Making Storytell smarter: Enhanced instructions and new Chrome extension feature

May 16, 2025

In this week’s engineering demo, our team shared updates on key improvements to Storytell, including an overhaul of instruction handling for our AI, enhancements to the Chrome extension for faster and more relevant data interaction, as well as targeted bug fixes that significantly improve platform stability and reliability. These updates were driven by invaluable user feedback, and they represent our continued commitment to making Storytell a more effective, user-friendly tool for everyone.

Improved instructions for smarter, more relevant answers

This week, Alex from our engineering team presented a series of improvements to the way Storytell handles prompts and uses data references when answering questions. Historically, users sometimes experienced overly narrow answers—particularly when the platform would limit its response to only the references provided, even if more context would have yielded a more complete answer.

What changed?
Alex revised the instructions that our models use to process user prompts, especially when a conversation combines both personal collections and public knowledge. For example, in a previously challenging use case, a user might have uploaded NBA stats for LeBron James and then asked Storytell to compare them with stats for other players. Before this update, if the dataset only included LeBron, Storytell would refuse to answer about players not in the references.

Now, with the new instructions:

Storytell will attempt to supplement available references with relevant public knowledge when needed, but will always identify the source.
If an answer combines data from both the uploaded reference and public data (or training data), Storytell provides clear attribution in its response.
When users inquire about the Storytell platform ("Can Storytell do X?"), Storytell now guides them to our official docs page—helping users easily access authoritative information.

Benefit for users:
You’ll receive more complete, contextually aware answers, with transparency about where the data is coming from. This reduces unnecessary refusals and makes Storytell a more valuable research and discovery assistant, whether you’re working entirely within your own documents or leveraging broader public data.

Reducing prompt errors for a reliable experience

Reliability is a core pillar of the Storytell platform, and Alex also detailed engineering work that has resulted in a notable drop in prompt-related errors.

What changed?

Integrations were refined, particularly with large language models, to improve stability.
A key metrics dashboard in Grafana now tracks the error percentage: recent changes have reduced error rates from about 13%-15% to roughly 3%, even with thousands of prompts processed per day.
Improvements in document ingestion and data extraction—especially for PDFs—have made these processes much more stable.

Benefit for users:
With significantly fewer prompt errors, Storytell users can trust their queries and data processing to complete reliably. This improvement particularly benefits power users who work with high data volumes and expect seamless handling of various file types.

Persistent cookies and preferences across pages

A frequent user pain point was the inconsistent saving of preferences and cookies on the platform. Before, settings (such as model selection or prompt improvements) would reset on each page, slowing down workflows and causing frustration.

What changed?

Now, once a preference is set—say, choosing the Gemini 2.5 Flash model—it persists across all pages, not just the one where it was selected.
This fix ensures a more consistent and efficient experience as users navigate and experiment with different features.

Benefit for users:
Settings "stick" throughout a session, letting users focus on their work instead of repeatedly re-applying choices. This small but impactful change saves time and smooths out the user experience, especially for those who frequently switch between models and feature toggles.

Chrome extension: Instant access from text selection

Andi, another core team member, introduced a game-changing update for the Storytell Chrome extension. The extension now lets users instantly engage with relevant content from any web page via a text selection pop-up.

What changed?

Any time text is selected on a page—with the extension installed—a pop-up appears, providing instant options to interact with Storytell based directly on the selected content.
Users can quickly "chat" with Storytell about the selection, request a summary, or receive highlighted takeaways, all without needing to manually initiate a new conversation or find the relevant prompt.
The extension creates a new thread for each interaction, attaching just the selected content to keep conversations focused and relevant.
There is built-in flexibility for those who want control over the feature. Plans include a settings button allowing users to disable or adjust the pop-up if desired, with clear links to documentation for managing the extension.

Benefit for users:
This creates a more seamless, integrated workflow in the browser. You can now pull insights, generate summaries, and interact with web content faster—without context drift or manual navigation. The feature also solves a specific pain point highlighted by Storytell’s frequent users: when browsing complex pages with multiple articles, you can target precisely what content Storytell should analyze, resulting in more relevant outputs.

Ongoing refinements:
Andi acknowledged that, while the feature works very reliably in most cases, there are occasional issues when the sidebar is open while sending content due to timing mismatches. These are actively being addressed to further smooth the user experience. Feedback from Chrome Store reviews (such as summaries being too verbose for small content selections) is also influencing prompt design to create snappier, more concise summaries tailored for the browser context.

Conclusion: Continuous improvement, focused on you

This week’s engineering demo underscores Storytell’s ongoing dedication to a reliable, intuitive, and powerful AI assistant experience. Major highlights—smarter handling of instructions and references, dramatically lower error rates, persistent preferences, and a vastly upgraded Chrome extension—are already delivering measurable improvements for users.

As always, product evolution is guided by real user feedback and active listening. If you haven’t yet tried Storytell, now is a great time: the platform keeps getting smarter, faster, and easier to integrate into your daily information workflows.

Ready to experience Storytell for yourself? Visit our website or install the Chrome extension to see these updates in action.