI wasn't formatting on tablet devices.
• Fixed bug where auto-highlighting would stop working after auto-advancing.
• Fixed bug where account deletion would take too long and time out.
• Fixed bug where Term Lookup (and a few other Ghostreader prompts) wouldn't grab the proper context on mobile.
• Fixed bug where PDF and EPUB cover images wouldn't load on the mobile home screen.
• Fixed parsing issues with theverge.com, medium.com, economist.com, bloomberg.com, bbc.com, nytimes.com, theatlantic.com, wsj.com, tikr.com,

Creator Content 📼

Maneetpaul Singh

Last update, we shared the first video in a series where Maneetpaul Singh would break down the intriguing but maddeningly complex BOOX e-ink product lineup starting with the phone-sized Palma. Since then, he's also reviewed the Boox Tab Mini C and the Boox Page. He even found the time to create what might be the best video yet made on Reader: Readwise Reader: Top 15 Features, Tips, and Tricks!

Tiago Forte

Led by Steven Johnson, Google recently launched its own AI-native note-taking app called NotebookLM. Our friend Tiago Forte made a walkthrough video teaching newcomers how to get the most out of the tool including feeding it his Readwise highlights via our Google Doc export integration.

Brandon Boswell

Brandon Boswell released a new video demonstrating Reader's paged scroll mode on his BOOX Note Air 3c tablet alongside his preferred settings for optimal reading experience.

David Kadavy

Long-time friend of Readwise and author David Kadavy is the first person we know to try Reader on a Boox Poke5. In a recent blog post, he shares the pros and cons of the device, as well as how Reader performed.

Farewell 👋

Every time I get to the end of a product newsletter approaching 4,000 words I think to myself, "We should send these more often so they're not as long." We're working on doing better with our communications, I promise!

As a reminder from the introduction, we've begun recruiting for a new Principal Product Designer. After several years with us, Jesse recently rejoined some old friends so we need to fill his (very large) shoes. If you or someone you know might be a good fit, we'd love to hear from you!

Thank you again for your continued support and please reach out any time 🙏

– Dan, Tristan, & the Readwise team