Latest / Found in AI: AI Search Visibility, SEO, & GEO / AI Agents Are Taking Over. Yahoo Wants to Keep the Links.

Transcript

AI Agents Are Taking Over. Yahoo Wants to Keep the Links. · Found in AI: AI Search Visibility, SEO, & GEO

• 0:41
  SPEAKER_00: Hey, welcome back to Found in AI.
• 0:43
  SPEAKER_00: I am Cassie Clark, an AI Search Visibility Consultant.
• 0:47
  SPEAKER_00: Today is September 24th, and today we're talking about two developments that get to two sides of the same question.
• 0:55
  SPEAKER_00: What does the future of AI powered discovery actually look like?
• 1:00
  SPEAKER_00: First, Meta has introduced Muse.
• 1:03
  SPEAKER_00: It's a personal AI agent designed to do things on your behalf.
• 1:07
  SPEAKER_00: So not just answer questions, but actually take action.
• 1:10
  SPEAKER_00: And then we're going to talk about Yahoo.
• 1:13
  SPEAKER_00: Yes, Yahoo.
• 1:15
  SPEAKER_00: Because while a lot of the industry is focused on building AI experiences that keep us inside the answer, Yahoo's CEO is making a very different argument about the future of search, publisher traffic, and how the open web gets paid.
• 1:32
  SPEAKER_00: And I really want to spend some time on that because I don't think enough people are talking about it just yet.
• 1:38
  SPEAKER_00: Let's get into it.
• 1:47
  SPEAKER_00: So Meta introduced Muse on September 8th, and they describe it as a personal AI agent built to handle task and work towards goals on a user's behalf.
• 1:58
  SPEAKER_00: It appears that Muse isn't being positioned as another chat bot you open when you have a question.
• 2:04
  SPEAKER_00: Meta wants it to be something that you delegate work to.
• 2:08
  SPEAKER_00: According to the announcement, Muse can send emails, book travel, open a browser, fill out forms, negotiate on a user's behalf, and complete purchases.
• 2:20
  SPEAKER_00: It can also keep working after someone closes the app and then return when it needs a decision or approval.
• 2:28
  SPEAKER_00: It's powered by Meta's MuseSpark model and runs on what Meta calls Muse Secure VM, which is a dedicated virtual machine with its own browsers.
• 2:39
  SPEAKER_00: Users can interact with it through the Muse app or through WhatsApp.
• 2:43
  SPEAKER_00: And Meta says users control which services it can access and then it asks for approval before sensitive actions like sending an email or making a purchase occur.
• 2:55
  SPEAKER_00: Now, some of those capabilities sound familiar if you have been following the development of AI agents.
• 3:01
  SPEAKER_00: Sounds a little bit like Google.
• 3:03
  SPEAKER_00: But I think it's the interesting part of Muse is how Meta is packaging them.
• 3:08
  SPEAKER_00: It's trying to make the delegation feel like a normal customer-consumer behavior.
• 3:14
  SPEAKER_00: You don't have to build your AI agent, you don't have to figure out an elaborate workflow, you just tell it what you want done, and it goes off and does the work for you.
• 3:24
  SPEAKER_00: This is really similar to what Google has been building with Gemini.
• 3:29
  SPEAKER_00: And that matters for marketers because the way someone discovers a business could start looking very different depending on adoption of these AI chatbot agents.
• 3:40
  SPEAKER_00: Agents is the word we're looking for.
• 3:42
  SPEAKER_00: So Muse works like this.
• 3:45
  SPEAKER_00: Let's say I want to book a weekend trip.
• 3:47
  SPEAKER_00: Now, in the traditional search journey, I might search for destinations, I might read a couple articles, like a couple different hotels, visit a couple different booking sites, and then eventually make a reservation.
• 4:01
  SPEAKER_00: In the AI search journey, though, I might ask a chatbot, where should I go?
• 4:06
  SPEAKER_00: I might get a few recommendations and then visit one of those businesses.
• 4:11
  SPEAKER_00: But when agents are involved, I could potentially tell muse what kind of trip I want, give it my budget and preferences, and ask it to handle the research and booking so I don't have to do that.
• 4:23
  SPEAKER_00: That changes the role of a brand's website entirely.
• 4:27
  SPEAKER_00: The website is still pretty important, but the person who needs to use it might not be the person who actually makes the purchase.
• 4:35
  SPEAKER_00: It might be the AI agent acting on behalf of the user.
• 4:39
  SPEAKER_00: And that's where I keep coming back to between the differences with AI visibility and agent readiness.
• 4:48
  SPEAKER_00: Being mentioned and recommended in an answer is valuable, and that's one thing.
• 4:52
  SPEAKER_00: But what happens after that recommendation?
• 4:55
  SPEAKER_00: Can an agent find accurate information about your product or service?
• 4:59
  SPEAKER_00: Can it understand your pricing, availability, policies, and next steps?
• 5:05
  SPEAKER_00: Can it complete the action the user asks it to complete?
• 5:09
  SPEAKER_00: If it can't, then being recommended might not be enough to win the business.
• 5:14
  SPEAKER_00: Now I want to be careful here.
• 5:16
  SPEAKER_00: Meta's announcement describes what Muse is designed to do.
• 5:20
  SPEAKER_00: It doesn't really establish that agents can reliably complete every type of transaction across a website today.
• 5:28
  SPEAKER_00: In fact, when I was playing around with it yesterday, it mentioned that captures on transaction sites often cause these work tasks, like purchasing, to fail.
• 5:38
  SPEAKER_00: Now, however, the announcement does show the direction Meta wants the consumer experience to move.
• 5:45
  SPEAKER_00: So there's that.
• 5:47
  SPEAKER_00: There's another piece though that I find really interesting.
• 5:50
  SPEAKER_00: Meta says Muse can remember details about a user and make proactive suggestions based on that context.
• 5:57
  SPEAKER_00: Its announcement gives an example of turning a saved Instagram recipe into a grocery list and then remembering guest dietary restrictions when helping plan a dinner party.
• 6:07
  SPEAKER_00: As someone who is gluten-free, not by choice, that would be really helpful.
• 6:13
  SPEAKER_00: But think about what that can mean for discovery though.
• 6:16
  SPEAKER_00: Today a lot of marketing is built around trying to reach someone at the moment they express intent.
• 6:22
  SPEAKER_00: They type in a query, they browse a category, they start comparing options.
• 6:26
  SPEAKER_00: That's how it's been for a long time.
• 6:29
  SPEAKER_00: But if a personal agent already understands someone's preferences and their routines and their goals, it could begin helping them make decisions before they ever make a traditional search.
• 6:42
  SPEAKER_00: But that doesn't mean that search disappears.
• 6:44
  SPEAKER_00: It means though that some of the moments that we currently think of as search moments become part of a longer ongoing relationship between a person and their agent.
• 6:55
  SPEAKER_00: And from a brand perspective, it does raise a list of new questions.
• 7:00
  SPEAKER_00: How does the agent decide which business to consider?
• 7:04
  SPEAKER_00: What information does it rely on?
• 7:07
  SPEAKER_00: How does it evaluate whether a business can fulfill the request?
• 7:11
  SPEAKER_00: And how much of that journey will a marketer actually be able to measure?
• 7:16
  SPEAKER_00: We do not have definitive answers to those questions yet.
• 7:19
  SPEAKER_00: And these are the same questions that I'm working through with Google's personal intelligence that also affects how Gemini works.
• 7:26
  SPEAKER_00: But I think they're questions worth asking now and thinking about rather than waiting until agent-driven transactions become a much larger part of the consumer behavior.
• 7:37
  SPEAKER_00: And if you go back and listen to the episode I had with AJ Gergic, you know that's coming sooner rather than later.
• 7:44
  SPEAKER_00: Now, before I move on to Yahoo, there is one other thing I don't want to gloss over with Muse, and that is the level of access this kind of product requires.
• 7:55
  SPEAKER_00: For an agent to manage your email or use your accounts or make purchases on your behalf, it needs to access sensitive parts of your digital life.
• 8:06
  SPEAKER_00: Meta has been a little bit proactive about this according to the announcement and describes a number of safeguards, including user-controlled permissions, a separate agent that reviews actions, approval for sensitive tasks, and an audit trail.
• 8:22
  SPEAKER_00: Those are Meta's description of the system, how well the protections work in practice will matter though as people begin using it.
• 8:30
  SPEAKER_00: So we'll see that shake out with time.
• 8:32
  SPEAKER_00: And that trust question isn't really separate from adoption.
• 8:37
  SPEAKER_00: People have to be comfortable delegating before the agent-driven experience Meta is describing can become a routine and everyday practice.
• 8:47
  SPEAKER_00: So my takeaway from Muse isn't that brands need to drop everything and optimize for this one particular product right away.
• 8:54
  SPEAKER_00: We just need to watch this one because the shift from answers to actions is becoming more concrete by the day.
• 9:02
  SPEAKER_00: And brands need to start thinking about what happens after an AI system or an AI engine decides they're relevant.
• 9:08
  SPEAKER_00: Which brings me to Yahoo.
• 9:11
  SPEAKER_00: I want to spend a little more time on this update because I think it deserves more attention than it's getting.
• 9:16
  SPEAKER_00: On September 23rd, which was just yesterday, Yahoo CEO Jim Lensones spoke at Human X in Amsterdam about Yahoo Scout, the company's AI engine.
• 9:28
  SPEAKER_00: If you are a longtime listener to the news updates, you know that I like to record them the day before.
• 9:34
  SPEAKER_00: I ran out of time yesterday.
• 9:36
  SPEAKER_00: That's why I sound a little sleepy because it is early early here in the States.
• 9:41
  SPEAKER_00: I'm glad I ran out of time because this Yahoo update is massive.
• 9:45
  SPEAKER_00: And one of the points Yahoo CEO made was that Yahoo intends to keep Scout full of links back to publishers.
• 9:54
  SPEAKER_00: Links, people, we are getting our links back with Yahoo.
• 9:58
  SPEAKER_00: We're not talking about just citations tucked away somewhere in the interface.
• 10:02
  SPEAKER_00: Those links are supposed to remain visible and they're supposed to remain a part of the search experience.
• 10:09
  SPEAKER_00: According to the next web, Lanzone gave two reasons for that approach.
• 10:13
  SPEAKER_00: First, links let people inspect the sources behind the AI-generated answers, which matters when an answer may be inaccurate or outdated, so they're trying to keep the trust there.
• 10:25
  SPEAKER_00: Second, he argued that sending traffic back to the websites, providing that information, is part of the relationship that has historically made the web work.
• 10:35
  SPEAKER_00: Now, we've spent quite a bit of time talking about what happens when an AI engine answers someone's question without requiring them to click through a website.
• 10:46
  SPEAKER_00: From a user's perspective, that can be convenient.
• 10:50
  SPEAKER_00: You get the answer you need without opening six tabs.
• 10:53
  SPEAKER_00: I was shocked, to my goodness, shocked when my dad said that he loves AI answers because of this very reason.
• 11:02
  SPEAKER_00: But from a publisher's perspective, there's a much harder question that we have to deal with.
• 11:08
  SPEAKER_00: If your content helps produce an answer, but the user never visits your website, how do you sustain, reasonably sustain the work that went into creating that content?
• 11:19
  SPEAKER_00: And this isn't just about traditional news publishers.
• 11:23
  SPEAKER_00: Think about independent creators, industry publications, review sites, niche experts, and businesses investing in original research.
• 11:33
  SPEAKER_00: The economics are very different for each of those groups, but they all have some version of the same problem.
• 11:40
  SPEAKER_00: Producing that kind of content costs money.
• 11:44
  SPEAKER_00: If AI engines become a major way people consume that information, the relationship between the answer engine and the original source matters.
• 11:54
  SPEAKER_00: That's what makes Yahoo's approach to this interesting to me.
• 11:58
  SPEAKER_00: Yahoo isn't saying that AI answers shouldn't exist.
• 12:02
  SPEAKER_00: I mean they're out there building a whole AI answer engine on top of what they already have.
• 12:08
  SPEAKER_00: But Lanzone is arguing that links should remain central to the experience, both so that users can verify what they're seeing and so publishers have an opportunity to receive traffic.
• 12:20
  SPEAKER_00: That is a different product philosophy from treating the AI-generated answer as the entire search destination.
• 12:28
  SPEAKER_00: And I think we should pay attention to whether users really respond to it.
• 12:33
  SPEAKER_00: Now, a little context here because I know Yahoo really isn't necessarily the first company that comes to mind when we're talking about AI search.
• 12:42
  SPEAKER_00: Yahoo Scout launched in beta earlier this year.
• 12:46
  SPEAKER_00: According to the NextWeb's reporting on Lanzone's remarks, Scout uses a combination of Yahoo's own data and search infrastructure, including its knowledge craft and publisher relationships.
• 12:57
  SPEAKER_00: But the announcement also identified Anthropics Claude and Microsoft's being grounding API as parts of the system.
• 13:05
  SPEAKER_00: Lanzone said Yahoo uses a lightweight model to process data while Yahoo handles the ranking and presentation of the answer.
• 13:14
  SPEAKER_00: So Yahoo really isn't just saying, hey, let's put a chat bot on top of the search engine, but is describing an approach where the AI generated response and the links to underlying sources are both part of the product.
• 13:28
  SPEAKER_00: Now, that doesn't automatically mean Scout will send publishers meaningful amounts of traffic.
• 13:34
  SPEAKER_00: A page most likely will contain plenty of links and still satisfies the user's question without the user actually clicking out.
• 13:43
  SPEAKER_00: The important thing to measure will be what people actually do, not just how many links appear in the answer on the screen.
• 13:50
  SPEAKER_00: Do those users go and visit the sources?
• 13:53
  SPEAKER_00: What kinds of queries lead to clicks?
• 13:55
  SPEAKER_00: Are those visits useful to publishers?
• 13:58
  SPEAKER_00: And does the experience help users discover sources that they otherwise wouldn't have found?
• 14:04
  SPEAKER_00: Those are the questions that I would like to have an answer to, and those questions will tell us whether the approach changes anything economically.
• 14:12
  SPEAKER_00: Lanzone also addressed a related issue that I want to bring up.
• 14:17
  SPEAKER_00: He was talking about how publishers should be compensated when AI companies use their content, which, giving the timing with Google's AI contribution pilot program that just was announced last week, this is interesting.
• 14:31
  SPEAKER_00: His position was that micropayments from AI companies won't be enough to sustain publishers at scale.
• 14:38
  SPEAKER_00: Instead, he argued for an advertising-supported search model.
• 14:43
  SPEAKER_00: The idea, as reported by the Next Web, is that sponsored results could be presented with the AI search experience as groups of advertiser links.
• 14:53
  SPEAKER_00: Lanzone acknowledged that click-through behavior could change, but argued that search advertising remains a more viable economic foundation than relying on small content licensing payments alone.
• 15:06
  SPEAKER_00: Now, I don't think that we should treat that as a settled answer to publisher economics.
• 15:12
  SPEAKER_00: Advertising revenue depends on a lot of things, including whether a user clicks, how the ads are priced, and how revenue is shared.
• 15:20
  SPEAKER_00: There's a lot of things up in the air about this.
• 15:23
  SPEAKER_00: And a publisher getting traffic from an AI answer engine isn't necessarily the same as that publisher getting paid for his content being used to generate an answer.
• 15:34
  SPEAKER_00: They are related problems, but they're not identical.
• 15:37
  SPEAKER_00: Still, I think that Lanzone is asking a pretty important question.
• 15:42
  SPEAKER_00: Can AI Search develop a business model that gives the broader web a reason to keep producing the information these systems depend on?
• 15:50
  SPEAKER_00: Yahoo has a particular perspective on that because, as their CEO pointed out, it operates on both sides of the issue.
• 15:59
  SPEAKER_00: It has an AI search product, but it's also a publisher whose content can be used by these AI engines.
• 16:06
  SPEAKER_00: That doesn't mean its proposed solution is automatically the right one.
• 16:10
  SPEAKER_00: I want to be very clear there.
• 16:11
  SPEAKER_00: It does mean though that Yahoo has a direct business interest in what happens to both AI search and publisher traffic.
• 16:19
  SPEAKER_00: So you are probably asking me today, this morning, okay, why should we care about Yahoo Scout if we never use it?
• 16:26
  SPEAKER_00: Um, well, that's a good question.
• 16:29
  SPEAKER_00: Because a lot of the current conversation with AI Search visibility focuses on whether a brand appears in an AI answer.
• 16:37
  SPEAKER_00: We're asking, was the brand recommended, was it mentioned, was it cited?
• 16:41
  SPEAKER_00: Those are all useful things to measure.
• 16:44
  SPEAKER_00: But if you've been in this long enough, or maybe you've just started researching it, maybe the one of the first things that you've noticed is that those questions they don't really tell us the entire story when they get answered.
• 16:55
  SPEAKER_00: A brand could be cited frequently and receive very little referral traffic.
• 17:00
  SPEAKER_00: Another brand might be mentioned less often but receive visits from people who are further along in the decision-making process.
• 17:07
  SPEAKER_00: And for publishers, being cited without receiving traffic may not support the business model that funds the content in the first place.
• 17:16
  SPEAKER_00: And that's why I don't think we can treat every AI search experience as interchangeable.
• 17:22
  SPEAKER_00: The way an agent or an engine, agent or engine, depending on which one we're using, presents information, it changes what visibility can actually deliver.
• 17:33
  SPEAKER_00: If Scout prominently displays publisher links, that is one kind of experience.
• 17:40
  SPEAKER_00: If another product answers a question and encourages the user to continue the conversation inside of that interface, like ChatGPT or Gemini, that is a different experience.
• 17:53
  SPEAKER_00: If we have a personal agent like Muse that goes out and completes a task without the user visiting a website at all, again, very different.
• 18:03
  SPEAKER_00: For marketers, these experiences all require different definitions of success.
• 18:09
  SPEAKER_00: And I think that's where the conversation really needs to start going next.
• 18:14
  SPEAKER_00: Not just, hey, are we visible?
• 18:16
  SPEAKER_00: But what does that visibility lead to in this particular environment?
• 18:21
  SPEAKER_00: So here's why I wanted to cover Muse and Yahoo in the same episode.
• 18:26
  SPEAKER_00: Meta is building a world where you can delegate more of your digital life to an AI agent.
• 18:33
  SPEAKER_00: Yahoo, though, is making a case for an AI search experience that still gives people visible paths back to the websites that supply that information.
• 18:42
  SPEAKER_00: Those aren't necessarily competing visions.
• 18:45
  SPEAKER_00: An agent could use those sources and an answer engine could help users take an action.
• 18:51
  SPEAKER_00: But they put the emphasis in different places.
• 18:54
  SPEAKER_00: Muse makes me think about the future of the customer journey.
• 18:58
  SPEAKER_00: Yahoo makes me think about the future of the information ecosystem that that journey depends on.
• 19:04
  SPEAKER_00: And brands, whether we like it or not, we sit in the middle of both of those.
• 19:09
  SPEAKER_00: Brands need to be discoverable.
• 19:11
  SPEAKER_00: They need to provide information that AI engines can understand and use.
• 19:16
  SPEAKER_00: They need to make it possible for people or agents acting on their behalf to complete that next step.
• 19:24
  SPEAKER_00: But they also need to understand what they're getting back from the environments when they actually are visible.
• 19:30
  SPEAKER_00: And as AI discovery evolves, I think we're going to have to get much more specific about which outcomes we're measuring and why.
• 19:38
  SPEAKER_00: Okay, that is all that I have for you today.
• 19:41
  SPEAKER_00: I'm going to go find some coffee next.
• 19:43
  SPEAKER_00: I'd love to know what you think about Yahoo's approach in particular.
• 19:47
  SPEAKER_00: Would you rather use an AI answer engine that keeps publisher links front and center?
• 19:52
  SPEAKER_00: And do you think an advertising-supported model can preserve the economics of the open web?
• 19:58
  SPEAKER_00: Or are we moving toward a world where more people simply want an agent to get things done, regardless of whether they ever see the websites involved?
• 20:07
  SPEAKER_00: I think both of those developments are worth watching, and I'm really interested to see whether Yahoo Scouts approach leads to meaningful traffic for publishers.
• 20:16
  SPEAKER_00: Speaking of traffic, publishers and revenue, if you have been invited to Google AI contribution pilot and want to talk to you, please email me, I have some questions.
• 20:27
  SPEAKER_00: If you enjoyed this episode though, follow Found in AI wherever you're listening.
• 20:32
  SPEAKER_00: I would love you to pieces if you left a review or shared it with someone who's trying to figure out what AI search means for their business too.
• 20:40
  SPEAKER_00: And if you want more of my analysis on AI visibility, GEO, the changing search landscape, all of it, you can subscribe to the Visibility Report.
• 20:49
  SPEAKER_00: It is my weekly newsletter, it goes out every Friday.
• 20:53
  SPEAKER_00: Okay, that's all I have for you.
• 20:54
  SPEAKER_00: Thank you for listening.
• 20:55
  SPEAKER_00: Until next time, stay visible.