How TIME Is Building for AI Agents, w/ COO Mark Howard

0:00Hello and welcome to AI Curious. My name is Jeff Wilson. I'm a journalist. I'm a human and I'm curious about AI. And here's something that is true right now and that almost nobody is noticed or talking about. There are two internet. There's the one you see, photos, fonts, layouts, all of it built for human eyes. And then there's this second one. Strip down to plain text that only machines ever read. Same articles, different web. Time magazine is now running both. And on the version you will never see, they've started selling ads to the robots, to the AI agents. My guest today is Mark Howard, Times Chief Operating Officer. And this conversation started on a stage. Back in March, Mark and I shared a panel at the Human X conference, the biggest AI conference

1:05out there about how leaders actually scale AI. And he said something that struck with me, something I've been kind of quietly stealing in my workshops ever since. Mark is using AI to do more, not to cut. Now I have wanted a longer, deeper version of that conversation ever since and here it is. We get into all of it. How Mark is driving this kind of surprising fit or marriage of 103 years of tradition of time in all kinds of content, a million pieces of content, stranded across five databases, PDFs, magazines, the 1920s and so on, this old legacy with cutting edge AI tools. He also discusses and kind of shares his playbook, why their first AI experiments was not some low stakes thing someone would never see some internal only process. Their first big AI experiment was Donald Trump as person of the year.

2:10That is not what I would call easing into it. We get into how time now has more AI agents than human employees. Why they blocked the AI crawlers, then let 70 of them back in. We discuss what sacred and what can and should never be touched by a machine and why every company has their own version of that. So without that, please enjoy my conversation with Mark Howard, Chief Operating Officer of Time. Mark, welcome to AI Curious. Thanks so much for having me. So all right, let's in a way pick things right back up where we were on stage together at human acts, which was only like I think in March, but now feels like 75 years ago, given how fast AI moves. But one thing you said on stage has really stuck with me that you and time you're using AI to do more not to cut, to do things that were not possible for at time magazine.

3:18Let's start there. That seems to be a North Star of yours. What are some things that you've done at time that would not have been possible in the age before AI? Yeah, that's great place to start. And I think the most important thing is really thinking about this is 103 year old story and iconic brand. And for over a century, it's meant a lot to a lot of different people all over the world in terms of that red border signaling, a level of trust of trust of journalism, trust of facts, trust of curation of information that people need to know to best understand the world that we live in. What I was back in the early 20s when I launched, or still true today, 103 years later, what's changed, of course, is the access to information, consumer behavior, and all of that layers up to what we saw as the opportunity just a few short years ago.

4:18I joined time three years ago a little over when we were going to have that point was the beginning of a major transformation. We viewed that consumer behavior was changing. We needed to do certain things to reinvigorate this iconic brand and where I come in is really on the technology infrastructure later. What we had at that time was our entire archives split across five different databases, rendering in three different front ends, and had no control over the unified experience. We just just excited to come there. Am I right? 750,000 pieces of content in all in a mix of PDFs and images and text files. You don't have to be one set into the fancy term is unstructured data. That is unstructured as hell across over a sensory just to kind of underscore. That's absolutely right. You think about magazines from the early 20s that were just simply PDFs of those magazines

5:20sitting in a database. We viewed that as one of the big opportunities and really even taken one step back as chat GBT came online at the end of 2022. All of our lives were going to change and we all realized that. But to what extent we were we weren't quite aware but we knew that change was coming and it was coming fast. We looked at it at a couple of different levels. One there's an infrastructure situation. Where was our IP? What makes time time is our content. We also understood especially as it pertains to the website is more and more we were starting to see LLM's coming to the website and looking for that information. They were crawling our pages. They were training on that data. So we needed to pay direct attention to that. We also knew at that time that there was opportunity to partner with the LLM's and so we took a very proactive business development approach on that front. And then lastly, it was how is this going to transform the way that our entire company

6:23operated? What are the tools and how are we going to introduce them into the organization in a way in which it allowed us to do exactly what you said in the intro. We were going to be able to enhance the productivity, give people the ability to do the unimaginable before, be able to take on projects and complete them and take ownership of things that they never would have signed up for before because they didn't have the backup and the resources in the time to be able to do it. And that empowerment of what we were able to do with our team became a critical part of how we operated. So first foremost, we ended into a partnership with a company called SCALAI, amazing company. We took this database of 100 plus, 100 plus years, almost a million pieces of content, totally unstructured. And we built a private LLM just for time, only utilizing our archive and built that in a vectorized database.

7:24We indexed the entire archive and we made that all accessible. So that was one of the major projects. So at the same time, I'm curious to see that this custom LLM you built that's built exclusively on time's content. Is there a starting place that start though with a basic map of the world or basic, like is there kind of like a vanilla LLM, open source model you use that knows kind of basic stuff. And then on top of that, you bolted the custom almost million bits of time content. Is that kind of how it works? Yeah, that's right. So going back a few years ago, when we did the announced Donald Trump as the person of the year in 2024, we introduced the first version of this where we had taken Donald Trump in the previous three people of people of the person of the year. We trained every piece of content that we had generated off of Donald Trump, Taylor Swift,

8:28Automar's Lensky and Elon Musk, and we created a kind of a mini model as the proof of concept for this. And when we released that person of the year with Donald Trump, we released it as a completely new experience where the entirety of the article was written by the author was present and available to be consumed as you normally would. But we had also introduced this toolbar. And the toolbar created all kinds of enhanced functionality. We gave the user the ability to translate into 13 different languages to create audio briefings, short and medium length to be able to summarize the story and give key points, to be able to just get a straight summary and tax format to have a chat on the side where you could interact with the content for any of those four people. All of that content and that chat and all of those features all grounded exclusively on the content of those four people as published by time. Nothing went outside. We did put it on top of chat GPT and OpenAI was our partner in the development of that

9:31toolbar. And that was the one of experimenting with how we could use these new tools to give people the ability to customize the experience in the way to get access to that information in the way in which they best wanted it. What's interesting and important is while we use AI to give people choice and optionality of how they best wanted to engage, the facts remain the facts. We never manipulated the actual factual data of any type piece of time content. And that I think is a critically important component of this story, especially as we'll get a little bit later and to kind of where the web is today and where it's going. Because that's an important part and it's an important part that's differentiator between how we build our model and how the major LLM operates today. And just not just to kind of put a finer point on to it. So by grounding it in time content, am I right? This virtually guarantees it's not hallucinating, right?

10:33You're not giving it free reign to say, what do you think is most likely that happened in January 24? It's saying you can't exactly actual hard facts and you can only choose these facts. You can answer based off of only content that's been published on time. And as you can imagine, we worked very, very closely with scale and the red teams to ensure that the guardrails were tuned appropriately. You can only imagine with those four individuals the kind of questions that people wanted to go to the AI and to try and answer. The reality was it was really designed very. You're saying some controversy. You've done all Trump, possibly some controversy there. You know, even when we live in the Latin mirrors of Lensky, there were comments that people made and the guardrails were intact there to reroute them back to information that it was capable of answering based off of time reporting. And that was the important piece. Also, I love what you're like, okay, our first experiment, let's do a person of the year, Donald Trump.

11:33You did that for the start. The single biggest content moment of the year. That lowest rate, minor non-polarizing figure. Well, we figured when we pulled it off with that, it then opened us up to being able to do the full blown build out of this model. And that then led to the middle of the following year. We introduced audio briefings to all of our top stories of the day. That then led to the fall of 2025 when we released site-wide what we call time AI. And time AI is the universal release of a chat window, but it's really more of an agent because it's got multifunction capabilities where on any article page on the site, you have the ability to go deeper, to add richer content to any topic, to have that information presented back to you, audio briefs, multi-language, summaries, table content, full points, whatever

12:39you desired format for receiving that information. And it's a fascinating tool. You can see it at time.com slash time AI or on any article page, it'll come into play. But we also pre-prompted to show you the depth of what the agent capabilities are. You can say, give me a history over the last three decades of presidential tariff policies and give it to me in a five minute audio brief in Mandarin and Spanish. And the agent will go to work. And the agent will go to work. And in two, three minutes, it'll come back. It'll give you a short summary. And then it'll deliver to you these two audio briefs at the time that you asked a five minute brief. It'll give it to you in each of the respective languages. And so what we were really trying to do, one, was build that muscle and build that technology in-house, figure out how to leverage this hugely valuable archive for the purposes of trying to build products and experiences that are moving in the direction in ways in which consumers

13:43now want and seek information. So that sort of cut in, I care, I think this is a massive undertaking. Did you, and you mentioned you partnered with scale. How much did you use in-house tech talents? Did you hire new AI-focused engineers? Did you bring in small army of consultants? Like what was the, because a lot of listeners with podcasts, you might not be in charge of magazine, but they have their own business and they're thinking they want to do some version of what you're doing. They're like, oh shit, I'm not really sure. Yeah. The actual like, playbook for pulling us off. What kind of, how would you summarize what you did? It's a little bit of a combination of both. We have an incredibly small, but extremely talented tech team here at time. Given that our product is journalism, we do create software and ship it. We're not a tech company by nature. We partner with a company called Code in Theory. They've been amazing product and engineering partners to supplement our internal team.

14:46And that combination of our internal team, our partners of code and our partners at scale, are really the team that built this. The other part of it, which was critically important, which goes back to that guardrails and voice and tone part, is the editorial team was also directly involved in this process from the get go. It was something that we deemed crucially important because how time was represented, the voice of time, especially when it's being told through AI voices and AI chat, is critically important that we got that part right. We sampled all kinds of different voices for the audio briefings to be able to determine how we wanted them to sound, how we wanted them to handle transitions, how we wanted them to handle positive versus neutral versus negative stories. All of those things come into play and the editorial team played a critical role in our editor, Sam Jacobs, was involved in all of these final decisions to ensure that when we finally released this, we all felt confident that the voice of time being told through

15:51this new application was going to be something that we all stood by and supported. Because that is also a place where if you overlook any of those minor details, you could come out with a product that is not reflective of your brand and ultimately do damage to the credibility of our brand. Some minor laps can have massive egg all over your face and suddenly it's a laughing stock not successful. And to your point, this is very public-facing. There are plenty of people out there who would love to try to catch you doing something that is some kind of AI blunder. So the stakes are very high, especially with a brand-like time where all we truly stand for is that red border means trust. And a lot of other things follow that, but that is what we stand for and that's our currency. I want to call, invoke something you mentioned on stage back at HumanX on our panel, which I really like. You said that you guys thought deeply about what is kind of sacred at time.

16:56And it's that red border that trusts in the editorial voice and integrity of time. Okay, that is not something you can compromise. To make this not impossible and everything starts with that point. Exactly. And I use a version of that in my own AI keynote speech. I do workshops, the companies like most people listening to this podcast are not running a magazine, but almost every business has some version of what's sacred to them. Like if you're a hospital, it is the doctor patient relationship, right? And that should stay human to human. You need to have that trust, but there are also things that I would side of the scope of that doctor patient relationship where AI might be useful, right? And so I think almost virtually every company, every project, it's like the challenge is figuring out what is sacred and what are some areas AI should never touch.

17:57And then be really bold and ambitious in the areas that are not sacred where AI can make a big difference and be massive help. So I love that framework. Yeah. And what's even more important about that when you are a media company, similar to that doctor patient relationship is the journalist source relationship. We have no business. People do not trust our reporters and don't want to speak to them. Anybody can get generic updates on the news. What people are looking for today are original reporting that shares something with you that you didn't know before that could only be obtained by a reporter doing the hard work of knowing their sources, getting to the bottom of the stories and their ability to tell those stories. And what's amazing is not only does that make total sense, although it's not necessarily the model that a lot of open web publishing follows today, we've doubled down on that. We've limited the volume of content we published to put it through that rigorous standard

18:58of still being reported, still being fact. And as we start to think about where the future of this agentech web is going, that's one of the reasons because of that original reporting, because we can present original facts that don't necessarily exist in other places as part of our reporting. The AI crawlers come to us at disproportionately high amount of times because of our domain authority and the fact that we're presenting not regurgitated information, but original reporting and original facts. And I think that that is a critical opponent of where the future is going, especially for media companies on the agentech web. We'll get to AI crawlers much more in a moment. Before we do, though, I'm curious, given that kind of those guardrails you're mentioning, Mark, I'm guessing there were some tough calls, though. You guys have had in the last couple of years of like, there are some things where a bundtionly clear, AI should not do this. And there are some things that are a bundtionly clear. Okay, this sure, AI can do this stuff. No one wants to do.

19:59There's probably a lot of trickier, gray area stuff, but we're some of the harder calls you guys had to make a ruling on of maybe anywhere you were like, well, we're tempted to use AI for X. But ultimately, we realized it's too fraught or too problematic for whatever reasons. Yeah. The big hot topic in media, especially journalism, is AI use in newsrooms, of course. And how are you going to push on that? Because of that partnership that I mentioned with Sam Jacobs, our editor and our involvement in doing some of these other projects along the way, one of the major undertakings that we recently had is our team successfully completed a CMS migration. We had been on the same platform for 15 years. We moved to a new headless solution, modern technology, a lot of the basic tools around AB testing headlines and some of the summary tools and creating auto generating briefs and out of the documents.

21:01Some of those things that are kind of table stakes now are just built into the platform. And so that there for the taking, everybody can use them. How are reporters using AI to handle their reporting is on them. We have not mandated, certainly not out of the product and tech or any kind of usage or deployment of AI as it comes down to the actual creation of the content. Now, certainly there are reporters who use it for research and for ideation and for brainstorming on how to evolve some of their, there's all kinds of different ways. Obviously, there's the AI note taking and the transcription and the ability to build catalogs that they can then refer back to and draw, right? There's all these logical deployments that people are using to varying degrees. And that's amazing. Our job was to provide them the tools and to give them the ability to use them accordingly. But we do not in any way generate journalism using AI.

22:02Whether that will become more mainstream in the future and it will creep into the way that the production process works in the future. Still to be determined, I assume it will continue to play a bigger role. But right now, that was one of those places where we have said that is something that each reporter within collaboration with their editor will work through. We provide the tools they determine the workflows. And so that's been a very successful approach to us. And naturally, every reporter has a million ideas. They want to do the best job of within their beat, understanding how to tell the best stories and to be able to deliver that original reporting. We want to provide the containers. We want to provide the distribution. We want to have, as the AI crawlers become a bigger part and AI citations become, we want to be optimized for them so that they're getting that distribution and that visibility and those citations.

23:03That's where my team comes in to help provide that support. Is that still a red line for time that AI, that the actual text of news articles are written by humans, not AI? Is that still a red line that's kind of absolutely? Absolutely. Yeah. Right. So we might have, you might have broken a bit of news at our human next panel when you may, I think, it may have been mentioned before, but you, I think astonished folks when you say that there are more AI agents at time than there are human employees. When I posted that on LinkedIn, that got some reaction. They were like, whoa, say, say more. So, hey, can you fact check me there? Is my memory accurate in B? Can you elaborate on that and unpack that? What are all these AI agents doing at time? Yes. Yes. I'm not sure if you're accurate and even more agents today than that, than back then.

24:04The important thing is you cover this a ton in your podcast, which is how are, how are companies deploying AI and how are they proving ROI on those efforts? How do you get it into the hands of the people until people are embracing it? I heard one of your past guests say you should be using AI not five times a day, but five times an hour. Boy, I was thinking about myself. I wish I was five times an hour, but I'm definitely more than five times a day. What we want is that similar behavior change of all of our employees. Where we started was really kind of multi or single function task agents thinking about your repetitive task in the course of every day, week at month. What are those things that you do that takes up time? You would love for somebody else to do? Well, that's where we started. Our CIO shared mills. She worked across all the department leads and is now worked at the individual level across the company to identify both in proactive engagement.

25:05But now we also have people coming to us proactively saying, hey, do you think you could help us create an agent to do XYZ? And it cuts across every department in the company. When I made that comment back then, what you had were a handful of power users. I myself included that had a half a dozen, eight, ten of these agents doing different things. I have a news curation agent where every day I haven't go out and I look at the intersection of AI and media and it comes back and it both goes to the web and it looks at my inbox at newsletters. And it creates a custom brief for me that I breed every morning when I wake up. I have one of the same things. I got to say, on behalf of George, AI Curious, my agent is AI Curious George. And George, every night George is looking at AI news. And I want to say, Mark, George was very excited about having you on the podcast. So it's amazing. Love George. Love George. And so, right, that's like one basic example.

26:05That's one agent that I have. That's several other people have. Now what we're finding is, while it used to be top heavy several months ago, it's almost universally spread where almost everybody at the company has at least one agent working to do something that helps them with their job. Most common, I love the example you share for yourself, but across the company, what would you guess is the most common use case for most people's agents? If they have one agent, what's it usually doing? There's all kinds of data pools. Everybody needs data, right? It doesn't matter what department you're in and there's specific data sets that are relevant to your role. And so as we think about how we are able to function and everybody empower everybody to no longer need to ask a data analytics team or send a request over, slap or email to somebody, they can now go in and they can literally fire up their agent and they can have it go and pull whatever that data is that they need.

27:07We're also kind of taking a one step further. We think a lot about all of the different partners that we have that bring data to the operations of time and mostly via MCP and the out chat bots that's it on top of it. We're making a lot of that data accessible via chat to broader swaths of the organization and the best part. You can permission it based off of what data they have access to and you can make it really simplified. So this idea of in the past, even for complex data pools, that idea that you start with either the agent or the chat bot that we built for the respective organization departments, you start there, you build from there and then we've given them all the tools we have, a licensing partnership with OpenAI. We've been a partner with them for almost three years. We're a great partner. One of the components of our great partnership is we have access to tech credits that we use internally to build things like that time AI agent.

28:09And now we're also deploying it universally, chat to the enterprise to our entire staff. So they've got that powerful tool at their fingertips and it's one of the things that because of the MCP connections, because of the way that we've been able to build the privacy and safety guardrails into all of the data sets, we now are telling people start by asking chat GPT or we use Glean as an AI search internal search tool. Those are your two starting points. And the vast majority of requests that previously would have required some kind of analytics team or analysts to pull and analyze for you is at your fingertips. And so we're deploying these workflows and this sense of autonomy and the ability to act with agency to our individual team members and the impact of that of course is not only significant time savings, but oftentimes as we all know, people have very specific requests

29:11for very specific either customers or use cases. Their ability to follow that thread through and not have to be dependent on other people is only accelerating the pace with which we can operate that we can get back to our customers, we can get back to our partners, etc. Yeah, quick super quick digression on agents given how everyone wants to do agents all of time this year. I think so many companies struggle with and are some of the surprise to learn that a lot of the maker break of are they working or not is really about the kind of mending issue of context or agents being fed enough kind of and there's a lot of like complicated orchestration involved, even in my own as a one person shop, even for me it's complicated figuring, okay, here are the various text down or mark down files, asus read, newbies and files systems and a lot of it's like, oh, so I have the right file structure set up, which is like very pro-zeic and mundane, but it's up in your experience also that like there's that so much of this is about finding the right context and feeding the right context for

30:12your agents. Certainly, certainly and there's been two major projects that have solved quite a bit of this. The first was that building the agent over the entire archive and structuring all of our editorial content into structured data set that's retrievable and accessible via this time AI agent was the first part because that database was critically important for all this. You think about all the different ways that journalists and editors need access, easy access to any and everything that we published and it could be anything. It could be how did we cover Iran in the 70s and how does that compare to what we're doing today and what's happening between the current regime and the previous regimes and can be some factual stuff. It can be adding contextual factual information to any story narrative. That's one part. The other is we went through a massive data cleanse slash data project moving all of our

31:13data historically cleaning it up, moving it into clean instance, a BigQuery, plugging in all of our partners via API, getting it to a clean instance that allowed us to put AI tools sitting on top of it to be able to make it accessible. Now all of our data feeds into that data warehouse. It's in a clean instance that we just set up literally a year ago to make it more manageable. You think about time at one point was the flagship of the time-inch media empire and then time-inch was sold to Meredith and Meredith declared that time was not going to be part of their go-for, but they were there for a period of time. They operated within that organization. Then it went off independent and went to an independent. We still had technology and data that dated back to 15 plus years prior. We needed to have clean instances that were this version of time that we could build on top of. Both the content library and our clean data warehouse initiatives put us in a position

32:21to be able to be in a much stronger position, to make all of it accessible, to have that ability to apply that contextual relevance and knowledge that we think is so critical. Again, big kudos to my tech team who led that initiative to set this up and to get it all in a place where we felt confident that we could enable this kind of access to the entire data organization. Okay. You mentioned the word crawlers earlier. Let's go there now. It's a big part of what you guys have done this year. It's all kinds of literally breaking news in the last couple of days. Let's go there now. Talk about your time journey with crawlers. You blocklisted some, you whitelisted some. There's obviously crawlers give if crawlers take it the way. There's now two internet in a sense, one internet for humans, one internet for crawlers or AI agents. We could spend five hours alone, I know, just on this topic. But let's go there now.

33:21Talk about your time journey with crawlers. So the journey started almost three years ago. We worked with a couple of different partners to lay infrastructure. We worked with a company called Tollbit and a company called ScalePost, which both sit on top of our CDN. And both of them track every single day all of the different crawlers that come to the site where who they come from, what their purpose is, is it training, is it indexing, is it AI or WebAsix, meaning real time, what pages are they going to, what volume are they coming? So that infrastructure was laid several years ago. We've been tracking and monitoring that. One of the great coalescing forces of a little over a year ago was July 1, 2025, Cloudflare organized amongst a broad swath of the media community, what they call media independence day. That was on July 1, 2025, by default, bot blocking was turned on. It was, everybody was blocked by default.

34:24So we participated in that philosophically. We joined in on July 1, we did the default blocking several months later. For us, our market, just to kind of like just for a moment to kind of set the context for why, right? I mean, I'm guessing that like, I know you have your relationship with OpenAI and licensing, but there are many media companies who I think is fair to say are, there will be delegates frustrated. Some would say outrage. Some would say they think it's a crime that their contents has been pillaged and used for training by AI that is then going to be a direct competition for them. So many media companies are saying, if this, we are going to stop the block, stop the crawlers from coming. And that will then choke off the content being used. And that will force AI companies to negotiate and pay for some kind of partnership. You're too polite to say that. So I want to say that.

35:25Yes, you're right. And there's basically two concepts that are in development, but are taking an extremely long time to materialize. One is the, go in the legal route. And absolutely those companies have been taking the content. Everybody's got the proof. You can see it in all of your logs. They came in. And so there's great, well, you know, many lawsuits and some of them will start to be resolved. Presume in the next year plus. And I think it will set, set some standards around what, what, you know, really is going to be the go forward fair use copyright material. There's, it's very complicated and there's a lot happening there. And there's a large number of lawsuits at this point that are outstanding. The other was, could there be marketplaces? All iTunes and Spotify when there was an appster. And then you could get the content for free, but that was stealing. And people knew it, but you could do it. And then these other options emerge that allowed you to legally source that content.

36:26It gave you quality controls. It gave you ownership. It gave you all the things that a great market should to move everybody's motivation from taking the free. But then moving into this model where we paid, but then it supported the music industry, the artists and it built a home ecosystem. There was a hope and there still is a hope that a marketplace will emerge on a paper crawl or paper use basis for LLMs and agents. We know they're common. We know they're crawling. We know they're indexing and we know oftentimes they're using web assist or AI assist for real-time retrieval and display. Yet, there is no payment system in place for all of that activity. And as Cloudflare declared, over 57% of the internet is now bots versus humans. They have some pretty astronomical growth percentages that they come out with recently in terms of what they think the explosive growth of the agentic web is going to be.

37:27You think about all of them going to websites, hitting your servers, crawling your pages, pulling your content. Not only are the LLMs, but there's a whole ecosystem of crawlers and scrapers out there that are taking this content, repackaging it and monetizing that, the sale of that data, not content to either developer communities and different developer agents that really don't even think about what the process is in terms of sourcing all of that information. They just need the information and they need a package delivered in a tidy way. So there's a lot of different challenges. And then, of course, the whole thing is a bit of an arms-restory game of whack-a-mool, where we can deploy our defenses, those technology companies, whether it's the frontier labs or the scrapers or any of the...

38:27They have technology to circumvent your blocking. Right. So how long do you can play that game? Yeah, absolutely. Just to jump ahead for a bit. Sadly, I know our time is limited. What's so interesting is, and I appreciate you articulating the reasons for blocking. I think your average, someone, anyone who has even a passing-interesting media will say, well, sure, yeah, block those suckers. You did that, but also said, okay, we also need a whitelist, a good chunk of crawlers. I believe 70 was from my understanding. And there's a lot in here, we get in the world of GEO. And there's lots of reasons why we actually want crawlers, as well as we are... There are threats, but there's an opportunity. Talk about... So now we've evolved and we've gone to this allow list of 70 plus. And with that, we look at the volume of AI pages, pages that they are crawling every single day.

39:27And that volume continues to increase, as you would expect. And so one of the things that's well published is because of the way that AI mode works in Google search, Google is sending less traffic out across the open web. Every media publisher is experiencing some form of traffic declines. What is increasing are AI crawls every day. And so we look at the agents coming, we look at the pages they're going to, and one of the things that emerged is several months ago, we started routing those agents and bots to mark down versions of our website. Humans get the full HTML, agents and bots get the mark down page. That mark down page represents a new source of inventory if we were able to come up with the right way to introduce a new ad product. And we have a partner called Moby and they have been a partner of ours for over three years doing all things measurement, intangible advertising.

40:28In partnership with Moby and we worked on what do we think is the appropriate version of introducing an ad unit into that mark down page. And what we came up with is this idea of brand verified and sourced facts. And we know brands ultimately are trying to get in front of AI answer agents is accurate information about their products. When brand looks at their own GEO, they see that their content from their blogs from their website is in those answers. But what they also find are really random third party sources. Oftentimes that are dated information that aren't accurately or positively reflecting what their brand and their products are. And so the way that we've built this product is we work with the brands to establish a very structured data page that gives facts about the company, facts about the history of the

41:31company, the management team, FAQs about certain brand attributes and certain products that they want the facts to be accurately reflected about who they are when it surfaces an AI answer. And what we're able to see and why this GEO ecosystem is becoming so big is brands do have lots of problems with the accuracy of which the ways in which they're represented. And they definitely want to improve their GEO also. And so if you're giving them a way to do that. They definitely need to. And their own and operated can only get them so far. And media can only get them so far. And so one of the opportunities that we found is that our domain authority, the fact that we are cruel, but so many of these AI agents on such a high volume every single day. And the fact that we have so many citations that are occurring every single day and week in these AI answer platforms represent this opportunity. We have this brand, these brand facts only appear on the mark down pages.

42:36They're very clearly labeled, sponsored content, verified facts brought to you by the brand with a date. We lay out this very structured data table of all of the information so it makes it easily digestible by the agents to be able to come and take when they crawl the page. And then we label when it's when it ends and then where the content begins. So it's very, very transparently labeled, which was important to us. We know that these agents have the capability of deciphering between the two of understanding that when we label something is what it is. And we think that there is a big opportunity here for this concept of agent ads on targeting only agents in this explosion of agents across the agentic web as a potentially big opportunity. And it is very early. But we're we're first and we're excited about this. Yeah, let me just something make sure I have as right. I think it's worth underscoring.

43:36This is truly for listeners. This is truly like tip of the spear like frontier bleeding at stuff in the world of media advertising. So Mark, what you in time have done is realize, okay, there's essentially now two internet where we're going to have for every article we have. We're going to have one optimize for good old school old school flesh and blood human beings that care about things like photos and visual nice fonts. And then we're going to have one for AI agents who don't give a shit about what the font will look like. They don't care if the borders are nice and they don't care. They just want and so the mark down was just more or less a fancy word of saying a text file with a slight differences in format and so on. And so you're creating these like mark down slash glorified text files for agents where you can be using whatever we don't get on matter. It's like 8,000 characters or 10,000 really. I mean, a slight token price increase, but there's not the same constraints like you have in a website where you have, you know, only this where you it's a much more constrained real estate.

44:36And you're also then saying we're going to have a whole new inventory when advertised for agents and a real in demands product that people want to pay for is better geo and for listeners who aren't quite full that genera basically SEO for AI, right? So it's like when people go into their a chatbot or chat to be to your clawed and say, I need a good plumber in Cleveland's in the plumber plumbing companies in Cleveland want to make sure they're in it just was in your days. They want to have a high search result for their plumbing shop. They want to have the AI state this. And so you're doing is saying cool on in time articles plumbing shops in Cleveland can pay money to have and I assume you've got this and make sure it's accurate. Oh, founded in 1985, blah, blah, plumber in Cleveland, right? So I'm at my goodness, right? Yeah. You're all right. And I think the important piece is here are a couple things. One, these are brand verified facts.

45:37So they are in fact signing off. These are all totally accurate about their business. The other is all of them have to be sourced. So in the agent ad at the bottom are all the sources of where all the facts come from from their own website. So it is, it is maybe overly cautious, but it is super important to us in this stage of the game that we can prove that when an agent goes out and uses web assist and is instructed to go find facts about the brands that they get it right a much higher percentage of the time, then when they don't have the web assist on and they don't have the ability to go get these facts. And so we're doing all kinds of research right now to look at causal outcomes. We're looking at model impact outcomes. This is a long-term game. There's no silver bullet to GEO, but we do know that it is a huge problem for almost every single brand.

46:37We know how they market themselves to their advertising versus what the AI platforms say about them. And this is where mobile and comes in, they measure AI, sorry, they measure ad messaging and what brands are saying about themselves. They have a massive database for years of how brands have messaged, but then we compare that to what the AI is saying. And inevitably there's a big gap. So we are spending all this time and energy to build these wonderful campaigns for people. And that's messages are coming through. However, when those people are using AI as their first place to do research and to find information about relevant information for themselves, those same, that same information that a brand is trying to convey through their advertising is not coming through in the AI platforms. So we built this end-to-end platform with mobile and to do that analysis, to talk to the brands and for the first time to give them a place where they can actually post these brand

47:37verified facts and address those gaps. We had AI, the financial services company and project management institute. So one B2C company, one B2B company as our launch partners in this and we've got a handful of others that are lined up behind it. But you can only imagine how exciting the conversations are that we're having with so many brands right now about what this means for them. Oh yeah. So last question for you, Mark. And kind of a true part of take a hill over you'd like. Where do you see this headed? Any thoughts on kind of where how the internet will continue to evolve, how media will evolve and then how are you guarding against internal cannibalization, right? So you're doing all this cool AI stuff. How are we making sure that people still go to time.com to read the article itself as well? Yeah. So working backwards. So the website itself is still critically important and it's not only important as a representation of our brand and our brand identity and also a place where you can find the entirety of our catalog and everything we publish.

48:41But the web is by fricating. The human web is now by fricating into you've got wall gardens and social platforms and syndication partners like the Apple news and the Google news is of the world. And so people are finding information in all kinds of different places and it's important that we show up there and we show up there with our best version of time in each one. And our team works very hard with those partners to be able to do that and to build new monetization opportunities. The agentic web is absolutely still this new environment where between again lawsuits and marketplaces and hopefully this concept of agents that are now operating with agency on behalf of individuals on behalf of companies that are finding information. We know for sure that when they come in the crawler page, even if they didn't display that information in that moment and site us, we know that they're retaining that data and when relevant other searches happen, they have the ability to produce that.

49:45And I can prove it because I see on a lot of articles that when they was scraped, it didn't show up in citations at a high volume, but we have more referral traffic to that article than we have citations. That's not actually possible unless the AI is actually storing that information and able to present it in other relevant searches. So it's in a very exciting time. It's very early in terms of the agentic web and those opportunities, but you really have to have like we have and I feel great about this. The foundation of everything that we've been building to get to this moment to be in an opportunity to be able to introduce this new idea. And like I said, the conversations have been phenomenal. It's new. It's is, you know, when search first started, there was all kinds of concerns about would the paid advertising influence the organic results? And clearly that has turned into the greatest business model ever introduced in the world today. And so we're very, very much at the very first step of this.

50:45It will continue to evolve. The product will evolve. The relationship and we have conversations going with several of the LLMs right now. Those conversations will evolve. They're all building their own version of monetization and ads businesses as well. So it's going to be interesting, but we've got great partners who are helping us in mobile and we've got great partners that took that leap of faith in Ally and Project Management Institute. And so we're excited that all of this work over the last couple of years to build those foundations, to build partnerships has led us to this moment where we could do this. And like I said, you know, we'll see where it goes, but the excitement is extremely high. Mark, that's extremely exciting stuff. I just want to echo that what's so cool is you're combining this bleeding edge of AI advertising and internet, agentic internet with a 103-year old iconic brand. And you mentioned you had almost a million bits of content. A few dozen of those million were written by me.

51:46So I appreciate that. Amazing. Yeah. We appreciate that. We appreciate that. And really interesting stuff you guys are doing. We really enjoyed this. Thanks so much. Yeah. Thanks for having me. Really appreciate it. Well, there you have it. Thanks again to my guests, Mark Howard, Chief Operating Officer of Time. Thank you to our show's producer, Jasper Chua, and to Garrett Lang for our show's D-Music. If this is your first time at AI Curious, please consider subscribing. Welcome to a friend, rated five stars, all the good stuff. Thanks again and see you next time.