Exclusive: How The New York Times’ Granular Gen AI Tool Drives Campaign Performance

Over the past three months, The New York Times has been using generative artificial intelligence to align a brand’s campaign message with the most relevant articles and interested audiences, executives told ADWEEK. And it’s seeing promising campaign results.

The gen-AI-powered ad tool, called BrandMatch, uses large language models to decode an advertiser’s brief, uncover more specific audience segments and match that brief to the most relevant articles to run ads against. The publisher would not share specifics about how it defines its audience.

“Premium and proprietary ad products like [BrandMatch] have contributed to the success of [our] ad business,” said Joy Robins, global chief advertising officer, New York Times Advertising. “It allows us to help solve for challenges that advertisers previously hadn’t had solutions for. How do we help them reach the perfect audience if you can’t define it with the existing targeting segments?”

The Times ran BrandMatch tests between April and June with six brands, including Paramount+ and Ferragamo, across the tech, finance and luxury sectors. These campaigns outperformed the publisher’s performance averages and benchmarks, netting an average click-through rate of 0.40%, while native display units saw an even higher CTR of 0.72%. The publisher said this was meaningfully higher compared to campaigns without using BrandMatch targeting. Brand lift increased by an average of 8.4%, while consideration and preference experienced a 3.1% lift.

Brands and publishers are increasingly experimenting with gen AI to improve tailored ad-targeting capabilities. Last year, the Times hired Zach Seward, co-founder of Quartz, as editorial director of AI initiatives to test gen AI within its newsroom. The publisher updated its terms of service last August to prevent AI bots from scraping its content, and it is currently locked in a copyright battle with OpenAI.

The Times has over 10 million paid subscribers and more than 100 million registered readers. According to Robins, it reaches 50 million to 100 million people weekly.  

Evolution of preset category targeting tools

The publisher already offers first-party audience targeting and contextual-targeting solutions. However, with these, advertisers select from a limited set of predefined targeting options such as gender, age range or household income, resulting in “very finite, investable assets,” said Robins.

The new gen AI tool is an evolution of traditional category-based targeting methods. For instance, finance investment brands like Robinhood and Fidelity Investments would likely target similar audience segments interested in trading, typically found next to business content. BrandMatch gets more granular with audience interests.

“When you add the brief into BrandMatch, the articles it suggests go beyond traditional business alignments, into culture, entertainment and sports, and find audiences receptive to that message,” said Robins.

More granular targeting

Paramount+ was a launch partner for BrandMatch, using the tool to promote the final season of its show, Evil. The goal was to reach two distinct audiences—fans who had already watched previous seasons of Evil, and people who hadn’t seen the show but were fans of supernatural.

Traditional audience targeting couldn’t achieve this level of nuance, leaving Paramount+ to use similar targeting criteria for both groups, according to Mohit Lohia, senior vice president and head of digital advertising at the Times.

However, using BrandMatch, Paramount+ submitted two separate briefs to describe the differences between the audiences. One brief detailed why Evil fans like the show, and the other described the preferences of supernatural genre fans.

BrandMatch found articles likely to resonate with each audience and modeled two audience segments based on the readers of those articles. Lohia wouldn’t share article specifics.

The campaign achieved a CTR over 20% higher than similar entertainment ads. One of the highest CTRs was within the first-party audience segment defined as “Entertainment Subscribers.” Brand lift showed a positive trend, with more than a 2.2% increase in preference.

More specifically, the existing Evil audience outperformed CTR benchmarks by more than 35%, the publisher said. The Times saw peaks in CTRs driven by audiences defined as in the market for “Movie Theater Tickets” and “Entertainment Subscribers.” One of the top content environments for Evil fans was articles tagged with the emotion “Adventurous,” the publisher said.

A higher adjacency score indicates a match

BrandMatch uses an adjacency score to analyze the advertiser’s brief and identify the most relevant content, said Lohia. A higher adjacency score indicates a closer alignment between the article and the brand’s messaging.

Once it identifies the content most aligned with the brief, the publisher determines the most engaged audience within that content, said Lohia, but the company wouldn’t go into specifics on how it defines its audience.

The first campaigns using BrandMatch targeting should be live in the second half of September. The Times plans to make the tool available for purchase to more clients in the fourth quarter of this year, offering it as part of its suite of ad solutions.