Overview
Problem
Design Considerations
Iterations
Solution
Reflection
Future Work

AI-Powered Newsletter Assistant

Designing an automated newsletter digest system to accelerate AI learning and combat information overload.

Timeline
2 weeks
(Jun 2025)
Objective
Gain hands-on experience in designing with and for AI

The Problem

Newsletters have become a popular way for professionals to stay current with the rapid advancements in AI, but this has created a critical paradox. While millions of people subscribe, most don't have the time to read them.

Even the best-performing newsletters achieve daily open rates of just 34–50%, revealing a significant gap between the high user demand for information and their actual ability to consume it.

This project addresses the common challenge of making continuous learning in a fast-moving field both efficient and manageable.

Digging Deeper

To understand this challenge deeper, I first deconstructed the current process of engaging with multiple daily newsletters. This manual workflow, combined with the volume of content, revealed several critical pain points.
1
Inbox Overload
Newsletters easily get lost in a sea of other emails. Newsletter readers must first sift through countless messages before even beginning to engage with the content.
2
Unsustainable Time-Commitment
Readers reported that the sheer volume of daily newsletters required a commitment of up to an hour each day. This becomes a tiring, unsustainable habit that ultimately hinders a user's original goal of consistent learning.
3
Finding Personal Value
Without a strong foundation of prior knowledge, readers struggle to make deeper connections between general AI news and its direct application to their individual learning goals.
4
Redundant Content
Readers often have to navigate through repetitive content across multiple newsletters creating a frustrating trade-off: re-read the same information or risk skipping a crucial or unique insight.
5
Information Overload
The desire to not miss out makes subscribing easy, but managing the subsequent influx of information is not. With each new newsletter, readers must repeat a series of tedious steps, which quickly becomes an overwhelming and unmanageable burden.
Analyzing these pain points made it clear that a new approach was needed, leading me to two foundational questions that would guide the design process:
  1. How might I leverage AI to make these newsletters more valuable and relevant to my specific learning goals?
  2. How might I use data to make informed decisions on which newsletters are actually providing the most value?

Design Considerations

This section can only be viewed on desktop :)

Having established a clear understanding of the problem and the key questions to be answered, I began exploring the design considerations that would shape the solution.

1/ LLM Platform

Choosing the right AI platform was critical for creating a customizable assistant that could consistently deliver UX-focused insights. I primariliy evaluated platforms based on customization options and cost-effectiveness, but also considered other factors such as API integration, contextual strength, and scaling potential.

Features
Gemini Gems
ChatGPT Custom GPTs
Open AI Assistants API
Claude Projects

Setup

Custom instructions and system prompts

Custom instructions and system prompts

Code-based

Custom instructions and system prompts

Pricing

Available with free tier

Requires subscription (starts at $20/month)

Pay-per token usage

Requires subscription (starts at $20/month)

API Integration

No direct API integration

No direct API integration

Full programmatic control and API integration

No direct API integration

Key Strengths

Integrates with Google Workspace

Handling complex prompts and tasks

Mature GPT Store (monetization potential)

Broad user adoption

Enterprise-grade API infrastructure

Complete programmatic control

Handling information from long documents

Excellent at coding tasks

Decision: Gemini's free custom Gems enabled unlimited experimentation with UX-focused newsletter analysis without subscription costs, making it the optimal choice for rapid prototyping and iterative prompt engineering. Custom API integration could be explored in the future, but is not necessary for the current scope.

2/ Gemini Interface

After selecting Gemini as the core platform, the next step was to navigate the distinct capabilities of its various interfaces. This was a critical design decision, as the right choice would directly enable the solution's core functionality, while the wrong one would limit its potential. I had to choose between the seamless, in-context experience of the Gmail interface and the powerful, flexible capabilities of the desktop application.

Features
Gemini for Gmail
Gemini Desktop App

Intended Use

Contextual assistant designed for email-based tasks

A full-featured environment for comprehensive tasks

Instruction Handling

Limited to simpler instructions

Handles complex, multi-step instructions

Output Formatting

Does not support detailed formatting or tables

Produces detailed, formatted output, including tables

Decision: The Gemini desktop app was ultimately the more suitable choice because it offered greater control over the output, superior formatting capabilities, and the flexibility for more complex use cases.

3/ Guiding Principles for Gemini's Output

To ensure the solution effectively addressed the core problem and delivered a valuable user experience, I established a set of guiding principles for its core functionality:

Contextual Understanding
The assistant should understand user context and provide clarification when prompts are unclear, recognizing implicit references like "newsletters" without requiring constant explanation.
Personalization & Efficiency
The assistant should remember user learning goals to eliminate repetitive input and deliver personalized experiences.
Actionable Insights
The assistant should actively connect newsletter content to user objectives rather than simply summarizing.

Iterations

1/ Introducing Automation

Choosing the right AI platform was critical for creating a customizable assistant that could consistently deliver UX-focused insights. I primariliy evaluated platforms based on customization options and cost-effectiveness, but also considered other factors such as API integration, contextual strength, and scaling potential.

2/ Refining the Process

While the first iteration of the design successfully addressed all core pain points and served as a functional MVP, I discovered an important opportunity for refinement through ongoing use. The Gemini desktop app could integrate with Gmail directly, a capability I was not previously aware of. This new knowledge allowed me to eliminate the need for Gmail filters or third-party automations like Make, dramatically simplifying the workflow. My focus shifted to refining the Gemini Gem’s instructions, for a smoother and more consistent user experience. This more streamlined solution is illustrated in the below flow chart:

A Breakdown of the Solution

Intuitive Interaction

The Gemini Gem understands user context, so you don't need to specify learning goals or newsletters for analysis.

It can summarize newsletters from a natural language request and will automatically guide or ask for clarification as needed.

Personalized & Contextualized Digests

The Gemini Gem intelligently organizes each digest by relevance to your goals, using clear headings and logical formatting.

It also includes a dedicated section that connects news to your specific objectives, acting as a personal tutor to bridge knowledge gaps.

Relevance Scoring

Each newsletter is scored on its relevance and compatibility, empowering readers to make informed decisions about their subscriptions.

Reflections

This project helped me build practical skills in AI integration for design workflows, which was exactly what I hoped to achieve on my journey toward AI fluency. Through researching and experimenting with LLM platforms and designing the Gemini Gem, I learned to balance baseline contextual information in prompts... too little yielded inconsistent results, while too much created confusion. Across two major iterations with multiple refinements, I gained proficiency in formatting instructions for optimal AI comprehension, determining when to provide examples, and calibrating specificity levels.

The experience shifted my perspective toward the technical side of AI, designing the backstage components that dictate user experiences with LLMs. Understanding AI limitations helped me identify where these tools can streamline workflows versus where human oversight remains critical, and how the prompt architecture shapes user interactions.

Navigating the rapidly evolving AI landscape proved challenging initially. Discovering new tools could completely derail my workflow for a day, because there was so much new information to take in. However, by persevering and immersing myself in the AI ecosystem, I became more efficient at evaluating new tools. Over time, I developed the skills and confidence to quickly assess whether a tool was worth deeper exploration.

Future Work

Moving forward, I plan to focus on three key areas to enhance this project's impact and usability:
Scaling for Broader Adoption
I'll explore strategies to accommodate a larger user base, focusing on system architecture and data management approaches that will support increased usage without compromising the user experience.
Onboarding Experience Design
The next phase will involve designing UI components that collect information about users' goals and newsletter subscriptions during onboarding. This data will be used to customize Gem instructions, ensuring the tool provides personalized and relevant assistance from the start.
Continuous Evaluation and Refinement
I'll implement ongoing usability testing and establish metrics to track the tool's success over time. Through regular personal use and user feedback collection, I'll identify pain points and opportunities for iterative improvements to the overall experience.
Keep Reading
PC Express

Bridging the gap between online and in-store grocery shopping experiences

Rent Smart

Helping renters feel prepared and organized when searching for housing in Toronto

About Me

Who am I and how did I get here? Learn about why I transitioned from engineering into UX and more fun stuff