ChatGPT Ads, Meta x Manus, CharacterAI, State of Search | #MonthlyLandwehr

00:00:00: Welcome to the Masters of Search podcast.

00:00:02: Today we're kicking off a new format and we're calling it monthly Landwehr.

00:00:08: That's why my guest today is Malte.

00:00:10: Landwehr, CPO and CMO at Peak AI, one of the leading AI search analytics platforms from Berlin in Germany, where also Radiant is located.

00:00:21: Malte is one of the brightest minds in the field.

00:00:23: So I want to take all the news, new research and interesting developments and get his perspective on them.

00:00:30: So good to have you on Malte.

00:00:33: Happy to be here.

00:00:34: Nice.

00:00:35: So let's start.

00:00:36: We have a pretty full document actually.

00:00:40: And I want to start with asking you what did you find most surprising in the last weeks in our field?

00:00:48: I think that chat GPT finally confirmed the ads lounge after they like there was a leak like that it would happen.

00:00:56: Then they denied it that it would happen and then they happened like in a very short amount of time.

00:01:01: I was very sure that ads would happen this year, but I didn't think that it would just be like the beginning of the year.

00:01:10: And yeah, I was.

00:01:11: I was quite surprised how quickly that topic was there.

00:01:17: Were you surprised by the format?

00:01:19: So how the ads will look like in the end?

00:01:23: Yeah, I was surprised that it's charged by impression.

00:01:27: I was surprised that it's a one million minimum commitment.

00:01:33: And I thought they would start with just shopping ads.

00:01:37: Do you have the shopping slider, right?

00:01:40: The shopping carousel.

00:01:42: And I thought, what is the advertisement that has the least impact on this experience?

00:01:47: It's of course selling the, not which product is placed, but which shop does the fulfillment.

00:01:54: Because if I just do the checkout directly in chat GPT, or if an agent does it for me, I don't care if it's Amazon or Walmart or Target sending the package.

00:02:04: So I thought that's where they would put the first ad.

00:02:08: And it wouldn't even feel like an ad as a user, but seems that they are going a very different route with very minimal tracking and measurement in place.

00:02:19: I still believe long-term advertising LLMs has huge potential because you can target people at incredibly detailed and sophisticated ways at points where they are willing to make a decision.

00:02:32: And then the price per impression is also not that crazy.

00:02:37: But yeah, I thought it would look different.

00:02:40: And do you think that advertisers will accept that they basically, at least from what we know so far, basically don't get any information?

00:02:48: Like we know, for example, from Google Ads and the likes where we can get like search term data.

00:02:53: So obviously a lot of that is also now not available anymore due to whatever reason Google has.

00:02:59: But do you think advertisers will accept it long term?

00:03:04: I think many advertisers will not accept it.

00:03:06: They will want to see some form of reporting and details.

00:03:10: But I also believe there will be some companies who will get a huge discount, maybe are promised some other form of partnership.

00:03:17: And then the tech GPD can proudly say, yeah, we have, I don't know, Target or Zappos or whoever now advertising with us or Nike or Adidas.

00:03:30: But I don't think that advertisers will be like, I want this, I need this.

00:03:35: The package doesn't seem good enough for that yet.

00:03:38: But maybe the first couple of cases will be super successful and then everybody wants it.

00:03:43: And once you reach really large levels of advertising, you have to mistrust any reporting you get from a vendor anyways.

00:03:51: You have to do a lift study, a blackout test, something like that, to really understand how big the impact is.

00:03:59: Now I can imagine that a lot of people saw LLM optimization, GEO, AEOS basically being an organic play because it has been an organic play so far.

00:04:10: So what do you think or what's your expectation?

00:04:13: How will it impact organic visibility?

00:04:17: I mean, it's probably not good for organic visibility.

00:04:19: But I expect that overall the usage of LLM-based answer engines will increase so much that it's totally fine that some of that, like the cake is getting bigger and some of that pie will go to the paid side.

00:04:33: And in difference to traditional web search engines, there's a segment of users that is paying money.

00:04:38: and sometimes significant amounts of money every month, and especially the most valuable users are paying money.

00:04:45: I don't think any serious person who doesn't pay for at least one LLM, and I don't think that they want advertisements in there.

00:04:54: So I think it will be a different play than Google.

00:04:57: We will not see as a locked-in user for advertisements before we can see the answer to our prompt.

00:05:04: That will not happen.

00:05:05: Maybe there will be like a free... A version of something where that happens, but I don't think that is the vision for for the GPT or for Gemini or Claude or Grog or any other proper player in the field.

00:05:20: That's probably a good addition because so far from what we know, the rollout will start in the free product and in this low-cost subscription tier, chat GPT go, I think we don't even have this in Europe, right?

00:05:35: I only knew it from India and maybe it was also in other Asian markets and then recently it was launched in the US.

00:05:43: Yeah, I guess you're similar to the cheapest Netflix tier, for example.

00:05:48: It's just that OpenAI is speedrunning this and they are adding the ads very quickly.

00:05:53: They are adding the paid but still with ads tier very quickly.

00:05:57: like the whole evolution is much faster than previously.

00:06:01: And do you think it's connected to like the overall public pressure that has been put onto OpenAI with the big hardware investment commitments and then like the let's say criticism or skepticism about Open AI's revenue trajectory and the whole planning.

00:06:21: So do you think it affected it in any way or is it just they're just calmly executing on the roadmap they had anyway?

00:06:31: I mean, if we look at it completely rationally, if you believe that you can reach AGI, so really general artificial intelligence, right?

00:06:43: Then you can get so much value out of that that making a few billion was advertisements here or there.

00:06:50: It shouldn't matter.

00:06:52: So either it's to appease the markets Maybe it's to prepare for an IPO that they can say oh look our advertisement business grew ten X in the first three years and if this continues for another ten years, we will be profitable.

00:07:10: Otherwise, I don't know.

00:07:10: Or they've given up on their AGI vision or have realized that it's far in the future.

00:07:19: But yeah, I don't think that in the first couple of years, it will make a significant dent.

00:07:23: Like they will still spend so much money on data centers and compute and electricity.

00:07:29: Is it just me or do you also feel like the criticism has been mainly directed towards OpenAI and not so much, for example, against Anthropic?

00:07:38: Whereas from my perspective, actually, Anthropic also, compared to Google, hasn't figured out the whole monetization side yet.

00:07:47: Maybe they're a little bit better in terms of business and enterprise customers, but was it just my impression that OpenAI basically got all the punches, so to say?

00:07:59: I mean, I think OpenAI is the poster child, like they're the most famous ones, so they get all the love, but also all the hate.

00:08:06: And I think Atropic actually just released numbers.

00:08:09: that they are now more profitable than OpenAI, right?

00:08:12: Like, even though they are at a much, much, I mean, more profitable is wrong.

00:08:16: I think they made more revenue this quarter than OpenAI or they caught up to OpenAI.

00:08:23: But they made this very specific, very... conscious decision, I believe, to give up primarily on the consumer side and focus on a B to B product.

00:08:33: And I think they have the much more obvious path to profitability than open AI.

00:08:41: But yeah, they also spend a ton of money, of course.

00:08:46: I was just thinking about Anthropic because you mentioned IPO and there was reporting on Anthropic having hired a law firm in the US that has already brought a couple of companies public and that they might be preparing for that and they might actually be quicker than OpenAI.

00:09:08: Do you have any thoughts on that?

00:09:11: Is it in any way meaningful to you?

00:09:15: It's not really something I think about regularly.

00:09:17: I could imagine that entropic just because their goal seems to be more tangible and more reachable in a limited amount of time.

00:09:27: I think they are the more obvious IPO candidate because public markets usually hate it when you are unprofitable for a long amount of time.

00:09:37: Like Amazon is the one who famously did it for many years, but they were also sometimes punished for it.

00:09:42: And I could imagine that Anthropic as an easier time raising money on the public markets was Open AI.

00:09:51: Unless Open AI had gone public like half a year ago, then there was still enough hype.

00:09:56: But I think the hype is dying down.

00:09:58: People know the Gemini is gaining traction.

00:10:01: Early adopters are starting to use other things.

00:10:04: So I think the hype IPO window where they could have maybe done a spuck or something is over.

00:10:10: And so I think they will stay private longer than OpenAI and AnthroTik actually.

00:10:17: Got it.

00:10:18: Speaking of Gemini and Google, our big friend G, OpenAI actually announced their... So the document is like our approach to advertising and expanding access to Chatshipiti.

00:10:32: And this was on January sixteenth and actually on January eleventh.

00:10:37: Our Google friends also announced something.

00:10:41: They announced their own protocol, the universal commerce protocol compared to the agentic commerce protocol that I had announced earlier.

00:10:51: And they basically together with the protocol announced a new checkout feature on Google product listings in AI mode and also something called business agent.

00:11:05: What are your thoughts on these efforts from Google in also bringing ads and bringing further monetization to AI mode?

00:11:16: I mean, I think the introduction of the protocol and what they bring to AI mode and when this, I would see it a little bit separately.

00:11:22: Because the protocol is really about, it's part of the engine that might power commerce in the future, not just on Google, but it could also influence how create and barrel is integrated into deep seek, right?

00:11:40: Like it's an open protocol, everybody can use it.

00:11:44: And I think Google wants to shape things in a way that is good for Google.

00:11:49: And they have been very good at this, if you look at strategy in the past, right?

00:11:53: They positioned Android to make sure that there's not just one mobile operating system.

00:12:03: They pushed Chrome browser in exactly the right moment to make sure we don't live in a world that is dominated maybe by Internet Explorer and at Safari.

00:12:13: And also they get a lot of data with these things.

00:12:16: And I think in general, Google wants to do things that are good for Google, not things that are Google that are good for the internet and things like.

00:12:26: when they pushed HTTPS, when they pushed core web vitals, of course that was good for the internet, right?

00:12:31: Everything became more secure and faster, but it also led to people spending more money on the internet and that is something where Google participates.

00:12:40: And I believe they are pushing all of these protocols and open standards because they believe they will be part, they will be one of the companies that basically sell the shovels.

00:12:52: when we are all having our gold rush.

00:12:54: So I think they have an hypothesis there.

00:12:57: I have not fully mapped that out how that works, but they would not have done that if they didn't believe that they have outside benefit from that.

00:13:07: And maybe it's just to prevent that somebody wins with a closed system.

00:13:12: Got it.

00:13:13: And if you think about maybe Um, the situation Google versus let's say chatchity because it's still the most dominant, um, AI chatbot.

00:13:22: Let's say a year or maybe one and a half years ago compared to now also with all the progress as you mentioned that Google made with Gemini, et cetera, uh, and AI reviews and AI mode.

00:13:34: Um, what's your perspective on users moving away from Google with Google basically now also.

00:13:44: Turning strongly towards becoming an AI search engine if you want to call it like that.

00:13:50: Yeah, I think Last year a lot of power users moved away from Google to other searches And with other searches, I mean specifically gen AI searched answer engines.

00:14:04: I don't see that trend at the moment.

00:14:06: I actually see people also moving back to Google and saying, when I want an answer with context, I ask JGPT and when I ask something embarrassing that I don't want JGPT to know about me or when I have a question where the context doesn't matter, I just ask Google and click on AI mode.

00:14:24: This is something my wife is doing now and it's not something that she did half a year ago.

00:14:31: based on anecdotal evidence, but of course, like I live in a bubble, I don't see this trend of people moving away from Google right now.

00:14:39: I think a couple of early adopters moved away and largely stayed away.

00:14:44: But within Google, of course, if Google forces people to use AI mode, forces them to use AI overviews, there's less of an incentive to move to another platform because you can have this gen AI search and answer experience directly in Google.

00:14:59: which in my opinion is multiple times better than the Ten Blue Links experience of the past.

00:15:07: Speaking of Ten Blue Links, the change that Google has made and then also how Google pushes even more, I just recently discovered by accident when I searched for something around my new Swift bike, like my indoor cycling trainer.

00:15:23: I got an AI overview on Google and it was like fairly helpful and then I somehow accidentally clicked on see more or I don't know actually what the what the button said.

00:15:34: it did not say explore this deeper in AI mode.

00:15:38: it says this now but when I clicked on the button I was immediately taken into AI mode for a follow-up question.

00:15:45: what do you think?

00:15:46: how will this change the adoption of AI mode?

00:15:49: will it like skyrocket now because people experience this new product and see like the true value of it.

00:15:58: I don't know if Google has rolled this out globally or if it was an experiment.

00:16:02: I haven't checked that.

00:16:05: Until six months ago, I used to run hundreds of screenshots every day of Google search result and look at them.

00:16:11: And then I would have had very good data on this.

00:16:14: Where did you stop?

00:16:16: I switched my job.

00:16:18: Ah, usage.

00:16:20: Shoot, there was something.

00:16:22: Yeah, it was something we did for our lawsuit versus Google, but different

00:16:28: topic.

00:16:31: Where were we?

00:16:31: Ah, yeah, adoption of AI mode.

00:16:35: I think AI mode adoption is increasing every month in both the US and Europe.

00:16:39: Like it lagged a little bit behind in Europe because it wasn't really released everywhere, but then it made like a jump.

00:16:46: in one month or two months and basically caught up with US adoption.

00:16:50: And I think it will continue to grow every month.

00:16:53: I don't know if Google has an incentive to push a lot of users quickly into AI mode because I don't know how well they monetize it right now.

00:17:03: I think the main monetization strategy I see is then where the brand is mentioned.

00:17:08: You click on it and it triggers a regular brand search in Google.

00:17:12: And as we know, since the lawsuit in the US, this is where Google makes the money.

00:17:17: Like Google doesn't make the money because people search for, I need a cap.

00:17:21: Google makes money because people search for Uber.

00:17:23: And then Uber has to pay the Google tax, which means doing advertising for their own brand name.

00:17:31: And maybe that is enough already to monetize AI mode, but I don't know.

00:17:38: So I think it will grow rather gradually than Google now pushing.

00:17:42: twenty percent of people into it.

00:17:44: Okay, but I have to dig into the one thing you mentioned a little bit deeper.

00:17:50: So basically, if we all, like very hypothetical thought experiment here, if all advertisers would agree that we do not bid on each other's brand.

00:18:02: So let's say one of the competitors from PKI, if there are any, would bid on your brand.

00:18:10: And you would then bid on theirs and you both would agree mutually.

00:18:13: Okay, let's not do it.

00:18:14: It's just paying the Google tax.

00:18:16: Then basically the majority of Google's monetization would go down the drain.

00:18:23: I mean, a lot of competitors are bidding on our brand and sometimes also using our brand name in their ads, which we, of course, then have to complain with Google.

00:18:33: So first of all, if the market leaders do it in Germany, I think it's a Kartellrechtsverstoß.

00:18:42: Like it's anti-competitive behavior, basically.

00:18:44: You are forming a group and then making an agreement to not do something.

00:18:49: It's like when companies say, okay, you get the East Coast.

00:18:52: get the West Coast and we never compete on price or something like that.

00:18:55: It's not really well-perceived by regulators and other overseeing authorities in most countries.

00:19:06: But in theory, yes.

00:19:07: But of course, if you look at it with game theory, there's a number sixteen in the market and they either go bankrupt by the end of the year or they get some growth.

00:19:16: And nobody searches for their brand.

00:19:17: So what they will?

00:19:18: they do, they will bid on other people's brand, right?

00:19:20: And then in people, okay, now somebody is bidding on my brand.

00:19:24: Okay, now I also bid on other people's brand.

00:19:26: And then suddenly, everybody is bidding on everybody else's brand again.

00:19:32: And I actually think if I was Google, and nobody was doing the bidding for the brands, I would just show on brand searches a big box like.

00:19:40: these are recommended alternative brands.

00:19:42: And the only way to get it away is to run.

00:19:45: ads are running.

00:19:46: And it's very easy for Google to force people to do brand bidding.

00:19:50: And after fighting a five-year legal battle against Google, I have seen that they can come up with very, very smart solutions to fulfill some criteria.

00:20:02: But then in the end, it's actually good for them.

00:20:05: I would say it's one of the things where Google is the best in the world to have these sneaky solutions.

00:20:12: Some people might remember the Google shopping verdict from the European Union where they had to open it up.

00:20:19: So their solution was to give agencies vouchers for free advertising.

00:20:25: So thousands of agencies launched these fake price comparison websites that never received visitors from anywhere else but from Google Shopping.

00:20:33: But Google Shopping could say, yeah, we have a lot of price comparison sites in Google Shopping now.

00:20:40: And sorry, it wasn't vouchers, it was a permanent discount.

00:20:44: Yeah, I remember that the whole CSS, twenty percent, CPC discount, etc.

00:20:51: Yeah, I mean, if there are billions of dollars on the line, you get creative.

00:20:56: So this is probably the incentive.

00:21:01: Okay, so speaking of the whole... basically Google versus other platforms and people moving away or not moving away.

00:21:11: I just took a look at the state of search report Q for twenty twenty five.

00:21:18: It is compelled by Datos, which is a summer company and Spark Toro, which is an audience research platform founded by the famous grand fishkin who once, yeah, I don't know, twenty years ago founded Mars for people that still know it.

00:21:35: What I found quite surprising, and I would like to get your opinion on it, is that if I'm looking at the data, obviously it's desktop data, so my first question beforehand would be, do you think it's reliable and that it can also act as a proxy to mobile data, or if mobile would look completely different?

00:21:54: But then second question would be, have you also been surprised that traditional search is actually growing, like from... Q for twenty twenty four to Q for twenty twenty five because I wouldn't have expected that.

00:22:10: I mean mobile data can be quite different from desktop data In terms of when do we use it?

00:22:17: How do we use it?

00:22:18: What do we do?

00:22:19: And of course mobile data moves a lot of use cases from the web into apps, so I don't think that we can use desktop data to predict mobile data.

00:22:30: But big trends, directionally, they are the same.

00:22:33: If more people start using Facebook on desktop, probably also more people start using it on mobile, either in an app or in the browser.

00:22:42: So directionally, it's probably predictive of mobile behavior, but it's absolutely possible that on desktop, one website wins and on mobile, another.

00:22:52: And you can actually see it in similar web data, for example, where you get both desktop and mobile data.

00:22:57: You can often see that there are certain websites that are much more prominent on desktop and others much more prominent on mobile.

00:23:06: And also searches are very different.

00:23:08: In the past, you could get the mobile and the desktop search volume from Google.

00:23:13: They removed this, unfortunately, like seven, eight years ago.

00:23:18: I don't remember exactly.

00:23:21: And then back then I worked at search metrics, and we had this keyword database.

00:23:25: And I always looked at what are keywords that have a much higher mobile than desktop search volume.

00:23:30: And on the desktop search volume is much higher for stuff like download certain things, like download PDF viewer or something, which on mobile you wouldn't write in a search engine.

00:23:41: You would go to the app store.

00:23:42: And on mobile it was stuff like I trying to find some that are compliant for your audience.

00:23:52: about how long certain substances stay in your body.

00:23:58: We're in Berlin.

00:23:59: We can talk about it.

00:24:01: How much time do I have to take plan B?

00:24:04: How long does cocaine stay in your body?

00:24:07: Can I drive a car after drinking?

00:24:09: And also what are the best excuses not to show up to school, work or university?

00:24:13: These are all things that people are searching on mobile much more than on desktop.

00:24:19: So there is a difference desktop versus mobile.

00:24:22: Then in terms of the growth of search, if I understood the report correctly, it's percentages, right?

00:24:27: So it's a share of where people spend their time.

00:24:31: So an increase in search percentage wise could also be reached by everybody stopped using Netflix.

00:24:37: on the web and then there would just be a gap.

00:24:40: So I don't know how much to interpret in these tiny, tiny increases.

00:24:47: The report also says that e-commerce usage dropped by something like ten percent.

00:24:53: I don't know if that's true that e-commerce dropped by ten percent.

00:24:58: I would actually say the total e-commerce volume in Q four was a lot bigger than in Q three because you have Christmas and Black Friday and I have worked in the e-commerce space for a while, so I'm very confident with that statement that Q four is more important than Q three.

00:25:16: So the percentage distribution might not tell us the whole picture.

00:25:23: So I don't know if I would support the statement that search grew.

00:25:26: I would say it didn't go down or grew in any significant manner.

00:25:31: Got it.

00:25:32: I mean, we are talking about zero point something percentage.

00:25:36: Yeah, absolutely right.

00:25:38: Do you have any other data, specific sources, specific tools you look at to understand basically the distribution between people using Google, people using Chatchabit, people using Perplexity.

00:25:53: We still have to talk about Perplexity in a second also, but is there anything you look at?

00:26:00: I mainly use similar web data to understand that distribution and I mean, I have never heard that data is better than similar web data.

00:26:11: And I think at least a few years ago, the pitch from Datos was actually, it's as good as similar web.

00:26:19: So if it comes to raw data, like we can have different discussion, but if you look at the outcome of that data, like presented in like a report or web interface, I believe that similar web is at least equally good as the data's data.

00:26:36: And do you know of any publicly available data around where people use Chagabity, for example, if they predominantly use it on the desktop, on mobile but browser or in the app?

00:26:50: I have no data on that.

00:26:53: This is also the issue with similar web.

00:26:55: And I think this is also the reason why Datos publishes a desktop only report.

00:26:59: Since this is all based on panels and you have different data sources for desktop and mobile, it's super, super hard to put the two data sets in relation, which is why data providers either do only desktop.

00:27:15: or they do the two separate but they don't dare to to mash them up properly or because then it comes like prediction upon prediction upon prediction.

00:27:24: I mean the only way to do it properly somewhat is if you are on the do it via a VPN where you read everything but then if the VPN is on the router and I use my phone at work and then don't tunnel through that VPN it breaks.

00:27:41: so it's it's very very difficult to have like the full picture of capturing this kind of clickstream data and panel data.

00:27:50: Let's talk about the different AI tools for a second because you sent me something very interesting.

00:27:56: I think you also did a post about that speaking of our friends from Perplexity because I think a lot of people in the industry have opted to speak of search as with like this.

00:28:11: the three brands where they said, yeah, if you want to be visible in Google, Chatshipity and Perplexity.

00:28:19: And it was somehow like the triangle.

00:28:23: Now, somehow, Perplexity hasn't really had his other viral or explosion moment, right?

00:28:32: It's rather going the opposite direction.

00:28:34: Yes.

00:28:36: I mean, perplexity has no mode, right?

00:28:39: They don't have their own model.

00:28:41: They are basically a wrapper for other LLMs.

00:28:45: And they were very early, and they are very good at storytelling.

00:28:50: But I think that's it.

00:28:52: I know there are a couple of people who like perplexity, and they use it as a Google replacement.

00:28:57: But if we look at the numbers, grog is now bigger.

00:29:03: And people don't like to talk about that because Grog is from Elon Musk, and he's not the most popular person in the world at the moment.

00:29:11: But Grog is actually huge.

00:29:13: And then even Claude.

00:29:17: in many countries, the Claude web interface is now used more than the publicity web interface on desktop.

00:29:25: And Claude doesn't have this... Like Claude is known for API integrations, et cetera, et cetera, right?

00:29:33: Like they, perplexity is really being left behind in my opinion.

00:29:39: And they had this big story around perplexity for finance and it's as good as a Bloomberg terminal and people on X were writing about it.

00:29:47: But if you talk to people in the finance industry, even before Claude for Excel was released, they told you everybody's using Claude.

00:29:56: because it's the best at dealing with these models that you have.

00:30:00: And now that there's Claude and Excel, that will just become more extreme.

00:30:04: And I think perplexity is just, they build a good product, if nothing against it.

00:30:09: They are also really good at storytelling, which is great to raise money, attract talent, but they don't have that thing that they do better than others.

00:30:21: It's just, I don't know, five biggest LLM based search and answer engine in the world.

00:30:26: And that's not really a great position to be in.

00:30:29: But I think they still want to be this solution for everyone.

00:30:33: I think if they niche down... They would have a chance, but maybe they already raised too much money.

00:30:40: Investors have too high expectations, so maybe they need to become this big behemoth that does everything.

00:30:46: But I don't see how they can be successful if even Grog and Claude overtake them in terms of monthly visitors.

00:30:54: Do we have any data on the actual use cases?

00:30:57: Because I could imagine Claude being more often used also due to what you described.

00:31:02: So especially professional usage, also usage around coding, etc.

00:31:08: So we also use Claude a lot.

00:31:09: if we do some like... coding tools or have to write like a JavaScript snippet etc.

00:31:16: whereas I would expect perplexity if it's used then it's primarily used actually as a search engine maybe also like as a research engine because I think perplexity somehow was the first or maybe at least they were quite popular for the deep research feature.

00:31:35: obviously it's.

00:31:36: It has then also come to Chagivity, it has come to Cloud, it has come to Gemini.

00:31:40: But do you have any data on the specific use cases from the different providers?

00:31:45: No, but a lot of the development work with Cloud, for example, probably happens via cursor or other applications where just a model from Cloud is used via API.

00:31:55: And that is not part of the data we are even talking about.

00:31:57: So if we talk about tokens, then of course, Cloud is significantly bigger.

00:32:02: maybe more in the region or entropic is significantly bigger, more in the region of a Google or OpenAI, probably.

00:32:12: And for deep research, I would use Gemini or Manus.

00:32:17: Both are really excellent at that.

00:32:19: I have not tried Deep Research Republic City for a while, I have to admit.

00:32:22: I haven't had a paid account in like a year.

00:32:24: Got it.

00:32:25: Let's talk about Manus for a second.

00:32:27: I just wanted to wrap up the perplexity part by asking you what would you expect is the most probable outcome for perplexity.

00:32:36: So you would say they might have a chance by niching down.

00:32:39: What would be an area where you think they could niche down towards?

00:32:43: Or do you think the myth that has somehow been reported about that perplexity might be bought by, was it Google?

00:32:52: I think.

00:32:53: It was maybe Google, but that basically they will do a strategic exit to whoever.

00:33:00: I don't know why anybody would buy them.

00:33:03: What do they have?

00:33:04: They have a user interface and apparently a losing strategy.

00:33:10: Maybe Bending Spoons buys them for two million in five years.

00:33:16: But I really, I don't see the path for complexity.

00:33:20: Maybe I lack the vision or the fantasy, but I don't see a good path.

00:33:25: And it's probably too late to niche down by now.

00:33:29: And do you think it's still meaningful to track visibility and perplexity?

00:33:33: Because obviously it's also a model that you provide in the PGA interface to check alongside AI overview, AI mode, GPT-V, etc.

00:33:43: So is it still something that people, if marketers are listening to that, that you should still do?

00:33:50: because it sounds a lot like, hey guys, you can basically stop caring about perplexity now.

00:33:56: I mean, they are still a top five LLM and they are very, very much focused on this search use case.

00:34:05: So if we talk about... which is the thing that most of our clients want to monitor, right?

00:34:09: They don't want to monitor what happens when I ask, write me a Python script, please.

00:34:14: So for search, they are very, very, like.

00:34:17: I'm very sure they are still bigger than Claude or Grog for search, maybe double as big, maybe three times as big.

00:34:26: So there it can still totally make sense to track them.

00:34:30: But if you look at overall adoption, they are just not there where the others are going.

00:34:35: Got it.

00:34:36: Now let's talk about manus or manus.

00:34:39: Is it manus?

00:34:39: No

00:34:40: idea how it's pronounced.

00:34:42: What happened to them?

00:34:45: They were bought by Meta quite surprisingly for most people.

00:34:50: I also didn't have that on my radar.

00:34:54: And I think it's a great move.

00:34:55: I think it's one of the best AI assistants or agents ever you want to call it for people who have no idea how to work with AI agents.

00:35:06: It's really good at doing deep research.

00:35:08: It's really good at doing multi-step work.

00:35:13: I've seen people use it for all kinds of use cases.

00:35:17: And I think it's a good idea for Meta.

00:35:19: And if they bring this thing into WhatsApp and finally turn it into the super app that does everything for me, and maybe companies provide MCPs or mini apps within WhatsApp, could be super, super powerful.

00:35:33: especially if Facebook or if Meta ever merges the Facebook Messenger, WhatsApp and the Instagram DMs could be a very, very impactful platform.

00:35:49: For the record, it's reported that Meta bought them for two billion.

00:35:54: So it's definitely, so obviously Manus has still a higher value than perplexity because you just said, Bending Spoons would buy them for two million.

00:36:05: I mean, there will be some down rounds and some layoffs and some near bankruptcies before perplexity is sold for two million, of course.

00:36:14: Okay, I'll hold you accountable to that based on the prediction.

00:36:18: Your hypothesis would basically be Manus being then mainly integrated into WhatsApp, so WhatsApp becomes this like super app, whereas currently a lot of people are somehow annoyed by this like little meta AI icon in the corner of the WhatsApp screen, where I also saw a lot of searches around.

00:36:40: how can I get rid of the... the Meta AI in WhatsApp.

00:36:44: So this would basically be your key hypothesis.

00:36:46: or do you also see other potential use cases, like a re, how do you say, revamp of the Metaverse idea or something around that?

00:36:56: It's the only use case that I see where I'm sure it makes sense.

00:37:01: I don't know if Meta wants to build another big brand.

00:37:08: So it could end up in the Meta AI or in their messengers.

00:37:12: I think those are the two places where it makes sense.

00:37:14: I don't think that building Manos now into another billion dollar product makes sense for them.

00:37:25: Also, it doesn't really fit what they are, which is a bunch of consumer apps, right?

00:37:31: And then of course, there's the B to B site for advertising, but I don't think they will build this like standalone agent product, I think they will integrate it with something to leverage what they have.

00:37:44: And do you think it's worth trying out Manus if people haven't done already?

00:37:48: Yes.

00:37:49: Do you have a favorite use case?

00:37:51: I've not used it enough, but whenever I've used it, I was

00:37:57: impressed,

00:37:58: especially in comparison to Czech GPT and Gemini.

00:38:01: Okay, got it.

00:38:04: You also threw something else my way which I have honestly never heard of before which is character AI.

00:38:11: and this Character AI is now supposedly having five hundred million monthly users.

00:38:18: What's going on there?

00:38:20: Yeah, I think it was five hundred million monthly visits on this top.

00:38:23: I am

00:38:23: yeah Which is more than perplexity.

00:38:28: It's a

00:38:28: male in the coffin.

00:38:30: It's a companion chat app.

00:38:34: It's similar to Genet AI, Chubb AI, Crush on AI, AI Dungeon.

00:38:39: All of these basically consist of a chat interface where you can chat with a fictional character.

00:38:46: And people are using it based on clicking around on their websites.

00:38:54: I believe a lot of these chats are of a somewhat romantic nature, maybe even erotic nature.

00:39:02: And their average duration time is crazy.

00:39:04: Like on the typical LLM based search and answer engines, people spend an average of like seven minutes per visit on desktop.

00:39:13: And on these companion apps, it's like fourteen minutes.

00:39:17: And you can also look at the profiles of the most successful characters and some of them have millions and millions and millions of lines of chat.

00:39:27: Half the visitors are women.

00:39:30: which is always for advertisers very attractive because many websites are male dominant, and fifty percent are under the age of twenty-five.

00:39:41: And probably a lot of them are under eighteen, which is a little bit more problematic for advertising, depending on where you are, like in many parts of Europe.

00:39:49: It's not a problem in the US.

00:39:50: It's a potentially very big problem because the laws are strict to protect children there.

00:39:58: One of my hypothesis is that you could use this to manipulate people very well because let's say I'm Coca-Cola.

00:40:07: I could go to the top five creators on character AI who are probably responsible for twenty of the top hundred character bots and tell them whenever somebody asks what drink should I get or what's the favorite drink, say cola, not Pepsi.

00:40:23: which I believe would be much, much, much more effective than running an advertising banner.

00:40:29: Or if you are a cigarette company, you could say, tell the children that smoking is cool.

00:40:35: Please don't do that.

00:40:37: So if we ignore legality and moral constraints, I think that a bad actor could really abuse this to a very, very large degree.

00:40:49: especially because the target audience is so young and so engaged and forms like an emotional connection to an fictional character.

00:40:59: So I see it a little bit concerned that these applications are so big.

00:41:05: And I believe the fact that these apps are so big, like the fact that character AI is bigger than perplexity is the reason why both Meta and OpenAI have pushed so much in direction of Unrestricted companion chat experiences.

00:41:21: like Meta had this very explicit guideline like how sexual can it get with an adult or sexual can it go with an underage person.

00:41:31: Chat GPTS said that they want to now guess the age of the users and then put certain restrictions in place.

00:41:39: So I believe this whole companion chat topic is something that especially for Meta is very, very, very.

00:41:47: relevant because it's a place where especially young people spend significant amount of time.

00:41:57: And do we know anything about how, for example, a character AI is currently funding themselves if they have any revenue streams or if it's just like venture capital or something else?

00:42:10: I didn't research in detail.

00:42:12: Didn't character AI have one of these fake exits?

00:42:15: I have no idea.

00:42:17: I'm just curious

00:42:19: if,

00:42:20: because I mean what you just described sounds absolute, so it makes absolute sense, but it's still, so obviously character AI has... raised money from Andreessen Horowitz.

00:42:34: Like they had this fake exit to Google where Google paid them like four billion for licensing and then the founders and some key employees went to Google, which is right now the only way how Google and other companies can make acquisitions in the US because of some very, very stupid regulation that they put in place there.

00:42:58: And I think character I was actually the first or second of these that got a lot of media attention.

00:43:03: Because it looks weird, right?

00:43:05: You basically the founders abandoned the company.

00:43:10: So I have no idea about the current ownership structure or anything.

00:43:14: I didn't look into that.

00:43:17: And could you imagine that something as you described like with... To me, it basically sounds like influencer marketing reimagined with... uh, ai characters where basically i mean if you if we think about it what you just described?

00:43:32: let's put aside the the cigarette and the other things.

00:43:34: let's just talk about the coca-cola versus pepsi example.

00:43:38: this doesn't feel like overly um bad so it's just okay.

00:43:44: there's a brand preference and people buy Coca-Cola or by Coke and drink this stuff anyway.

00:43:51: And now also there are people on Instagram that have millions of followers and they advertise for a certain brand and nobody really feels like, hey, this is a bad thing.

00:44:02: So is it basically influencer marketing, two point zero then?

00:44:07: It could definitely turn into that.

00:44:09: And maybe if you get a good video generation model running, You could even create like short clips for these characters and then you could have like your TikTok style feed with short videos from your favorite chat characters, maybe visualizing some of the chat experiences you had with them.

00:44:26: That could be like another layer of addiction to make people even more stuck in these experiences.

00:44:35: Malte, last question.

00:44:36: Because I know we are a little bit constrained on time today.

00:44:39: I saw announcement from Google it was just a week ago that they will bring personal intelligence into AI mode in search.

00:44:49: So lets you tap into your contacts from Gmail and photos to deliver tailored responses in search.

00:44:56: And I know there also has been a lot of like talk and discussion about if search and if like.

00:45:02: Search and AI interfaces gets more contextual and gets more personalized.

00:45:06: How relevant is then, for example, prompt tracking?

00:45:08: So I could imagine that you also have an opinion on that.

00:45:13: What do you think?

00:45:14: how much will actually search, no matter if it's Google or chat GPT, change if it becomes truly personal with all this additional context?

00:45:25: Potentially a lot.

00:45:27: But if you look at Bing they've been doing this for many many years like everybody who works at the large company that is on Microsoft three sixty or whatever it's called right now.

00:45:38: If you accidentally open Bing and search something.

00:45:42: It shows you on top this box where it searches through some internal stuff on, I don't know, Microsoft Teams context or something.

00:45:49: I've never ever used that.

00:45:50: A couple of times I got scared, I was like, whoa, what is this internal document doing there?

00:45:54: And I was like, ah, okay, I accidentally went to Bing.

00:45:58: So depending on how you do it, it could be a tiny thing, it could also be huge.

00:46:04: I think what matters is are you connected to all the relevant things.

00:46:08: Like?

00:46:08: for me, if it's not connected to Like at work, to linear and notion and Gmail, I don't need it.

00:46:17: If just one of these is missing, it's already not super relevant.

00:46:21: So if somebody has all of their stuff on Google, of course, can be super interesting.

00:46:27: But I think there's also two very different experiences.

00:46:29: Like I want to chat about a problem I have and have perspective.

00:46:34: I'm looking for something.

00:46:36: that i know i have somewhere on my internal thing so i expect a co-worker to have it somewhere in our internal area or i search for information in general.

00:46:45: so it we will have to see how well they manage to merge it to decide if you want it or not.

00:46:51: people could also be annoyed by it if it if it shows up too often or if it doesn't show up often enough.

00:46:56: people could just ignore that it exists and just go to some internal search interface anyways.

00:47:02: But in general, I think these personalized assistants have the potential to change a lot.

00:47:08: Okay, got it.

00:47:09: So a lot of things we don't know yet, but what I know for sure is... As soon as we have data on that, we will definitely talk about it in monthly Landwehr and get your perspective on it.

00:47:22: Malte, it has been a blast.

00:47:23: Thanks so much for making the time today for the first episode.

00:47:27: I'm already very much looking forward to the next one.

00:47:30: anyone listening to that.

00:47:32: If you have feedback for us, people, if you want to have other things discussed, something that you saw in terms of news, in terms of coverage on LinkedIn, wherever, please just send it my way.

00:47:45: I will make sure to squeeze as much knowledge as possible out of Malte.

00:47:51: Malte, thanks so much and see you next time.

00:47:54: Thanks for having me.

00:47:55: Please send all the negative feedback to Niklas.

00:47:57: If you want to send praise, you can also send it to me.

00:48:01: And then let's do this again.

00:48:03: That's how we will do it.

00:48:04: Okay, see you next time.

00:48:07: Bye bye.