Shu Zhang
中文
← Selected work
03 — AI & Personalization Systems

Designing AI-Powered Ad Personalization for Content Feeds

Exploring how ads could adapt intelligently to user context, content, and creative assets — designing a three-layer personalization framework for consumer advertising in news feeds and content surfaces that moves beyond static targeting to dynamic, context-aware experiences.

Role
Design lead — Personalization framework & strategic exploration
Partners
Gary Anderson (Head of Ad Design), Vincent Muedra, Global design team
Timeline
H1 2024 (exploration & framework development)
Scope
Consumer ad personalization system design for content feeds & news surfaces

Most advertising platforms personalize by targeting users. But what if ads themselves could adapt — changing format, content, and presentation based on who's viewing, what they're reading, and what creative assets are available? That's the question behind the personalization framework I designed for consumer advertising in news feeds and content surfaces in H1 2024.

The problem: personalization stops at targeting

Traditional ad personalization asks: "Which ad should we show this user?" The system picks from a fixed set of ads based on user profile, then displays the winner. The ad itself never changes — same creative, same format, same message.

But that model breaks down in modern content experiences like news feeds and story streams, where:

The real opportunity wasn't just which ad to show — it was how the ad itself should adapt to make the experience feel native, relevant, and contextually intelligent.

The three-layer personalization framework — ads adapt based on user signals, content context, and creative assets, working independently or together depending on what's available.

The framework: three layers of intelligent adaptation

I designed a personalization system built on three independent but composable layers. Instead of requiring all signals to be present, the system adapts based on what it knows — making ads smarter even when user data is limited.

Layer 1: User Awareness

When we know about the user — their profile, interests, preferences, past behavior — ads can adapt to match personal relevance. This layer handles traditional personalization signals but gracefully degrades when users aren't signed in or haven't built a preference history.

Layer 2: Context Awareness

Even without knowing the user, we know the context: what content they're reading, what query brought them here, where the ad appears on the page. Ads should adapt to feel native to that context — a travel ad in a travel article should look different than the same ad in a finance article.

Layer 3: Ad Awareness

Every ad has multiple creative assets — images, headlines, descriptions, CTAs. Instead of showing the same fixed creative every time, the system should adapt the ad itself based on available space, context, and what's likely to resonate. This layer makes ads feel tailored even when we don't know the user at all.

The key insight: these layers compose. If we only have context (Layer 2) and creative assets (Layer 3), we can still deliver a better experience than a static ad. If we have all three layers, the experience becomes deeply personalized.

Three strategic ad personalization concepts showing personalized content units, interactive personalization, and holistic responses mapped to user, context, and ad awareness layers.
Strategic concepts built on the framework — from personalized content units to interactive refinement to holistic responses.

Strategic concepts: "Contextual Assets Optimization"

Beyond the core framework, I designed several forward-looking concepts that explored how intelligent ad personalization could transform the consumer experience. I created a new strategic category called "Contextual Assets Optimization" to describe this opportunity space.

Personalized Content Units

Ad experiences that adapt their entire presentation based on query context, product category, and user preferences. Examples included dynamic "showroom" experiences and adaptive filters that change based on what matters most in that context (price for budget searches, features for comparison searches, availability for immediate-need queries).

Interactive Personalization

Rather than showing a static ad, users could refine relevance through lightweight interaction. Sponsored content units with opportunities to refresh offers, signal preferences, and better align results with interests — making personalization a two-way conversation instead of a one-way guess.

Holistic Responses

Blending organic content and sponsored content into unified answer experiences, where the personalization framework ensures ads feel like helpful recommendations rather than interruptions. The same three layers (user, context, ad) determine not just what to show but how to integrate it naturally.

Impact and influence: from exploration to strategy

This was north star exploration work — not a shipped feature with metrics, but strategic design thinking that influenced how the organization approaches ad personalization. The framework I designed became a reference point for future monetization strategy.

What I delivered

Presented to and validated by

The work was presented to Gary Anderson (Head of Ad Design), Vincent Muedra, and the global design team, with concepts prepared for leadership review. Gary Anderson explicitly recognized the personalization framework as the strategic direction he was looking for, and the work was consolidated into broader monetization innovation planning.

The exploration influenced ongoing personalization strategy discussions and became a reference for how AI capabilities could transform consumer ad experiences in news feeds and Microsoft content surfaces.

Why this matters: designing for intelligence, not just interaction

Most ad design work focuses on how ads look (visual design) or how users interact with them (UX design). This work operates at a different altitude: how ads should think — the logic that makes them contextually intelligent.

That's the hardest design problem in advertising: not "what button goes where" but "what signals should the system consider, how should they combine, and how should the ad adapt as a result?" It's system design, not surface design.

This work also completes a narrative arc across my AI-related projects:

Each case shows AI at a different layer: optimization → creation → intelligence. The progression demonstrates how I think about AI not just as a tool for automation, but as a system that should understand context and make experiences feel adaptive rather than algorithmic.

Exploration work, not shipped features

This case study describes strategic exploration and framework design, not a launched product. The value is in demonstrating how I approach complex, unsolved problems and design system logic that could guide future implementation. The framework influenced monetization strategy, but I'm not claiming shipped metrics or direct product outcomes.