A gloomy ad outlook, Apple’s ATT troubles, bot blind sports and Dotdash Meredith’s Lindsay Van Kirk on D/Cipher’s OpenAI assist

Summary

Digiday Podcast Summary
Episode: A Gloomy Ad Outlook, Apple’s ATT Troubles, Bot Blind Sports and Dotdash Meredith’s Lindsay Van Kirk on D/Cipher’s OpenAI Assist
Release Date: April 1, 2025
Host/Authors: Kameka McCoy and Tim Peterson
Guest: Lindsey Van Kirk, SVP and General Manager of Decipher at Dotdash Meredith

---

1. Introduction

In this episode of The Digiday Podcast, hosts Kameka McCoy and Tim Peterson delve into pressing issues affecting the media and marketing industries. From the bleak advertising forecast to regulatory challenges faced by tech giants like Apple, the conversation is rich with insights. The episode culminates with an in-depth discussion with Lindsey Van Kirk from Dotdash Meredith, focusing on how Decipher leverages OpenAI to enhance contextual ad targeting.

---

2. Economic Outlook and Tariffs Impacting Advertising

Tim Peterson opens the discussion by addressing the ongoing macroeconomic headwinds affecting the advertising sector. He mentions that despite expecting ad revenue growth, companies like Magna from the Interpublic Group have slightly revised their forecasts downward, indicating hesitancy among advertisers amid tariff-induced consumer spending concerns.

"Last week, Magna... ticked it down by like, I think, a few tenths of a percentage... but it's still a cut at this point in the year..."
— Tim Peterson [01:30]

Kameka McCoy adds that new tariffs effective from April 6th are causing marketers to reconsider large upfront commitments. The ripple effects extend to retail media networks, a $140 billion global business, questioning the sustainability of increased advertising spends when consumer purchasing power wanes.

"Ripple effects... how much I am going to be able as a... manufacturer to be able to spend on this without passing on those costs?"
— Kameka McCoy [04:54]

The hosts discuss anecdotes such as rising prices of everyday items like eggs and salmon, and personal stories about leasing new cars early due to anticipated cost hikes, underscoring the tangible impact of tariffs on consumer behavior and, consequently, on advertising budgets.

---

3. Tech Giants: Apple’s ATT Troubles and Regulatory Scrutiny

The conversation shifts to the regulatory challenges faced by tech behemoths, particularly Apple. Tim Peterson highlights that French regulators have fined Apple $162.4 million over its App Tracking Transparency (ATT) framework, though this fine is minor compared to Apple's substantial financial reserves.

"What stands out to me about this is Apple's getting slapped on the wrist, but it doesn't have to change its actual policy."
— Tim Peterson [08:09]

They explore the broader implications of such fines, pondering whether this might set a precedent for other regulators in the EU or the US to challenge Apple's ATT policies. The hosts express uncertainty about whether removing ATT would allow the mobile app advertising industry to recover or if the impact would be irreversible.

Kameka McCoy notes the broader context of regulatory actions, including Google’s recent settlement over overcharging advertisers, suggesting a pattern of tech giants facing fines yet continuing profitable operations.

"Does it set up any key traction?... Meta's still making billions of dollars from advertising."
— Tim Peterson [11:10]

---

4. Ad Verification Challenges and Bot Traffic

Tim Peterson brings attention to ongoing issues in ad verification, citing a Wall Street Journal report that DoubleVerify, Integral Ad Science, and Human Security have missed tens of millions of bot impressions over seven years. This revelation poses significant challenges for advertisers and publishers who rely on these firms to ensure ad placements are legitimate.

"...these companies have allowed ads to be served to fraudulent impressions to bot impressions."
— Tim Peterson [16:26]

Kameka McCoy reflects on previous discussions about ad fraud and the creation of “made-for-advertising” sites, indicating a persistent and uphill battle in combating bot traffic effectively.

"Seems like this is kind of like a... I don't know what the resolve is here."
— Kameka McCoy [16:43]

The hosts express skepticism about the ability of current ad verification solutions to fully address the problem, emphasizing the need for more robust and transparent measures within the industry.

---

5. Guest Segment: Lindsey Van Kirk on Decipher and OpenAI Integration

The episode features an insightful conversation with Lindsey Van Kirk, SVP and General Manager of Decipher at Dotdash Meredith. Decipher is a contextual ad targeting tool that has recently integrated OpenAI’s large language models (LLMs) to enhance its capabilities.

a. Understanding Decipher vs. Traditional Contextual Targeting

Lindsey Van Kirk clarifies that Decipher goes beyond traditional contextual targeting by utilizing extensive first-party data on consumer behavior, thereby enabling more accurate intent-based targeting rather than mere keyword matching.

"We are not just looking at the ads on the page or the words on the page. We are looking for all of the data around the users."
— Lindsey Van Kirk [31:02]

She provides examples demonstrating how Decipher identifies consumer mindsets by analyzing patterns in content consumption across Dotdash Meredith’s 40 brands, thereby allowing advertisers to reach highly engaged and intent-driven audiences.

b. Enhancing Decipher with OpenAI’s LLMs

The integration with OpenAI has significantly improved Decipher’s ability to make nuanced connections between disparate pieces of content. Lindsey explains that while Decipher previously relied on traditional data science methods, the incorporation of LLMs allows for a 30% enhancement in content connection accuracy.

"It's improved that connection set by about 30% by using some of those tools and using those large language model capabilities."
— Lindsey Van Kirk [28:33]

Through advanced language understanding, Decipher can now recognize sentiment and subject matter across different articles, enabling more sophisticated audience segmentation. This is illustrated with an example where articles about parenting and chocolate dessert recipes are linked not by direct keywords but by inferred consumer behavior patterns.

c. Addressing Challenges and Avoiding AI Hallucinations

Lindsey acknowledges the potential pitfalls of relying solely on LLMs, such as hallucinations, and emphasizes the importance of human oversight in the process. Decipher employs a hybrid approach where data scientists and human reviewers validate the connections made by AI to ensure accuracy and relevance.

"We still have a human element that looks at the output... oversight and human intervention into all of these processes..."
— Lindsey Van Kirk [40:39]

d. Future Prospects and Industry Implications

The discussion touches on how other publishers can adopt similar AI-driven tools by leveraging their own first-party data in conjunction with LLM technologies. Lindsey highlights that the synergy between proprietary data and AI capabilities is key to unlocking deeper consumer insights and more effective ad targeting.

"The key here is really to find the ways that your own first party data and the connections that you can use with LLMs can really help bolster each other."
— Lindsey Van Kirk [37:57]

---

6. Conclusion

The episode wraps up with Kameka McCoy and Tim Peterson reiterating the critical insights shared by Lindsey Van Kirk. They emphasize the transformative potential of AI in ad targeting while acknowledging the ongoing challenges in the advertising ecosystem, from economic headwinds to regulatory and verification hurdles.

"As always, we'll see..."
— Kameka McCoy [19:10]

Listeners are encouraged to explore more through Digiday’s daily newsletter and to engage with future podcast episodes for continued industry analysis and expert interviews.

---

Notable Quotes with Timestamps

• Tim Peterson [01:30]: "Last week, Magna... ticked it down by like, I think, a few tenths of a percentage... but it's still a cut at this point in the year..."

• Kameka McCoy [04:54]: "Ripple effects... how much I am going to be able as a... manufacturer to be able to spend on this without passing on those costs?"

• Tim Peterson [08:09]: "What stands out to me about this is Apple's getting slapped on the wrist, but it doesn't have to change its actual policy."

• Lindsey Van Kirk [31:02]: "We are not just looking at the ads on the page or the words on the page. We are looking for all of the data around the users."

• Lindsey Van Kirk [28:33]: "It's improved that connection set by about 30% by using some of those tools and using those large language model capabilities."

• Lindsey Van Kirk [40:39]: "We still have a human element that looks at the output... oversight and human intervention into all of these processes..."

---

Final Thoughts

This episode of The Digiday Podcast offers a comprehensive exploration of the intertwined challenges and innovations shaping the media and marketing landscapes. From economic uncertainties and regulatory pressures to cutting-edge AI integrations in ad technology, listeners gain valuable perspectives from industry experts navigating the digital transition.

For those interested in the future of advertising and media, particularly in how AI technologies like OpenAI’s LLMs are revolutionizing ad targeting strategies, this episode is a must-listen. The detailed discussion with Lindsey Van Kirk provides actionable insights into leveraging first-party data and AI to optimize advertising effectiveness in a complex and evolving market.

---

Note: This summary omits non-content sections such as advertisements, introductions, and outros to focus solely on the substantive discussions of the podcast episode.

Transcript

[Lindsey Van Kirk]

Foreign. Hello.

[Kameka McCoy]

Hello and welcome to another episode of the Digiday Podcast, a show about the business of media and marketing. I'm Kameka McCoy, senior marketing reporter here at Digiday.

[Tim Peterson]

And I'm Tim Peterson, executive editor of video and Audio Digiday Media.

[Kameka McCoy]

Tim, welcome back. How was dps?

[Tim Peterson]

It was cool. It was cool, not super cold. It was like a little on the warmer side compared to past DBS's, like coming from California, I just felt like I never packed enough clothes, enough beanies, enough scarves, enough gloves. This one, it was pretty nice. But I also kind of missed, like, having the snow like DPS in Vale. The DJ Publishing Summit, the spring edition is generally the only time I get a in snow, and so could have gone for a little bit more snow, but the conversations were great, the sessions were good. Sarah W. Leone, our senior media reporter, and Alex Lee, our senior entertainment media reporter, did a great job on stage. So overall I think it was pretty successful show.

[Kameka McCoy]

Love that. And for anybody that's just now catching up, you guys did some live sessions, including the one for this podcast episode later on with your guest, right?

[Tim Peterson]

Yeah. So for this episode, we have a live recording from a session that I did with Lindsey Van Kirk, who is SVP and general manager of Decipher over at Meredith. Decipher being this contextual ad targeting product. So last year, Meredith did a deal with OpenAI like a lot of publishers did. Content licensing was a big part of that deal, but another part of that deal that didn't get a lot of attention at the time was that dot dash Meredith is also using OpenAI's large language model to enhance its contextual ad targeting product, Decipher. And so that's what Lindsey and I spoke about on stage and what folks will be able to hear in this episode.

[Kameka McCoy]

Nice, nice, nice. Which ties in nicely with one of the topics that we'll be talking about. So, as always, we'll get to our conversation with the guests later on in this episode. But first we've got some juicy, juicy scoops, including what's going on with tariffs, what's going on with tech giants, and what's going on with ad verification.

[Tim Peterson]

Yeah, and there's a lot to talk about, especially like the tariffs bit and really just like macroeconomic trends. I thought, I thought macroeconomic headwinds was gonna be a term we were able to leave in the past, back in 2023, at least for a bit. There are always like everything cyclical, including the economy, but. But I didn't realize it was only going to be less than two years later that we'd be talking about macroeconomic headwinds. And those have hit. So last week, Magna, which is part of Interpublic Group, they revised their ad revenue forecast for this year. They still expect ad revenue to grow overall, and they only ticked it down by like, I think, a few tenths of a percentage, as opposed to, like, it being a significant cut. But it's still a cut at this point in the year and at a time when it seems like there is a lot of hesitance on the part of advertisers to commit money, to be spending money, to be worried about how much money they're going to have available because of tariffs, for one. And then also just the impact of tariffs on consumer prices and just people's willingness to buy stuff. We had the conversation a couple weeks ago about the price of eggs and the price of salmon. My sister just leased a new car and turned in her old lease early because she was worried about how expensive cars are going to be if she had waited until December to turn into her car and lease a new one.

[Kameka McCoy]

Yeah, the Trump, Trump's, excuse me, tariffs. And those effects start to go in. I think April 6th is the day that they go into effect, at least for like automakers and whatnot. So we've actually been reporting on that a little bit. But as I'm talking to marketers and media buyers, the big concern is about going into these big commitments, you know, upfronts, in fronts, new fronts, any front that is being required of a marketer, the question now becomes like, you know, for as much money is being pumped into the advertising space specifically for retail media networks, last year, Emarketer said that that was a $140 billion global business. So for as much growth as that's happening there now, the question becomes like, if there are tariffs that are going to impact how much consumers are spending, how much I am going to be able as a, you know, manufacturer to be able to spend on this without passing on those costs? Can I then justify spending millions upon millions year over year increased with the retail media network? So it seems like with all of this there's, you know, I say it every week, ripple effects.

[Tim Peterson]

Yeah, no, I mean, in some respects we're already seeing a bit of those ripple effects or certain pockets of the industry are. So during the Digiday Publishing Summit last week, we do these behind closed door conversations with publishers that we call town halls, which are like large group therapy sessions, really. And one topic of conversation was just how has the first quarter of 2025 been for publishers advertising businesses. In short, not great, not great at all. Especially a lot of publishers were saying it was weaker than usual, it was a soft quarter. QN is never necessarily like the best quarter, although there was one publisher who said QN last year was like a record quarter for them. And this year is absolutely not the programmatic open market with Seb Joseph, our executive editor News and Ronan Shields, our senior ad tech reporter have been reporting on the weakness in the programmatic open market. Publishers were confirming a lot of that and then I'm having a lot of kind of pre upfront conversations with buyers and sellers, TV streaming ad buyers and sellers and they're starting to be a little concerned about how much money is going to be on the table in this upfront. At this point it sounds like folks are still expecting there to be more money spent than last year that like for upfront to grow year over year. But they are revising down their estimates as Magna just did for the overall ad market, as Brian Weeser did a couple weeks ago for the overall ad market. So it's. The clouds are rolling in.

[Kameka McCoy]

Yeah. I am curious though, like for publishers specifically, you know, it's interesting that Q1 was reported as such to you guys when there was so much remember we talked like last year about the potential Trump bump 2.0 that could be in play here and it seems that that is, you know, the way that it was teed up. It didn't shake out the way that the hope was.

[Tim Peterson]

Yeah, at least there wasn't a Trump bump when it comes to the advertising side of things. Or at least no one was necessarily, I guess forecasting there to be tariffs that would really affect consumer spending, especially categories like auto. Yeah, that's a really tough one because auto advertisers spend a lot of money especially in the upfront. But beyond the upfront I think it was the CEO of Lululemon, you know, just within the past week of, you know, I believe was saying they're expecting sales to suffer a bit because. Because of tariffs and just these again, macroeconomic headwinds. So it just seems like pretty much everyone is expecting things to get a little tough when it comes to the economy.

[Kameka McCoy]

Yeah, me included with my eggs and you with your salmon.

[Tim Peterson]

Exactly.

[Kameka McCoy]

The next thing that we've got on the the docket here is what's going a check in on the, on the tech giants. You've got French regulators are finding Apple over att the Apple the app tracking transparency. Excuse me. And then Also xai, which is Grok Elon Musk chatbot functionality acquiring X. So that fine is pretty hefty. You're looking at 162.4 million. But the requirement for Apple to change policy doesn't exist.

[Tim Peterson]

And 1 62.4 million is a lot of money for you and I, not Apple, which has this stockpile of billions of dollars. I believe it's. It may be approaching a trillion. I know that Apple's valuation is more than a trillion dollars, but like, Apple famously is an extremely profitable company. IPhones are extremely expensive. All Apple products are. And Apple hangs on to that money seemingly so that when things like this happen, it's just like, all right, Tim Cook just, you know, tells us via Steamers legal team, write the check, let's move on with our life. Because, yeah, the thing that stands out to me about this is Apple's getting slapped on the wrist, but it doesn't have to change its actual policy. Now, whether this enables or paves the path for other regulators to go after Apple over att, that'll be interesting because, okay, now you got a case of one. A proof point of one. Does the eu, does the, the US go after Apple over att? And then what, if any, effect does that have? Because att, which is, you know, for anyone who doesn't know, it's app tracking transparency. It's. If you have an iPhone, a lot of times you'll open up an app and you'll immediately get the prompt of like, are you cool with this app tracking you?

[Kameka McCoy]

Yeah.

[Tim Peterson]

And what that has done, because people like me say, no, I'm not cool with this app tracking me. I don't know why it needs to track me. That's not what I'm using this app for. It's a calculator app. What the hell? And so then that kneecaps those apps, ad businesses, as well as the downstream ad tech companies that rely on being able to track people across apps in order to sell ads. And so there have been companies that have gone out of business because of Apple's att again, Seb Joseph wrote, has written a lot about this for Digita over the years, so I recommend anyone like go read his coverage. But if Apple were to have to get rid of att, which again, French regulators are not saying Apple has to do, but if more actions come and Apple gets rid of att, how much of an impact will that have, at least in the immediate. Does it allow this mobile app advertising industry to rebuild itself? Or does it. Is that window Closed because of att. No idea. Yeah.

[Kameka McCoy]

And there was a big hoopla, as you said, you know, about it a couple years ago where there was a lot of like the cottage industry that had built itself, ad tech that had built itself around this space was really concerned. And then also advertisers and you know, publishers, like if I don't have those insights that we were banking on, you know, what does that mean for my, my business related. Google also settled its lawsuit with advertisers for overcharging. So a lot of tech companies being fined. But to your point, 162.4 million. 100 million here in US dollars, everybody's just being slapped on the wrist.

[Tim Peterson]

Yeah, yeah. So it's again like not a ton of money for Google to have to part with. But does this set up any key traction? Maybe. Does it is the impact of just reminding advertisers of this controversy of this issue where Google wasn't actually giving the advertisers the discounts it was meant to give them, that it was serving ads in countries or regions that the advertisers didn't actually want those ads to be served. Yeah, maybe, maybe that has some impact. But like Facebook or you know, now Meta has a whole history of ad measurement issues. Had, you know, the thing like either last April or the April before where it was mistakenly like running ads and running up advertisers bills and there was a bit of a freak out on a Sunday. I don't think it was Easter Sunday, but maybe around that time. But Meta's still making billions of dollars from advertising. Doesn't seem to have had that much of an effect. So I don't know how much of an effect this is actually going to have still when you're throwing around nine figures, it's worth noting.

[Kameka McCoy]

Odin, this is fair. Maybe that's just like get their attention at the same time. Xai, which is Elon. This is a mouthful. Elon Musk chat bot grock feature. Right. I think I got that out. Right. Is acquiring X a lot of moves being made. How many companies does Elon Musk have? Is the real question here. Jesus.

[Tim Peterson]

Xai X, SpaceX, SpaceX, Tesla, Doge. Technically, if we want to throw that in there, used to have Solar City, but then he rolled that in, you know, already. So he's. Maybe he's in consolidation mode. Maybe the dosh stuff has him thinking of streamlining. Yeah, this one's one of the interesting things about this is just we now have an updated number on X's Valuation and, and this deal I guess puts it at 33 billion. Whereas when Musk acquired Twitter it was valued at 44 billion. So that's like roughly a quarter of its value that's lost, which I guess probably sounds about right given like all the coverage of X, you know, people talking about using X less since Musk acquired it, advertisers certainly pulling back on X that, you know, Musk is in a very menacingly way spend all that trying to change that. Yeah, exactly. So, but it does seem like this, is it an attempt to like turn XAI slash X into a tech giant? Because X certainly isn't a tech giant. XAI is less than 2 years old. Has GROK unclear how many people are actually using Grok Rock, but then also has this Colossus supercomputer that they've built in Memphis. So there definitely seem to be big ambitions with xai. Also Tesla's been having a tough go of it and so I don't know to what extent like in the future Tesla gets rolled in here, SpaceX gets rolled in here, whether XAI is the Musk holding company, but at least right now it's the holding company for X.

[Kameka McCoy]

Yeah, I think the, I mean there have been complaints with Elon Musk doing more with Doge, like what does that mean for the other companies? So on the one hand, yes, this could be a play to like kind of streamline all of the processes there, but also, no, to set X up as a, as a tech giant because who's to say? Well, not who's to say the bullying has not worked for marketers and advertisers, yes, they're spending more, but you know, if that's led to a better evaluation, that's been proven not to be the case based on these latest figures. So as always, we'll see.

[Tim Peterson]

Yeah, yeah. And so, and it also just maybe more money from advertisers will be freed up if there are more of these stories about the issues with ads running on the open web. So in February there's the analytics report on advertisers. Ads running on sites with child sexual abuse material I believe is like CSAM is the acronym for it. So that led to a big drop in a lot of ad tech companies valuations. It also coincided with like when the trade desk, you know, saw the stock hit. But like Pubmatic Magnite have also had their stock prices decline since then. And then within the past week Adelaide issued another report in this 1 claiming that DoubleVerify Integral Ad Science and Human Security 3 companies that advertisers rely on to manage where their ads can and cannot run online that these companies are often missing bot traffic. According to the Wall Street Journal's coverage of this report, there have been tens of millions of instances over the past seven years in which these companies have allowed ads to be served to fraudulent impressions to bot impressions.

[Kameka McCoy]

This seems to be an uphill battle because we talked about this at a town last year where we've looked at this idea of like made for advertising sites and like that being a problem and there was the colossus news and whatnot. So it seems like this is kind of like a. I don't know what the resolve is here.

[Tim Peterson]

Yeah, there's a bit of a whack a mole piece to it as they're so often with programmatic of. I don't know that like any of these companies would ever proclaim or be expected to eliminate 100% of bot impressions at the same time. Like this report seems to be claiming, like, well, they should be doing a lot better job now. The tens of millions of instances over seven years. I don't, I'd be curious to know like what percentage of traffic that actually amounts to because tens of millions of instances over seven years, that's if instances equals impressions. That's actually not that much because you would imagine if people are streaming an NFL game, that's tens of millions of ad impressions. Each ad that gets served in that stream within one game. If there's tens of millions of people watching. But at the same time, like it's not a good look for DVIs and human security who are already under pressure for. You mentioned the classes example, the Forbes example also from last year.

[Kameka McCoy]

Yeah.

[Tim Peterson]

Of advertisers. Agencies are turning to these companies to make sure their ads are running where they're supposed to be running and not running where they shouldn't be running. There continue to be examples of ads that get through. And then from the publisher side of things, publishers are not big fans of these companies. This was something that came up during DPS last week in the town hall. Again where publishers are just feeling like they're getting the short end of the stick from these companies where they're being told they have to pay DVE and ias in order to be able to like get the insights from DVE and ias so that, you know, a publisher can determine how its inventory is being assessed and to make sure like it's, you know, inventory is getting through, that it's impressions are getting through and being made available to advertisers, but they feel like so often they're. That's not actually following through. So they're paying DV IAS security for things that aren't actually helping the publisher's business. And so the frustration on the sell side has been there for years. It seems like this frustration on the buy side could be mounting, if it hasn't already.

[Kameka McCoy]

And then, as always, we'll see if it comes to it.

[Tim Peterson]

Yeah. Which is always, always the case. So.

[Kameka McCoy]

So remind us again of who you've got as our guest this week because it ties into the ad tech and whatnot.

[Jim]

Yep.

[Tim Peterson]

Yeah. So it's Lindsey Van Kirk, who's the SVP and GM of Decipher Meredith. So Decipher is Meredith's contextual targeted tool. And we get into this at the start of the conversation where I try to summarize it, ask Lindsay to correct me on anything that I get wrong. And we get into a little bit of a discussion on what's the difference between a contextual ad targeting tool and what she describes as an intent targeting tool. But it basically takes into account what content you read across Dash Meredith's websites and then makes connections across all of the different articles pages. You come across the keywords on those pages to then determine, like, okay, this is what you have an interest in. This is maybe a product you're in the market for. And so we talked about how they've enlisted OpenAI to enhance that tool, because OpenAI, their large language model GPT, is really good at making contextual inferences based on words. That's the whole point of a large language model. So it makes a lot of sense to take the technology that powers ChatGPT to be able to make sense of words and sound like an actual person, and apply that same type of technology to a contextual ad targeting tool that is largely based around making matches among keywords on different pages.

[Kameka McCoy]

Great. Well, that sounds like exciting conversation and gives us a break away from the OpenAI memes that have been circling the Internet for the past couple. Couple of days. So with no further ado, I'll let you and Lindsay get to it.

[Tim Peterson]

Cool, thanks. So let's talk about Decipher.

[Jim]

So it's been close to two years since you all launched Decipher, which is your contextual ad targeting product, basically indexing the content that Dot dash Meredith publishes across all its various sites, figuring out, okay, what is it that people are reading and using that as signals to then target ads, as opposed to, to what are these people's Household incomes. What shoes have they looked at on the site? I know everything's a gross oversimplification. Is that more or less on the mark?

[Lindsey Van Kirk]

Yeah, I think it's close, but there's a couple probably edits that I would make to that. So the first thing is that the word contextual is probably not exactly the right way to describe decipher. We talk about it as intent because what we're really looking at is not just what is the content of the page, but because we have the scale of dot dash Meredith, which has our 40 brands, and because 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 for a marketer, not just based off of the words on the page, but based off of all of that information that we know. So I'll give you an example of this. Right? So we know if you are on Investopedia reading about 529 plans, or if you're on travel and leisure reading about traveling with children, or if you're on Parents.com reading about disciplining your children, that you are a parent right then and there. There's no other information that we need about you. You are engaging in content that tells us that you are a parent. The other things that we can see are those people who are reading the disciplining children articles are also reading our chocolate dessert recipes on allrecipes, eight times that of usual consumers. Now you might be wondering, why is that the case? Maybe they are. Maybe those parents are stress eating because disciplining children is hard. Maybe they're lining up their carrots and their stick strategy, who knows? But all of that sort of allows us to understand when are the moments that people are actively thinking about parenting or when you can reach someone who is in a parenting mindset. And then we take that one step further and basically say, okay, from all of those articles and from that content, using the URL really as the key, what can we know about what those readers will do next from a purchase behavior perspective? So the other things that we know are because we have partnerships with other companies and we've been able to match our data to folks like Amazon's and others Things like people who are reading about traveling with young children are in the market for a switch. And that's probably basic for anybody who's here thinking like, yeah, when you're traveling with children, you need to distract them. Otherwise that plane ride is going to be extremely painful for you and probably for them and all of the other passengers on that plane. But what we're really talking about is not just how does the word on the page match to the advertiser and their message, but how do we unpack the consumer's mindset and what they are going to do downstream of that so that when you or a marketer are trying to reach them, when you're finding them in that moment, you're finding them in a high engaged and high intent moment where they will actually take an action next towards what you want to drive towards. And we see that over and over again.

[Jim]

Okay, I appreciate that you want to distinguish that from contextual. It sounds very much like just contextual to me and my historical understanding of it. But maybe as we talk into how Decipher has been evolving, then we can kind of understand. Maybe I'll understand the nuance a bit more because. So last year, I think it was the Q2 earnings call. Joey Levin, CEO of IAC, Meredith's parent company, mentioned how you all did the OpenAI deal. And at the time that deal was announced, I think a lot of us just kind of read it like, okay, another publisher doing a content license deal with OpenAI. But in that earnings call, Joey Levin mentions and like, it's actually helped to boost our advertising business. Or like, I forget the exact language, but basically that there was an advertising component to the deal related to decipher. How does OpenAI or GPT figure into Decipher?

[Lindsey Van Kirk]

Yeah, absolutely. So the deal that dot dash Meredith did with OpenAI had three components to it, like you mentioned. The first was a content licensing agreement because we believe that AI companies should pay publishers to train off of their content. Right. The second piece of that was when the answers that you are getting from something like ChatGPT are using specific sources to provide you with that answer, it now includes attribution to those specific articles so that you as the consumer can go in and see, okay, what is the source data of the answer that the AI just presented to me? And the third part of that, which we can get into in more detail, which is the one we want to talk about, is we're partnered deeply with OpenAI from a tooling perspective to help make Decipher a richer tool. Like I said, dot dash Meredith, all the way back to the dot dash days has been doing what we call intent targeting, which has been finding consumers in specific mindsets. What the scale of Meredith brought to the table when we did that acquisition was it really just opened up the amount of connections that we could make across our content. And we've been doing this using traditional data science for years. 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. So, like I said, DDM has 40 properties. And if you're trying to find articles that are related to one another in order to build out, scale and find out all of the spaces where you should be reaching a consumer in that specific mindset, what the large language models and the databases now allow us to do is actually draw connections not just from, again, the specific word matches on pages, but actually the sentiment and the subject that you're talking about. What the LLM allows you to do is say, okay, 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 Lindsey are talking about the same thing, even though we're using different language. And so the number of connections that we can actually make now between our content and content across the open web, when we're looking at content similarity between Dot dash Meredith articles and other articles, as we're extending Decipher into Decipher plus now, it's just a much richer and more powerful connection set and it's improved that connection set by about 30% by using some of those tools and using those large language model capabilities.

[Jim]

What's the metric there? 30%?

[Lindsey Van Kirk]

30% improvements on the connections that we can see across the pieces of content.

[Jim]

So could you give me a for instance in terms of what that means?

[Lindsey Van Kirk]

I'd have to go to our data science team to get the specifics of it. But let's say we're talking about, let's use maybe like a recipe example for this. And I might be using something that my team is going to come back to me and be like, that was not a good example, so everyone will have to forgive me for that.

[Tim Peterson]

Oh, they're not on stage. What's that?

[Jim]

They're not on stage.

[Lindsey Van Kirk]

But you could have two recipes that are about cooking for a specific moment. Right. You could be talking about a particular holiday, or you might be talking about even different holidays. But the sentiment is similar around cooking for a large group of people. And someone who's in that mindset needs specific things. Right. So you might be. Do you have A holiday in particular that you do a lot of cooking for. Do you not cook bad? A question.

[Jim]

I cook, but I cook just because.

[Tim Peterson]

I have to eat. If it's a holiday, that's an excuse.

[Jim]

To not have to cook.

[Lindsey Van Kirk]

Great. Okay. A recipe from a family member that you like for a holiday.

[Tim Peterson]

Sure.

[Jim]

Well, I can give you a challenging one. Yeah, my partner, she makes. They're called pigeons. It's not actual pigeons, but that's what they're called.

[Lindsey Van Kirk]

That is a food that people do eat, by the way.

[Tim Peterson]

It's Polish food.

[Lindsey Van Kirk]

Okay, all right, so there you go. That's a great one. Challenge accepted. So your partner is making pigeons and you said what kind of food is that?

[Jim]

So it's halepkes. It's basically pork. Or you could use plant based meat stewed, wrapped in cabbage, stewed inside a crock pot with tomato sauce.

[Lindsey Van Kirk]

Okay. So I think delicious. That sounds delicious. It's stuffed cabbage is my. Another name. Okay, yeah. All right, so we've just llmed it right here. Right. Like in the audience here, someone said, yep, that's stuffed cabbage. So you've got a specific set of language that you're using for that recipe. Right. And the ingredients are similar, the name is totally different, but the sentiment around that specific dish is related to like other recipes from other cultures. But they have something that's similar. And the LLM capabilities allow us to say, okay, we don't necessarily have to go look for an exact match against all of the ingredients to say that the article that you would be reading for that specific recipe matches someone else's stuffed cabbage recipe. The connection gets made in a much faster way because again, the LLMs allow you to understand that we're talking about the same thing as opposed to having to do an exact match to each of the keywords.

[Jim]

So it makes it a more contextual.

[Tim Peterson]

Product, you might say.

[Lindsey Van Kirk]

Well, it makes the contextual connections piece of it more robust. But again, coming back to the difference between what we do and what I think pure contextual is, is, again, we are not just looking at the ads on the page or the words on the page. We are looking for all of the data around the users.

[Jim]

But isn't that just based on the words that are on the page?

[Lindsey Van Kirk]

It's not just based on the words.

[Jim]

Maybe that's what I'm missing is what's the data outside of the data on the page that you're bringing in?

[Lindsey Van Kirk]

Right. So 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. So my parents reading about discipline.

[Jim]

So it's words on this page and words on that page.

[Lindsey Van Kirk]

No, in that example, it's about the humans who are reading the articles about disciplining children at scale. Also read these chocolate cake recipes, 8x that of our traditional readers. It's that first party understanding of the consumer patterns of content consumption.

[Jim]

But the raw data is just the words on the page.

[Lindsey Van Kirk]

No, it includes this layer of how do you look at the patterns of consumer behavior to draw out interesting insights. So I'll give you another example of.

[Tim Peterson]

This is the metadata.

[Lindsey Van Kirk]

It's the user data that we use. It's engagement data that we use from the consumers on the plat who are reading all of these articles.

[Jim]

Okay, so maybe I'm just a little lost cause on this one.

[Lindsey Van Kirk]

Well, let me give you another example of it that might help bring this to life. So we ran a campaign for a CPG brand who is selling a device that helps you to remove impurities from water. I can't name brand names here, so I'm going to be a little bit vague about all the specifics there. And one of the components for the audience that we were reaching with them was people like content about your kitchen space. Right? Because that's where that device or that type of product usually exists. We also ran that on teen dating content because what we saw was that people who were reading those articles around kitchens were also reading the teen dating content at scale. Now there's no contextual link between those whatsoever. Right? Like there's no words on the page that would tell you that the linkages around like the content about your kitchen and teen dating should go together, other than the fact that we see that from a publisher first party data perspective. And so when we ran that campaign, the ad engagement on the teen dating section was excellent. And a lot of people were like, why is that? When I was a teen, I wasn't thinking about water filtration. It's like, no, it's not you, it's your parents. Your parents are reading about that because they're getting ready to send you maybe to college. And you need a device that's going to help you make sure that the dorm room water isn't just absolute crap for you to drink. And so there's this connection that we can draw because we can see that pattern that helps that marketer actually find the consumer in all of the mindsets when they would be thinking about that type of device. So it's not just the content on the page, it's how do you understand those linkages between what consumers are reading and then use that and the contextual connections together to really unpack a mindset and build an audience at scale.

[Jim]

So there's two data sources here. There's the words on the page and then the user ID for the people.

[Lindsey Van Kirk]

There's a lot more that goes into it. There's like over 10 billion signals that power decipher in terms of user engagement, ad engagement, everything that we can see. But if you wanted to boil it down to those two things to really get at why it's different from just pure contextual, I think that's as fair.

[Jim]

Okay, yeah. I'm a simple person. The simpler things are, the more likely I am to be able to understand them. Hopefully that one now clicks in my.

[Tim Peterson]

Head at this point.

[Jim]

So the OpenAI piece, what's the process then of bringing GPT into decipher, exposing content to GPT for decipher? Could you break me? If we had a whiteboard up here, I would ask you to kind of diagram what is the workflow.

[Lindsey Van Kirk]

Yeah, absolutely. So again, there is not a real GPT engagement with Decipher. It is tools that allow us to make the connections between our content at scale in a much more robust way. So there is no feed between what you see in ChatGPT and what we do with Decipher. Those are totally separate. For us, it's really about how we use the capabilities that OpenAI provides to power those connections. Well, so it is large language models that We've worked with OpenAI on. It's database technology that we use to actually make those connections and house those connections and do that at scale. But the actual chatgpt side of it doesn't really factor into cycle.

[Jim]

What's the process here? Is it just like, hey, GPT bot, go crazy on Meredith sites and then just spit out some audience segments for us to use in Decipher?

[Lindsey Van Kirk]

No, no, no, not at all. Again, we don't really use GPT for Decipher at all right now.

[Jim]

It is what's there to use from OpenAI for decipher, then?

[Lindsey Van Kirk]

It's the technology that underpins it.

[Jim]

Isn't that GPT?

[Lindsey Van Kirk]

Well, so the main part of this is that we use the large language models, but it's not using the generative AI side of it. So we are using it to contain all of the connections that we have between the content and to make those stronger. But it's not necessarily saying like, okay, no one is going into ChatGPT and saying, Build me an audience of XYZ. We're doing that all in house.

[Jim]

But are you making API calls?

[Lindsey Van Kirk]

The team that's in the weeds of the tech is probably a little bit.

[Jim]

Because what I'm trying to understand is how OpenAI boosting Decipher actually works. And especially if I'm another publisher who has a similar product and I want to do something with, be it OpenAI or anthropic or Google on down the line.

[Lindsey Van Kirk]

Yeah, that's a great way to frame the question. So I think that the key here is that there's a ton of tooling that's out there that allows you as a publisher to understand your content better at scale. And OpenAI is one of those capabilities and that's what we're using it for. With Decipher specifically, there's tons of other applications of it. But from a publisher perspective, if you were thinking about how would I take advantage of AI tooling to help me better understand my content, it really is at using the LLMs to understand the connections between your content. Right. And that's where I think the benefit of taking your own first party data and layering it onto those types of capabilities is where Decipher becomes much more powerful.

[Jim]

So I guess what's the data that the LLM in this case, OpenAI's LLM is accessing and then from there, how is it able to tell you, okay, make these connections when it comes to decipher for this audience segment, or here's a new audience segment you can be creating based on these connections?

[Lindsey Van Kirk]

Yeah, I think I won't go too much into all of the details of all of the secret sauce for how we use all of these capabilities, but the key here is really to find the ways that your own first party data and the connections that you can use with LLMs can really help bolster each other. I think that that's really the way that as a publisher you can take advantage of this tool set is really to take what you know about your consumers and what large language models can tell you about content connections and find ways of bringing those things together. And there's so many different ways that you could do that. And it requires a lot of data science and it requires you to actually go in and make sure that the data that is feeding into it is really, really strong and very powerful and that it has a richness that actually allows you to extract the right types of insights out of it. But you can't rely only on the LLMs alone to tell that story, because there are still so many challenges with how AI tooling answers questions and operates that you really need to bring your own data and your own perspective and frankly, human level intelligence to the answers to really get the right outcome out of all of the AI tool that you want.

[Jim]

And so I guess I'll ask the question a different way. What's the output from the LLM that's leading to that 30% improvement? Like, is the LLM saying here's a new audience segment? Is it saying here's a new.

[Lindsey Van Kirk]

No, it's really about that content connection like I was talking about earlier. It's that challenge of, okay, if we've got two. Again, we were doing this before OpenAI to some extent using data science, but what OpenAI has allowed us to do is again, get to that level of all right, understanding language at a different level at scale. So again, the example that I gave earlier of if we're looking at two separate articles that are talking about the same thing but using different words, we can now connect those in a way that was much more challenging to do before. So those are the types of topics that we can now bring together in ways that were more challenging or difficult to get to. And so it really is about being able to just better understand all of the language and not have to do exact keyword matching in order for it to be an understanding that we're talking about the same topic.

[Jim]

Still, one last question then. I know they're going to kick us off stage. How do you deal with the potential for hallucinations from the LLM?

[Lindsey Van Kirk]

Yeah, that is a great question. So we have a huge team of people that are working on the connections and on this technology. And then frankly, when we go to build Decipher, 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, because outside of that, you're just running LLMs on top of LLM outputted content, and that ends up in a very bad cycle. So we're constantly finding ways of just training the models, having humans actually guide the decisioning, and then leveraging that to continue to improve.

[Jim]

We are, over time. But Lindsey, thanks so much for doing it.

[Lindsey Van Kirk]

Thank you, Jim.

[Tim Peterson]

Thanks for listening to this episode of the Digiday podcast. If you enjoyed it, please leave us a rating and review on Apple Podcasts, Spotify or Podcast wherever you're listening. Get more from Digiday with our daily newsletter sent out each weekday morning. Visit digiday.comnewsletters to sign up.