The New York Times is investing in generative AI for the same reason it’s been investing in first-party data for years.
It wants to be self-sufficient.
Because there’s a lot out there that publishers can’t control: the Chrome cookie saga, Apple’s ATT, Intelligent Tracking Prevention on Safari, GDPR, state privacy laws – and that’s just for starters.
All of these factors “have contributed to a world in which third-party signals aren’t just harder to come by; they’re less accurate,” said Valerio Poce, executive director of ad product marketing for New York Times Advertising.
“Tracking people as they move across websites – that’s going away in favor of alternative solutions,” he said.
One such solution is BrandMatch, a generative AI-powered contextual targeting tool developed by NYT that came out of beta over the summer.
BrandMatch uses a large language model to tease out which customer personas are most relevant to an advertiser’s brief. Then it analyzes NYT’s corpus of content to find thematically linked articles and create custom audience segments that go beyond the obvious.
Say a finance brand has a family-planning product. It makes sense to target people who are interested in personal finance and investing, but what about content related to traveling with kids or buying a house?
“Translating a brief into a set of targeting criteria is a difficult thing for marketers because they’re usually limited by the available options in terms of demographic and psychographic attributes,” Poce said. “But those attributes can’t really capture all of the nuances in a marketing persona.”
Poce went deeper with AdExchanger.
AdExchanger: How does BrandMatch, which is contextual, fit in with using first-party data for targeting, which is deterministic?
VALERIO POCE: BrandMatch is essentially creating a unique audience segment for each individual campaign composed of the people most likely to be interested in a certain message. Layering that approach together with first-party audience segments has a compound effect on performance and expands the reach of a campaign.
Are these audience segments reusable across different brands?
Each segment is unique to that marketing brief, and even a minor update to the brief could result in very different segments. The same would be true for marketing briefs from similar but competing brands.
We tested this on the backend by creating segments for marketing briefs from similar brands in very saturated and competitive categories. We found that even slight distinctions result in significantly different audience segments. Just because two jewelry brands both sell diamond rings, for example, doesn’t mean they have the same brand identity, the same values or the same target audience.
Which LLM does NYT use for BrandMatch, and did you develop it in-house or do you have a partner?
I can’t share that, unfortunately, but what I can say is that we’ve created a flexible system that can integrate some third-party solutions with our own first-party data solutions to achieve the results we get through BrandMatch.
So, let’s talk about results. Can you share numbers?
We’ve found that adding BrandMatch to a campaign that uses Flex XL [NYT’s name for its interactive display units] boosts the click-through rate by 30% on average. Some of the campaigns we’ve analyzed reached a staggering 0.8% click-through rate.
I hate to be facetious, but is an 0.8% CTR really “staggering,” and if it is, what does that say about the state of online engagement?
It’s true that CTR is only one measure. We’re also analyzing BrandMatch based on brand lift and attention, and we’ve seen a positive impact.
But to your point, we’re trying to reshape the advertising economy, by which I mean that not all impressions are created equal. Advertisers typically pay a CPM and get a CTR in return, but there’s a bigger picture here. We want to help marketers reach people who are more likely to respond positively to a message by connecting them to the most valuable impressions we can find.
Another way we’re doing this is by building more cross-portfolio media plans, because people who engage with us in multiple ways drive the highest results. For example, we’ve found that running a campaign that delivers on news and games drives a lift in action intent that’s three times higher than an average campaign on The New York Times.
The New York Times is arguably in a better position to collect first-party data than other smaller, less well-known publishers. What advice do you have for them?
Investing in a sophisticated first-party data system and targeting options is a big effort. It took us a long time and there were a lot of people involved.
But that doesn’t mean an investment in first-party data can’t also be a solution for smaller publishers. Every publisher has an opportunity not just to survive but to be profitable and continue doing the important work they do by knowing their audience and being able to articulate that audience to advertisers.
The main thing is establishing a personal relationship with your readers. Give them something they might even be willing to pay for and, over time, you’ll learn more and more about them. This is knowledge you can use to position your value to the buy side.
Anything you can share about NYT’s lawsuit against OpenAI and Microsoft over copyright infringement for training their LLMs on your content?
I can’t react to that one, I’m sorry.
Okay, then how is NYT’s advertising team preparing for a potential Trump bump, take two?
Can’t comment on that one, either.
Well, can’t blame me for asking!
This interview has been lightly edited and condensed.
