How Dotdash Meredith enlisted OpenAI to boost its contextual ad product, with Lindsay Van Kirk

This article is part of Digiday’s coverage of its Digiday Publishing Summit. More from the series →

Subscribe: Apple PodcastsSpotify

When Dotdash Meredith made its deal with OpenAI deal last year, it wasn’t only opening its content to ChatGPT. It was also enlisting OpenAI’s large language model to assist its contextual ad product D/Cipher.

“With OpenAI, we now have a tool that allows you to understand language at a much more granular level, and therefore make much more intentional connections between pieces of content when you’re looking for linkages,” Lindsay Van Kirk, svp and group gm of D/Cipher at Dotdash Meredith, said on stage in a live recording of the Digiday Podcast during last week’s Digiday Publishing Summit in Vail, Colorado.

In other words, Dotdash Meredith had already engineered D/Cipher to be able to infer a person’s interests and potential shopping intentions based on the article pages people visit across its 40 properties. But OpenAI’s LLM helps to make connections that may otherwise be more difficult to draw, by taking into account context beyond keywords.

“What the large language models and the databases now allow us to do is actually draw connections, not just from the specific word matches on pages but actually the sentiment and the subject that you’re talking about,” Van Kirk said. That enhancement has led to a 30% improvement on the connections made across pieces of content, she said.

This episode also recaps the top news stories from the past week, including the gloomy ad market outlook, Apple’s App Tracking Transparency troubles and ad verification vendors’ bot blind spots.

Here are a few highlights from the conversation, which have been edited for length and clarity.

Context is key

We have a scale of users who come to our properties in the hundreds of millions. We collect over 10 billion signals that actually allow us to understand consumer engagement and behavior with specific article types and then predict the outcome of what that consumer will do next.

The large language lift

What the LLM allows you to do is say like, OK, you and I could be having a conversation about the same exact topic. We could be using entirely different words to talk about that same thing, and LLMs are really, really good at helping you know that Tim and Lindsay are talking about the same thing, even though we’re using different language.

First-party data is the driver

As the publisher, we have the first-party data around the users, and so what we can see that others cannot in these scenarios is the people who read this also read that.

The need for human oversight of AI

We have a team of people that are working on the connections and on this technology. And then, frankly, when we go to build D/Cipher, we still have a human element that looks at the output from all of these connections and helps us make good decisions about whether something actually both makes sense and matches to the brand that we’re trying to actually build that audience for. And so there still needs to be oversight and human intervention into all of these processes.

https://digiday.com/?p=573768

More in Media

Creators are ditching Substack over ideological shift in 2025

The writers who left Substack in early 2025 represent a second wave after an initial burst of Substack creators left the platform in January 2024.

Google ad tech antitrust trial

Publishers left guessing how Google’s March 2025 core update will reshape search 

Google’s core updates, which happen multiple times a year, change its search algorithms and systems and have the potential to make or break publishers’ traffic.

The semantic slide of programmatic’s power word: ‘curation’

Not all curators are created equal, and not every curation pitch holds up under scrutiny.