Customize Discover with Natural Language

🔑 Enhanced Key Takeaways

•The natural language customization feature is part of a broader set of personalization upgrades announced by Google, which also includes a 'Preferred Sources' button for publishers to embed on their sites and customizable daily audio briefings in the Google News app.
•This new natural language input is facilitated through a chat interface within the Google Discover feed.
•Users can specify not only topics but also desired content formats (e.g., more videos), specific sources (e.g., from a particular publisher), and even the overall tone of their feed (e.g., calmer, more relaxed).
•The feature is designed to provide immediate updates to the Discover feed based on user input and to remember these preferences for ongoing personalization.
•Google Discover, originally launched as Google Feed, relies on advanced AI and machine learning algorithms that analyze a user's search history, app activity, web interactions, and location to curate its personalized content.

📊 Competitor Analysis▸ Show

Feature/Platform	Google Discover (with NL Customization)	Apple News	Flipboard
Personalization Method	Users describe preferences using natural language prompts (topics, formats, tone, sources) via a chat interface.	Users interact with 'Suggest More' or 'Suggest Less' buttons, follow specific channels/publications, and block unwanted sources.	Users select initial topics, 'tune' topics to focus on sub-topics, follow specific sources, and engage with stories (like/dislike) to refine recommendations.
Algorithm Basis	AI and machine learning analyze user activity, search history, web interactions, and location, enhanced by natural language processing.	Curates feed based on reading history, user interactions (likes/dislikes), and followed channels.	Recommendation engine learns from selected topics and user engagement, with explicit 'tune' options for deeper personalization.
Content Types	News articles, blog posts, videos, product recommendations, social posts (X, Instagram, YouTube Shorts).	News articles, videos, local news, and premium content from magazines and newspapers (with Apple News+).	Articles, videos, social media posts, and user-curated magazines across various topics.

🛠️ Technical Deep Dive

Google Discover's core recommendation system is powered by AI and machine learning algorithms that analyze various user signals, including Web & App Activity, Location History, search history, and interactions with Google products like YouTube.
Natural Language Processing (NLP) is a fundamental component, enabling Google to understand the intent behind user queries and the context of web content.
Advanced NLP models like BERT and the Multitask Unified Model (MUM) are crucial for Google's ability to interpret complex, natural language input.
MUM, specifically, is noted for being 1,000 times more powerful than BERT, capable of understanding 75 languages and processing information across multiple modalities (text, images, and potentially video/audio).
Natural language recommender systems often employ semantic embeddings, which are numerical representations of text that capture meaning, stored in vector databases to find items most similar to a user's natural language query.
Large Language Models (LLMs) are increasingly being used in recommendation systems to understand user intent expressed in plain language, moving beyond traditional keyword matching.

🔮 Future ImplicationsAI analysis grounded in cited sources

User engagement with Google Discover will significantly increase due to enhanced control.

By allowing users to precisely articulate their content preferences, the feed will become more relevant and less prone to showing unwanted content, leading to higher user satisfaction and interaction.

Content creators and publishers will need to adapt their content strategies to align with nuanced natural language queries.

As users specify preferences for topics, formats, and tones, publishers will be incentivized to produce content that directly addresses these granular demands to appear in personalized feeds.

Google's underlying AI models for content understanding and recommendation will become even more sophisticated.

The direct natural language feedback from millions of users will provide an invaluable dataset for training and refining advanced NLP and recommendation algorithms, leading to more accurate and contextually aware AI.

⏳ Timeline

1997

Google launched, with its search algorithm prioritizing relevance over keyword stuffing.

2016-12

Google Feed, the precursor to Discover, was introduced as a personalized content stream on Android devices.

2018-09

Google Feed was rebranded as Google Discover, featuring an updated design and enhanced personalization capabilities.

2021-05

Google introduced the Multitask Unified Model (MUM), an AI designed to understand complex queries across multiple languages and formats, foundational for advanced natural language understanding.

2025-09

Google Discover expanded its content to include posts from social platforms (X, Instagram, YouTube Shorts) and launched a 'follow' feature for publishers and creators.

2026-08-20

Google officially announced the 'Customize Discover with Natural Language' feature, allowing users to describe content preferences in their own words.

Customize Discover with Natural Language

⚡ 30-Second TL;DR

Key Points

🧠 Deep Insight

🔑 Enhanced Key Takeaways

🛠️ Technical Deep Dive

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline

📎 Sources (30)

👉Related Updates