Digital advertising is entering a new phase.
For years, dynamic product advertising has relied on algorithms to decide which products to show, which customers to target, and when an ad is most likely to generate a click or purchase. But as product catalogs become larger and customer behavior becomes more complex, traditional recommendation and retrieval systems are reaching their limits.
Artificial intelligence is now changing that equation.
New research into AI-powered Dynamic Product Ads (DPAs) shows that large language models can help advertising systems understand user intent more deeply and discover products that conventional keyword and embedding-based systems may miss. More importantly, researchers are finding ways to use AI selectively rather than running expensive AI models for every user.
One recent research project, called SMART (SeMantic-aware Adaptive ReTrieval), demonstrates this approach. Researchers found that a lightweight quality gate could identify users who actually benefit from semantic AI-powered product discovery and route only those users through the more expensive large-language-model pipeline. In experiments involving millions of users, the approach reportedly reduced LLM costs by about 90%, while a two-week online A/B test at Snap showed a 27.6% improvement in ad conversion rate compared with a strong embedding-based baseline.
The significance goes beyond one advertising system.
It points toward a future where product advertising becomes more personalized, context-aware, efficient, and increasingly automated.
What Are AI-Powered Product Ads?
AI-powered product ads are advertisements that use artificial intelligence to determine what product should be presented to a particular user and, increasingly, how that product should be presented.
Traditional advertising systems typically depend on signals such as:
- Search queries
- Previous clicks
- Product views
- Purchases
- Demographic information
- Browsing behavior
- Product categories
- Keywords
- Similarity between products
AI adds another layer: semantic understanding.
Instead of simply matching a user's behavior with a product's keywords or embedding, an AI system can interpret broader intent.
For example, imagine a customer who has recently searched for:
"Minimalist workwear for summer"
A traditional system may primarily retrieve products tagged with terms such as "summer," "workwear," "office," or "clothing."
A more advanced AI system can potentially understand that the customer is looking for a particular combination of style, occasion, season, and aesthetic.
That creates an opportunity for advertising platforms to move from simple product matching toward intent-based product discovery.
Why Dynamic Product Advertising Is Getting More Difficult
Modern eCommerce catalogs are enormous.
A major retailer can have millions of products, with constantly changing inventory, prices, descriptions, promotions, and availability.
At the same time, shoppers do not always behave predictably.
Someone may browse running shoes and then purchase a fitness watch. Another shopper may search for a birthday gift without knowing exactly what product they want.
This creates two major challenges for dynamic advertising:
Retargeting and prospecting.
Retargeting is relatively straightforward. The system already has evidence that a customer is interested in something.
Prospecting is harder.
The system has to identify products that the customer has not explicitly demonstrated interest in yet.
This is where semantic AI can become particularly useful.
The recent SMART research specifically examined this problem in Dynamic Product Ads. The researchers found that traditional lexical retrieval worked well for retargeting, while LLM-generated queries were particularly useful for prospecting and discovering new categories.
That distinction is important because it suggests that the best advertising system may not be entirely AI-driven.
Instead, different technologies can handle different parts of the customer journey.
The Problem With Using Large AI Models for Everything
There is an obvious problem with this approach.
AI inference costs money.
Every time an AI model processes a request, the underlying infrastructure consumes computing resources. At small scale, this may be manageable.
At advertising-platform scale, it can become enormous.
Imagine an advertising platform processing millions or billions of opportunities to display products. If a large language model has to analyze every user and generate semantic queries for every advertising decision, the computational cost can quickly become impractical.
This is one of the biggest challenges facing AI-powered advertising.
The question is no longer simply:
"Can AI improve advertising?"
The more important question is:
"Can AI improve advertising economically at massive scale?"
That is where recent research becomes especially interesting.
McKinsey has also highlighted the growing importance of reducing AI inference costs, arguing that the next stage of AI advancement may depend as much on making inference cheaper and more efficient as it does on developing more powerful models.
The Smarter Approach: Use AI Only When It Helps
The SMART research provides an interesting solution.
Instead of sending every user through an expensive LLM pipeline, the system first uses a lightweight quality gate.
The system asks, in effect:
Does the existing retrieval system already have good enough results?
If the answer is yes, there is no need to invoke the expensive AI model.
If the system detects a coverage gap meaning the traditional approach is unlikely to discover relevant products the user can be routed to the semantic AI pipeline.
According to the research, only around 10% of users needed this additional semantic prospecting path.
That produced a reported 90% reduction in LLM costs while retaining most of the semantic prospecting benefits.
This represents an important shift in AI architecture.
The future may not belong to systems that use the biggest model everywhere.
It may belong to systems that know when not to use the biggest model.
Why This Matters for eCommerce Advertising
For eCommerce businesses, smarter product advertising could have a significant impact.
The goal of advertising has always been to reduce the distance between a potential customer and the product they are most likely to buy.
Better AI can potentially reduce that distance further.
Instead of showing customers a generic collection of products, advertising systems can become increasingly capable of predicting what the customer actually wants.
This can improve several stages of the funnel.
Better Product Discovery
Customers do not always know exactly what they are looking for.
AI can help advertising systems interpret broad or ambiguous intent and connect users with relevant products.
This could be particularly valuable for fashion, beauty, home décor, electronics, gifts, and other categories where shoppers frequently discover products rather than search for a specific SKU.
More Relevant Ad Experiences
Relevance is becoming one of the most important factors in digital advertising.
A customer who repeatedly sees irrelevant products is unlikely to engage.
AI-powered retrieval can potentially improve the connection between the customer's interests and the products appearing in their feed.
Better Prospecting
Retargeting is useful, but advertising platforms also need to find new customers and introduce products they may not have considered.
Semantic AI is particularly interesting here because it can identify relationships between concepts that may not be obvious through simple keyword matching.
More Efficient Advertising Spend
Better product selection can potentially improve conversion rates and reduce wasted impressions.
The SMART experiment is notable because it combines both objectives: improving relevance while dramatically reducing the amount of expensive LLM inference required. Its online test reported a 27.6% conversion-rate improvement over the embedding-based baseline.
AI Is Also Changing the Creative Side of Product Advertising
The intelligence of advertising is not limited to deciding which product to show.
AI is also transforming the creation of the advertisement itself.
Generative AI can help brands produce:
- Product images
- Lifestyle scenes
- Ad variations
- Headlines
- Product descriptions
- Short-form videos
- Backgrounds
- Promotional graphics
- Different creative formats for different audiences
This means the future advertising system could potentially optimize two things simultaneously:
Which product should be shown?
and
How should that product be presented?
That distinction is important.
A product may be highly relevant to a customer but still fail to attract attention because the creative is weak.
AI therefore has the potential to create a feedback loop:
User intent → product selection → creative generation → ad delivery → engagement → learning → improved product selection.
This could make advertising systems significantly more adaptive than traditional campaign structures.
The Economics of AI Advertising Are Changing
There is another major implication.
When producing an advertisement becomes cheaper, businesses can experiment more.
Historically, creating a new advertising creative could involve photographers, designers, copywriters, editors, agencies, and production teams.
That created a natural limit on experimentation.
A brand might produce a few creative variations because producing dozens was too expensive.
Generative AI changes the economics.
Research into the strategic impact of cheaper ad production suggests that when creative production costs fall, advertisers can shift away from relying only on broad distribution of generic creatives and toward producing more targeted, higher-quality variations.
This could eventually result in advertising becoming much more personalized.
Instead of one campaign creative being shown to millions of people, brands may increasingly produce hundreds or thousands of variations designed for different audiences, contexts, products, and stages of the customer journey.
Product Images Become Even More Important
As advertising becomes smarter, the quality and flexibility of product assets become increasingly important.
An AI advertising system can decide that a particular customer is likely to respond to a certain product.
But the system still needs a compelling visual representation of that product.
For eCommerce brands, this makes product photography and product image production a strategic part of AI advertising.
A single product may need:
A clean marketplace image.
A lifestyle image.
A social media creative.
A vertical advertising image.
A seasonal promotional image.
A comparison graphic.
A video thumbnail.
The ability to quickly create these variations can become a competitive advantage.
This is one reason tools that automate product image generation, editing, and background removal are becoming increasingly relevant to modern eCommerce workflows. Instead of treating product photography as a one-time production task, brands can treat product imagery as a reusable creative system.
AI-Powered Ads Are Moving From Personalization to Prediction
Traditional personalization asks:
"What has this customer done before?"
AI-powered advertising is increasingly asking:
"What is this customer likely to want next?"
That is a much more powerful question.
The difference between the two is subtle but important.
A customer may never have searched for a particular product.
They may never have clicked on an advertisement for it.
But based on their broader behavior and semantic intent, an AI system may determine that the product is relevant.
This creates an opportunity for advertising platforms to become discovery engines rather than simply response engines.
The Yahoo research team's earlier work on audience prospecting for Dynamic Product Ads provides another example of this direction, using behavioral signals to improve product prospecting and reporting measurable delivery and revenue improvements in online testing.
What This Means for Marketers
Marketers should not think about AI-powered advertising as simply another automation tool.
The larger shift is strategic.
Brands need to start thinking in terms of data, creative flexibility, and machine-readable product information.
Product catalogs need to be accurate.
Product titles need to be descriptive.
Attributes need to be structured.
Images need to clearly communicate the product.
Inventory information needs to stay updated.
Pricing and promotions need to be reliable.
Why?
Because increasingly intelligent advertising systems can only make good decisions when they have good information to work with.
AI does not eliminate the importance of product data.
It increases it.
The Rise of AI-Native Advertising Campaigns
The next generation of advertising campaigns may look very different from today's campaigns.
Instead of manually creating a handful of advertisements and defining a fixed audience, marketers could increasingly define objectives and constraints while AI systems handle much of the execution.
A marketer might specify:
"Find customers interested in premium skincare, prioritize high-intent shoppers, test these five products, adapt creative based on performance, and optimize toward purchases."
The advertising system could then determine:
- Which customers to target
- Which products to show
- Which creative to use
- Which product image to display
- Which message to emphasize
- How frequently to show the ad
- When to stop showing it
- Which new audiences to test
This is the broader direction of agentic advertising.
The human sets the commercial objective.
The AI handles more of the decision-making loop.
But Smarter Ads Also Create New Challenges
AI-powered advertising is not without risks.
More personalization can raise privacy concerns.
More automated decision-making can make advertising systems harder to understand.
Generative AI can also introduce inaccurate product information or misleading creative if it is not properly controlled.
There is another important issue: consumer trust.
Recent research examining AI disclosure in digital advertising found that explicitly revealing AI involvement can sometimes reduce engagement because consumers may perceive the communication as requiring less human effort or associate it with lower product quality.
This does not mean brands should hide AI.
It means that how AI is used matters.
Consumers still want authenticity, quality, transparency, and confidence in what they are buying.
What Comes Next for AI-Powered Product Ads?
The most interesting development is not simply that AI is entering advertising.
AI has already entered advertising.
The important development is that advertising systems are becoming better at deciding when AI is actually necessary.
The SMART research offers a useful example of this principle.
Traditional retrieval handles straightforward cases.
AI handles the difficult semantic cases.
A lightweight system decides which path to take.
The result can be a system that is simultaneously more intelligent and more economical.
That is likely to become an important design principle across the AI industry.
The future will not necessarily be about putting a large language model into every part of the advertising stack.
It will be about building intelligent systems that combine different models, retrieval methods, rules, data sources, and decision layers efficiently.
The Future of Advertising Is Adaptive
Advertising has already evolved from static billboards to programmatic targeting, real-time bidding, personalized recommendations, and dynamic product ads.
AI is pushing the next transition.
The advertisement is becoming adaptive.
The product selection can change.
The audience can change.
The creative can change.
The message can change.
The system can learn from every interaction.
And increasingly, AI can help determine which parts of that process actually require expensive computation.
For eCommerce brands, this creates a clear opportunity.
The winners in AI-powered advertising will not necessarily be the companies generating the most ads.
They will be the companies generating the right product, for the right customer, with the right creative, at the right moment while keeping the economics under control.
That is the real significance of smarter Dynamic Product Ads.
AI is not simply making digital advertising more automated.
It is making advertising more capable of understanding intent, discovering products, adapting creative, and continuously optimizing the customer journey.
And as the cost of AI inference continues to fall while retrieval and generative systems become more sophisticated, the line between advertising platform, recommendation engine, and AI shopping assistant may become increasingly difficult to distinguish.
The future of product advertising is therefore not just dynamic.
It is intelligent, adaptive, and increasingly AI-native.