Digital advertising is changing quickly.For years, online advertising has mainly worked through keywords, audience targeting, product feeds, browsing history, and algorithms that decide which ad should appear to which customer.
That system is still important. But artificial intelligence is changing how advertising platforms understand customers and choose products.
Instead of simply matching a customer with a product based on keywords or previous clicks, AI can understand the meaning behind a customer's behavior and intent.
This is especially important for eCommerce, where advertising platforms may need to choose products from catalogs containing millions of items.
New research shows that AI can make Dynamic Product Ads more effective while also reducing the high cost of using large AI models. A recent study called SMART (SeMantic-aware Adaptive ReTrieval) found that an AI-assisted retrieval system could reduce large language model inference costs by about 90% while improving ad conversion rates by 27.6% in a two-week online A/B test at Snap compared with a strong embedding-based system.
At the same time, research from BCG X shows that advertising itself is moving into a new environment where AI assistants, AI search, and shopping agents can influence product discovery and purchasing decisions.
So what does this mean for the future of digital advertising?
What Are AI-Powered Product Ads?
AI-powered product ads are advertisements that use artificial intelligence to decide which products are most relevant to a particular customer.
A traditional product advertising system might look at:
- Products a customer previously viewed
- Products they clicked
- Search keywords
- Previous purchases
- Product categories
- Similar products
- Demographic information
AI can add another layer of understanding.
It can look at the broader meaning of a customer's actions and try to understand what the person is actually interested in.
For example, imagine someone searches for:
"Something comfortable to wear while traveling in summer."
A basic advertising system may look for products containing words such as "summer," "travel," or "comfortable."
An AI-powered system can understand that the customer may be looking for lightweight clothing, casual shoes, travel accessories, or other products that fit the situation.
This is called semantic understanding.
The system is not only matching words.
It is trying to understand intent.
Why Dynamic Product Ads Are Important
Dynamic Product Ads, or DPAs, are advertisements that automatically select products from a catalog based on a customer's interests or behavior.
They are widely useful for eCommerce because businesses can have thousands or even millions of products.
It would be impossible for a marketing team to manually decide which product should be shown to every customer.
Advertising systems therefore use automated retrieval and ranking systems.
The challenge is that the system needs to solve two different problems.
The first is retargeting.
This means showing customers products they have already shown interest in.
For example, if someone looked at a pair of running shoes yesterday, the advertising system may show those shoes again.
The second is prospecting.
This means finding products that the customer has not directly shown interest in but may still want.
Prospecting is much harder.
A customer may never have searched for a product before. The system has to discover a connection between the customer's interests and another product.
This is where AI can provide a major advantage.
What the New SMART Research Found
Researchers recently proposed a system called SMART: SeMantic-aware Adaptive ReTrieval.
The research focuses on one of the biggest problems with using large language models in advertising: cost.
Large language models can understand language and intent better than many traditional systems, but running them for millions of users can be extremely expensive.
The researchers therefore did not use an LLM for everyone.
Instead, SMART combines different retrieval methods.
For customers who can already be matched effectively using traditional methods, the system continues using those methods.
When the system detects that traditional search is not finding enough relevant products, it sends that user toward the AI-powered semantic retrieval process.
The researchers describe this as a lightweight quality gate.
In simple terms, the system asks:
"Do we already have good enough products for this customer?"
If yes, there is no need to spend additional AI computing power.
If no, the AI system is used.
According to the research, only around 10% of users needed the additional semantic prospecting path. This allowed the system to capture much of the benefit of AI-based product discovery while reducing LLM costs by about 90%.
This is one of the most important lessons from the research.
The future of AI advertising may not be about using the biggest AI model for every decision.
It may be about using AI only when it provides additional value.
Why Reducing AI Inference Costs Matters
AI inference means the computing work required when an AI model processes a request and produces an output.
For a small application, inference costs may not be a major problem.
Advertising platforms operate at a completely different scale.
Imagine an advertising platform serving millions of users and evaluating thousands or millions of products.
If an expensive AI model has to process every user and every advertising opportunity, the cost can become enormous.
That creates a difficult business problem.
An advertising platform wants AI to make recommendations better, but it cannot allow the cost of running AI to grow faster than the value created by better advertising.
This is why the SMART approach is important.
It treats AI as a specialized tool rather than something that needs to be used everywhere.
Traditional systems can handle simple cases.
AI can handle difficult cases.
Together, they can create a more efficient advertising system.
AI Can Understand Product Intent Better
One of the biggest advantages of AI-powered advertising is its ability to understand relationships between ideas.
Traditional keyword-based advertising depends heavily on matching terms.
If a customer searches for "minimalist office setup," the system may look for products that contain similar keywords.
AI can understand that the customer may be interested in:
- A clean desk
- Ergonomic accessories
- Desk lighting
- Laptop stands
- Minimal furniture
- Office organization products
The customer did not necessarily search for those individual products.
AI can connect the broader intent with potential products.
This can help advertisers discover opportunities that traditional keyword systems might miss.
The SMART research found that rule-generated queries performed well for retargeting, while LLM-generated queries were especially useful for prospecting and discovering new product categories.
That suggests that different technologies can be useful for different stages of the customer journey.
AI Is Changing More Than Product Selection
AI is not only changing which product appears in an advertisement.
It is also changing the environment in which people discover products.
BCG X describes three major parts of what it calls the emerging AI attention stack:
Search-embedded AI, such as Google AI Overviews and other AI-powered search experiences.
Assistant-native AI, including general AI assistants that help people research, plan, compare, and make decisions.
Retail and commerce AI, where shopping assistants help users find products and complete shopping tasks.
This means advertising could gradually move away from traditional banners and search results.
Instead, commercial recommendations may appear directly inside AI-generated answers and conversations.
For example, a customer might ask an AI assistant:
"I need a lightweight laptop for video editing under $1,500."
Instead of receiving ten pages of search results, the customer could receive a short list of suitable products with explanations and comparisons.
If advertising is included in this experience, the product recommendation itself could become part of the advertising environment.
From Search Ads to AI-Native Advertising
Traditional search advertising works around keywords.
A customer searches for something.
Advertisers bid on related terms.
Ads appear alongside search results.
AI changes this process because the system can understand a much broader question.
Instead of matching an exact keyword, an AI system can understand the user's goal.
This creates a new type of advertising opportunity.
BCG X identifies several possible forms of AI advertising, including ads placed inside AI-generated answers, advertising that appears during conversations, and sponsored options presented when an AI agent is helping complete a task.
This is a major change.
The customer may no longer need to actively search for a product.
The AI system could identify the need first and then present relevant options.
AI Shopping Agents Could Make Product Discovery More Important
AI shopping agents could push this change even further.
Today, a customer usually visits a website, searches for a product, compares options, and completes the purchase.
In the future, an AI agent may perform many of these steps.
A customer could say:
"Find me a good office chair under $300 that is comfortable for long working hours."
The AI agent could potentially:
- Understand the requirements.
- Search available products.
- Compare features.
- Check prices.
- Review customer feedback.
- Recommend several options.
- Help complete the purchase.
This creates a new challenge for brands.
Being visible on Google or Amazon may no longer be enough.
Brands will also need to make their products understandable to AI systems.
Product Data Will Become More Important
As AI becomes involved in product discovery, product data becomes increasingly important.
AI systems need information to make decisions.
That includes:
- Product names
- Product descriptions
- Specifications
- Sizes
- Materials
- Colors
- Prices
- Availability
- Reviews
- Images
- Product categories
- Shipping information
Poor product data can make it harder for AI systems to understand what a product actually is.
For example, a product with a vague title and incomplete specifications may be harder for an AI shopping system to recommend than a product with clear and detailed information.
This means SEO is also changing.
Brands should not only optimize product pages for search engines.
They increasingly need to make product information easy for AI systems to understand and interpret.
Product Images Are Becoming Part of the AI Advertising System
Visual content is another important part of this shift.
A product can have perfect data and still fail to attract customers if its images are poor.
Product images are often the first thing shoppers see.
This is especially important when products appear in:
- Social media ads
- Shopping feeds
- Marketplace listings
- Search results
- AI recommendations
- Display advertising
- Mobile shopping apps
As advertising becomes more personalized, brands may also need more image variations.
One product could require a clean marketplace image, a lifestyle image, a seasonal creative, a social media version, and several advertising variations.
AI-powered image tools are making this easier by allowing brands to create and adapt product visuals without organizing a completely new photoshoot for every campaign.
This could become increasingly important as AI advertising creates more opportunities for testing different creative approaches.
Better Targeting Does Not Mean More Personalization Everywhere
There is also an important issue that advertisers cannot ignore: trust.
AI systems can understand more information about customers, but that does not mean businesses should use every available piece of data.
BCG X points to growing consumer concerns around privacy, data use, disclosure, and the possibility of advertising influencing AI-generated recommendations.
This creates a simple rule for brands:
More data does not automatically mean better advertising.
Customers need to understand when commercial content is being shown to them.
AI-generated recommendations should also remain useful and relevant.
If an AI assistant recommends a product only because it is sponsored, while hiding better alternatives, users may quickly lose trust.
The long-term success of AI advertising will therefore depend on a balance between relevance, transparency, and commercial goals.
What Does This Mean for eCommerce Brands?
The rise of AI-powered product advertising creates several practical lessons for eCommerce businesses.
First, brands need better product data.
Second, they need high-quality product images.
Third, they need to think beyond traditional keyword targeting.
Fourth, they need to prepare for customers discovering products through AI assistants and shopping agents.
And finally, they need to experiment.
The advertising landscape is changing too quickly for brands to wait until every platform has a final advertising model.
Companies should start testing AI-driven discovery, product feeds, conversational commerce, AI-generated creative, and new advertising formats as they become available.
The Future of Dynamic Product Ads
The future of Dynamic Product Ads is not simply about making existing advertising systems faster.
It is about making them more intelligent.
A traditional system might ask:
"Which products are similar to what this customer clicked?"
An AI-powered system can ask:
"What is this customer trying to accomplish, and which products could help?"
That is a much broader question.
The difference could improve product discovery, especially for customers who do not know exactly what they want.
At the same time, the SMART research shows that this intelligence does not have to come at an enormous computing cost. By using a lightweight system to decide when AI is actually needed, the researchers achieved a reported 90% reduction in LLM costs and a 27.6% improvement in ad conversion rate in their Snap test.
The Future of Advertising Is Moving From Keywords to Intent
The biggest change may be happening at the level of customer intent.
For decades, digital advertising has focused heavily on clicks, keywords, audiences, impressions, and conversion data.
Those signals will continue to matter.
But AI adds something new: context.
It can help advertising systems understand what customers mean, what they are trying to achieve, and which products could solve their problem.
At the same time, AI assistants and shopping agents are changing where those decisions happen.
Instead of customers moving through search pages, websites, product grids, and comparison pages, more of the journey could happen inside AI-powered interfaces.
This means brands will increasingly compete not only for a customer's attention, but also for AI recommendation and selection.
Conclusion: AI Is Making Product Advertising Smarter
AI-powered product advertising is still developing, but the direction is becoming clear.
Advertising systems are moving from simple product matching toward deeper understanding of customer intent.
The latest SMART research shows how this can be done efficiently. Rather than using expensive AI models for every customer, the system uses traditional retrieval for straightforward cases and AI-powered semantic retrieval when traditional methods are not enough. The result is better product discovery with much lower AI inference costs.
At the same time, the wider advertising industry is moving toward AI search, AI assistants, conversational advertising, and shopping agents. BCG X argues that these systems could fundamentally change how consumers discover, compare, and choose products.
For eCommerce brands, the message is simple.
AI is changing what gets shown, who sees it, and where the buying decision happens.
Brands that prepare their product data, improve their product images, test AI-driven advertising, and focus on genuine customer intent will be better prepared for this next stage of digital commerce.
The future of advertising may not be about showing more ads.
It may be about showing the right product, to the right person, at exactly the right moment, and doing it intelligently enough to understand when AI is actually needed.