Marketers can use AI to ensure their ads are seen – here’s how

Somewhere between the article you’re reading and the ad next to it, a silent war for your eyeballs is being waged. Most display ads lose their ads because people simply don’t like them. Big tech companies like Perplexity and Anthropic are even looking for better monetization models to escape this burden.

But a new AI tool from researchers at the University of Maryland and Tilburg University aims to change that by predicting with disturbing accuracy whether a user will actually see an ad before someone even bothers to post it.

The tool, called AdGazer, analyzes both the ad itself and the web page content surrounding it to predict how long a typical viewer will spend staring at an ad and its brand logo based on extensive historical data from ad research.

The team trained the system on eye-tracking data from 3,531 digital display ads. Real people wore eye-tracking devices, browsed pages, and their gaze patterns were recorded. AdGazer learned from it all.

When tested with ads that had never been seen before, it predicted attention with a correlation of 0.83. This means that its predictions match actual human gaze patterns about 83% of the time.

Unlike other tools that focus on the ad itself, AdGazer reads the entire page around it. A financial news article next to an ad for a luxury watch will perform differently than an ad for the same watch next to a sports score ticker.

According to a study published in marketing journalads account for at least 33% of how much attention they receive and about 20% of the time viewers spend specifically looking at the brand. This is a big problem for marketers who have long assumed that creatives themselves do all the heavy lifting.

The system uses a multimodal, large-scale language model to extract high-level topics from both the ad and its surrounding page content and determine how well they match semantically (basically, the ad itself and the context in which it is placed). These topic embeddings are fed into the XGBoost model, which combines them with lower-level visual features to generate the final attention score.

The researchers also built Gazer 1.0, an interface that lets you upload your own ads, draw bounding boxes around branding and visual elements, and get predicted gaze duration in seconds, along with a heatmap showing which parts of the image the model thinks will get the most attention. Although this runs without the need for specialized hardware, topic matching that fully leverages LLM requires a GPU environment, which is not yet integrated into the public demo.

For now, it’s an academic tool. But the architecture already exists. The gap between research demos and production ad tech products is measured in months, not years.