Digital advertising was built around signals that were easy to see. A user searched for a product, visited a store page, or entered a predefined audience segment. The platform then ranked eligible ads using factors such as predicted relevance, expected value, and the advertiser’s bid, choosing from a set of finished creatives supplied in advance.
AI is not replacing targeting, ranking, or auctions. It is giving those systems more context. Recommendation models can examine longer behavioral sequences and identify potential demand before it is expressed directly. Conversion models can evaluate more of the customer journey and respond faster as new signals appear. Generative tools can then help adapt the execution once the opportunity has been identified.
The central promise is greater advertising efficiency: reaching a potential customer at the right moment and in the right place, even when that person has not yet directly stated an interest in the product.
From Expressed Intent to Predicted Demand
Search advertising remains the clearest form of commercial intent. When someone asks for running shoes, mortgage rates, or hotels in Rome, the need is visible and the matching task is relatively straightforward.
Many purchase journeys are less explicit. Someone may check the weather in another city and read about visas without searching for flights or accommodation. Individually, those actions reveal little. Together, and in the right context, they may suggest that the person is planning a trip and could be receptive to a relevant travel offer.
Earlier advertising systems had far less capacity to model long and varied sequences of behavior. They therefore often relied on recent actions, relatively small sets of signals, or broad audience categories.
Newer recommendation architectures can examine how interests develop, overlap, and morph over time. The order of events matters, as does the context in which they occurred and the feedback suggesting that a previous recommendation was accepted, ignored, or rejected.
This allows advertising systems to look beyond people who have already declared an intention to buy. They can identify patterns associated with emerging demand, estimate which commercial moments are most likely to produce a valuable result, and prioritize them accordingly.
Advertising Learns from Recommendation Systems
The basic mechanism is already familiar from music services, marketplaces, and social feeds. Listen to a different genre for an evening, spend longer looking at a particular type of product, or share an unfamiliar kind of post with a friend, and the recommendations change.
The system does not know with certainty what the user will want next. It updates the probabilities as new information arrives. A small action can alter the apparent set of interests for that particular moment — and the next action can alter it again.
Advertising platforms face a harder version of the same problem. A music or social platform observes interactions inside a relatively consistent environment. Advertising systems may need to interpret very different types of anonymized signals across content, commerce, search, apps, and other placements.
Concrete feedback is also sparser. People do not directly “like” an advertisement to explain that it was relevant or press a button explaining why they ignored it. The system must infer more from indirect signals and from the sequence in which they appear.
Better prediction therefore expands the range of potential customers an advertiser can reach. Instead of only responding to a clear purchase signal, the platform can recognize situations in which a person is becoming more likely to need a particular product or service.
Longer Memory, Better Forecasting
This shift is visible across the advertising industry.
Meta’s Generative Ads Recommendation Model, or GEM, is designed as a foundation model whose learning can improve other models throughout the company’s advertising stack. Meta says it contributed to a 5% increase in ad conversions on Instagram and a 3% increase on Facebook Feed soon after the initial rollout. The model processes behavioral sequences alongside information about ads, formats, advertiser objectives, and engagement.
Google’s Performance Max applies AI across bidding, audiences, creative, and attribution. The company says non-retail advertisers that adopt it see 27% more conversions or conversion value on average at a similar CPA or ROAS. Advertisers can also provide information about their highest-value customers, helping the system prioritize users more likely to generate long-term value.
At Yandex, Argus expanded the behavioral history available to advertising algorithms from 256 events to as many as 8,000, while Gorgona uses a broader range of conversion signals to forecast results and manage bids. Yandex says neural technologies across its advertising system increased overall efficiency by 38% in 2025. Together, the models illustrate the same broader shift: longer behavioral context, faster feedback, and forecasting that is more closely tied to commercial outcomes.
The details vary by platform, but the direction is consistent. Advertising systems are becoming better at interpreting long sequences, responding to recent changes, and distinguishing between activity that merely looks interesting and activity that is likely to produce business value.
Creation Enters the Advertising Stack
Once a platform recognizes an opportunity, the most appropriate execution may not already exist.
Amazon Ads offers a clear example of generation entering the campaign workflow. Its tools can use product information and advertiser-provided materials to create or adapt images, video, audio, and text within products including the advertising console and Amazon DSP. The company says Sponsored Brands campaigns using AI-generated images recorded an average 10.3% higher return on ad spend during the second quarter of 2025. Brands using its AI creative tools also advertised five times more products and used twice as many images per product.
These are Amazon’s own observed results rather than a guarantee for every advertiser. They nevertheless show how generation can reduce the effort required to produce and test more variations. Instead of manually preparing every possible execution, an advertiser can provide the product, source materials, brand requirements, and approved claims. The platform can then help adapt the advertisement to the format, placement, stage of the customer journey, or even the user’s preferred tone of voice.
Baidu represents another route into AI-native marketing. Its services include agents and digital humans that can produce and deliver commercial content in formats such as videos and livestreams. The company reported RMB 2.3 billion in revenue from AI-native marketing services in the first quarter of 2026, up 36% year over year, indicating growing commercial adoption of the category.
Not every impression will require an entirely new advertisement. Brands still need consistency, legal review, and control over their claims. In many cases, the platform may simply select a finished creative or assemble approved components rather than generate the entire message.
The important change is that creation is moving closer to recommendation and delivery. Prediction identifies the opportunity; generation helps the advertiser respond to it.
Why Better Prediction Can Expand the Market
Greater automation does not remove the advertiser from the process. Businesses still determine what they sell, which customers matter, how much those customers are worth, and what claims the brand is prepared to make. AI can improve execution, but it cannot decide the business strategy on the advertiser’s behalf.
Within those boundaries, better prediction can expand the range of opportunities worth pursuing.
Advertising economics are not limited to the first conversion. A customer who makes repeated purchases, adopts additional products, or stays with a service for years may justify a higher acquisition cost than someone who converts once. Recommendation systems can help platforms identify more of those high-value prospects and continue matching existing customers with relevant offers as their needs change.
This means that greater efficiency does not necessarily take the form of a lower price for the same result. It can also mean more revenue or conversion value from every dollar invested. When a platform can recognize more commercially useful moments and distinguish them more accurately from weak signals, a larger share of potential customers and placements becomes economically viable.
Advertisers are then more likely to increase spending — not because they are paying more for the same outcomes, but because the system has uncovered more outcomes worth buying.
The biggest change is that advertising systems no longer have to wait for users to show clear intent. They can identify promising demand earlier. Creative generation can help shape the response, but the harder and more consequential task happens earlier: understanding enough context to find the right customer, moment, and placement.
The strongest advertising systems will not be the ones that generate the most ads. They will be the ones that find more valuable opportunities and show that acting on them produced incremental revenue or stronger return on ad spend.
