ChatGPT Ads, Agentic Browsing and Websites Built for Agents

0:03Hello everyone. Welcome to our episode 11 of Marketing to Agents. And as always, I have Lalit with me. Hey Lalit, great to have you here.

0:13Hey Ashwani, great to be here always.

0:16Fantastic. So L, we've again as always we have so many things changing over the last one week in this space. today we are bringing in a few of those and of course as always would love to hear your thoughts and insights on some of these things. U perfect so with that let's let's get started. So our first item in docket today is a theme around monetization. Okay. So there are two separate threads coming around this monetization. So first of all, there's this big news out of Chad out of Chat GPD that chat GPD ads is hitting a $1 billion revenue run rate already. So it was I think hardly three months back when they started launching it. we recently heard that they've launched ads in India as well. of course those ads have been live in North America and parts of Europe. So looks like that charge GPD is able to now build another revenue stream and this is just very very rudimentary right now and I'm sure they will pick up pace on this as the usage of the platform also grows up. So that's one side of monetization. The second is that Google in their search ads, they're automatically upgrading all the search ads to AI max, which is where

1:35If the search happens on Google and there are AI overviews happening, if there's an opportunity to get links there, search links there, your ad will show up there as well. So, Google has automatically pushed all advertisers to enable AIAX as well. Now, is that a revenue thing or is that does that help brands as well? That remains to be determined. But what do you think about it now? So now looks like AI engines, answer engines are already maturing and they are now cracking towards monetization. What are your thoughts on this?

2:10Absolutely. I think these numbers are not insignificant. I think OpenAI is reporting a billion annualized run rate, right? revenue run rate and what Google is trying to do with AI overviews with AI max campaign. I think both these you know AI players have a very significant free usage. Google particularly right offers a you know the entire AI overview is free to use and charge GBD free. It is also a significant chunk of the user base right. I think ads are essential for them to you know keep funding free usage and while the free usage also keeps giving them you know the training data. I think what these numbers tell us is that on both sides. in fact there are three parties involved in this you know like you know in in the in the equation of the ad spend and the growth there right first is the platform right second is the consumer and third is the advertisers right I think we we are very clear on the consumer story there that u you know we have seen since last two years that the the trend on AI searches are leaning very strongly towards answers and zero clicks and that is not going to stop very you

3:27Know I would say that's the mainstream now so that's not not going to be reversing advertisers are now also showing confidence in the discovery right and I think the organic way to get recommended has a lot of liability on the platform side because when they are sort of recommending an answer they want to be safe what we are you know analyzing and looking at across millions of runs that we do for our customers I think there is a very strong let's say overlap in terms of brands being repeatedly recommended right so what what I mean to say is that AI engines are playing very safe in terms of let's say their organic recommendations and advertisement is I think is a is is a window that is open for people you know brands who are aggressive in their growth who cannot really wait for the organic optimization. So I think advertisers are also showing very strong confidence in terms of this channel. because as we have always seen like as you finally available for ondemand push button growth through the art

4:44Channels and to large geographies with the inclusion of India. So this is like a very positive sign for the AI search industry as such. I think organic will continue to be important but you know the paid channel gives some sort of a respite or maybe like a like a hedge against the time it takes to get organic. I hope these engines don't screw up you know the way Google actually did screw up the SE results with adding you know the the increasing or rather I would say the creeping real estate right you know it just keep you know creepingly becoming larger and larger right so there was a point where it was hard to even distinguish like whether what you're clicking on is an organic link or not so I hope the AI engines don't screw it up at that level because people tend to you know bring a lot of let's say longtail intent long tail you know what what we are observing is that people are pouring their heart out in terms of let's say when they are giving or rather delegating you know delegating is the right word it's not even searching search is a very shallow piece but they are probably delegating an aspect of their let's say life to an AI AI engine and over there if the monetization aspect sort

6:03Of takes over the the or the the spirit of serving their user you know genuinely then I think it might be a bad thing and organic would still prevail then but as of today I think it's a positive sign that this is getting its feet firmly into the future

6:22Actually excellent point you touched upon in the there that now with so Google when they launched search ads the the information they had about the user was still very limited and based primarily on the cookie data that the user was browsing on the net. But now chat GPD for example has hundreds and hours and hours of conversations with the client. They really know the persona. They really know what the person's you know real fears are or things that they want, things that they don't want. So this is very very interesting where where chat GPD can actually target the user really well and again I think they don't have a huge base of advertisers today because today the ads are extremely irrelevant and they are not even on the same topic but assuming that it gets better over time the ads can be actually very very on point and specific to something that you're talking about right now or concerning you right now. So yeah the opportunity will be there for I think for brands if you look at it from the brands angle it could be a new way if lot of if most of the if we are seeing the behavior shifting from search to asking chat GPD or asking these AI engines. I think ads that way is a is a

7:41Way for brands to then be present there as well. How good those engines are I think they this is just day one of those engines getting built. ads engine usually is is something that can be built very reliably. So I think we should see a lot of those opportunities for brands to capitalize on those intents or conversations which are happening with AI. But yeah, it'll be interesting paradigm for the brands also to explore how they can make the most of this new inventory which is available. and moving on the next major set of news that we got was just around agentic browsing. So we've heard from claude recently claude kova got an upgrade where they had their own they have their own inbuilt browser now we have claude available within chrome for all their plans and then of course perplexities hybrid compute m on Mac that they have launched recently. So a lot of news around browsing and how agents browse network internet now. So what's the takeaway from here like like what do you see where do you see this going like will eventually be 90% browsing on internet will be by agents and then humans will be what kind of will we just interact with internet to

9:00Express what we want and what we need everything else is done by agents. Where do you see this whole agentic browsing versus human browsing going? How does brands what is the implication on brands for example? Do we build website now for users or do we just build it for agents for example? Yeah, I think u just past few weeks I think many significant things have happened which which have which sort of give give us the answer about the future right so there have been on the consumer side there have been few startups like so there's this agent called instinct which is a iMessage and WhatsApp based totally textual interface there's the for few use cases they give you a dashboard URL but I think it's a purely iMessage or textbased interaction with an agent where users are actually delegating real aspects of their life, right? then we also saw like a in past few weeks Grogbot. Grogbot emerging on the horizon and then you know quickly you know getting the spotlight. The growth has been phenomenal. You know many people in my network myself included. a lot of my work is getting done through my chief of staff on Grogbot. so in some sense I think we have to start looking at this as u you know some sort of an agent

10:19Delegation of work right so we are delegating an aspect of our life or like a sub goal in our you know day-to-day like there are many things that we want to get done right whether it is let's say quick research on an upcoming weekend trip or like when is that concert find it out you know or keep scanning for let's say I want to buy this air purifier so keep scanning for let's say discounts, right? So, these are things that they were always either in your head or in your to-do list, right? And you now have an opportunity to just delegate it to a more persistent, more coherent and very amicable entity which is your you know agent. and I think this is so seamless or in some sense I think the the the efficiency with which you know things happen in this new mode of you know AI interaction I think it is it is irreversible in some sense right because just looking at my own personal self I got a lot of stuff done which was which was always like a thought right for example I just be a simple examples. so I have a lot of things that I want to go deep into and my default way to go deep into any new topic that I

11:38Want do not know anything about is to go create a notebook LM for that. Right? So I now have a Crockbot sub agent which is a specialist notebook LM architect. Right? So the moment I come across something new on the internet and you know where I I do think you know I must know something in depth about it. So I just you know drop a message to the Grogbot chief of staff which then delegates it to the notebook the notebook architect and in a few hours I have a readymade notebook built with four or five audio overviews ready for me to quickly consume and you know maybe you know just by next day I can get a lot of in-depth understanding of a topic which was sort of like a question mark in my head right so this is one thing which AI answers could not solved till yet, right? you know, obviously as search could not. So, agents have actually made it possible for humans like me or many people like me to to sort of get as I said an aspect of their life delegated, right? so we if you just start looking at this as not search and something beyond search, which is an actual action that you need to take. If you delegated something to someone most likely you want them to achieve

12:57Something on your behalf, right? So the browsing let's say updates that we are seeing in today's docket, right? things like these instinct like agents and Grogbot and Claude you know being now super compatible with claude sorry Chrome and you know today we just today we saw GP Astra being launched which is exceptional at computer use right so these are all hinting towards the delegation taking more and more of your let's say task list right so AI delegation taking more and more of your task list and your question around what is going to be the future of humans interacting with the web. I think unless something needs a double check, right? I think humans might not necessarily visit any website, right? if it is entertainment, obviously I can imagine, you know, maybe there's no point my Grogbot watching a Netflix movie on my behalf. It's going to be me who's going to watch the movie, right? and if it is something very crucial, for example, let's say buying a property, right? So I might see photos but I might also want to see you know photos to my till my satisfaction right. So I think it's going to the use cases with which would sort of really require

14:14Me to visit the website and have an in-depth session would be only for high-risisk categories where I just want to make make sure that things are fine but in that domain also I do think we will very soon see dynamic UI right so the agent should be able to spin up a UI for that use case which could which would be so much different from the website UI right So you don't have to necessarily depend upon how UX friendly is the website. Agent should be able to get the information present to you in the most user friendly way. For example, some users have this preference of viewing cards and then maybe like dislike on things, right? And some users might have let's say you know might want to read their stuff and then comment and maybe improvise on that, right? So the agent will learn about their users so much and maybe spin up dynamic UIs that are actual necessary to go to websites is going to be you know diminishing and in the B2B context you know which much of our audience is I think in many of our previous episodes we have talked about AI answers and you know the the in-depth let's say series of queries that people you know engage their AI you know AI agents on whether let's say it's

15:34Period or charge GPT the depth with which you go in it's not just going to be stopping at information right if earlier you were visiting the website to book a demo with the sales rep or let's say fill up a form for inquiry now you can let your answer engine or the agent do that also for you right so I think the move is from answer to action so from information to answers was the first milestone from answers to action right is the is the is the is is what we are seeing as a significant development.

16:07That's again very interesting thought that leads me to think that today the way platforms like AirPulse are helping brands get answered get mentions on the answer. I think the next evolution for visibility also will be agent accessibility. So for example the websites have to be compatible to be able to handle agents really well. Everything in the website should be agentfriendly. For example, if there are forms which are there, if there are lead forms, for example, a lot of SAS like it used to be a a basic tactic in SAS where some of your content was behind a login or behind a a form so that you can capture a lead for example. Now when agents are coming there you have to now start thinking about okay how does so so maybe platforms like airpulse can now start helping businesses also get ready for that paradigm where if it is agentic browsing how good your platform is behaving for an agent and is it able to supply all the answers which users might have all the quer so user might never come to the website it could be just their agent coming getting all the information and then presenting a summary to the to the end consumer to take a decision eventually

17:26And if that's going to be the case then I think more than the mentions because and again just extrapolating it assuming that compute cost goes down significantly. Okay, let's say we solve compute really well. Then I as a end user, I don't mind if my agent goes and searches for top 20 players, right? I don't mind doing them to do that as long as it doesn't cost me a lot of tokens to for me to do that and then they can come back and report a great table to me or or give a great analysis based on my needs that hey go and check all I will not limit myself to just top two, top three, top 10 in the category. I can actually expand that search also. But yeah, only if the a but I don't want to do that work. It'll be my agent who's doing the work. So if the agent can easily access the information on on the website, get everything which is relevant very quickly. I think that would be the next paradigm for helping build that confidence as well.

18:31Absolutely. I think what you touched upon is is actually very important and something recently gained lot of discussion and on on Twitter as well. This is web MCP right. So what you are essentially mentioning is that when agents are arriving at a website so how efficiently can they navigate and you know perform the action that is user that the user intended to do. I think WebMCP is is built exactly for that use case where these are not the backend tools that the the company or the organization exposes. is basically the website actions the actions that are that a user is expected to take on the website very simply let's say the Amazon add toart right it's it's a user action that is exposed on the website right and the user can easily quickly have a you know look at the website and then they can figure out where is the add toart button and click it it's very natural for the user like a human but for agents they have to you know take a screenshot analyze the image then you know figure out okay which x ycoordinate is the actual the the add toart button and then you know give the instructions for the cursor to move and then click right this is not very efficient right a lot of tokens get wasted and I think you

19:49Know AI's are going to be AI agents are going to prefer services websites which are token friendly right or to more token efficient right if I were to give an example of the you know example which sort of is analogous to this paradigm from the previous era. this is the website performance equivalent right so Google would Google always used to give references to the page load time right because a crawler when it when it used to come to crawl your or index your website it would come with a certain let's say compute budget or a crawling time budget right and if websites are slow then the the budget is not properly utilized right and the crawler doesn't really index the complete website or not even the pages if it is super slow it moves on. So you as a as a website owner, you lose out on potential search index and traffic, right? Similarly, I think when these AI engines are, you know, crawling across the web, getting to know the website, getting getting to know services, they would they would get to know which websites are MCP friendly because it is kind of like a so the protocol also is declarative, right? You can actually declare the MCP tool sets very much in the markup, right? And although the tool

21:07Is exposed as a a JavaScript function u but the the information about a web MCP being existent is very much in the markup right so agents would start to prefer tools you know for example let's say there are let's say you want to go get your booked travel is a commodity category so many travel websites u and if agents have this prior knowledge that let's say make my base webmcb compliant right or this one OTAA is webmc compliant they would you know prioritize going through that because it is easier right and they also want to they're also sort of in a way fine-tuned to delight the user by doing the job faster very interesting I remember early days of early days of web 2.0 where you know those lot of these websites were getting created where people could like e-commerce itself also you know there were so many things e-commerce was solving to help users build trust in that transaction to navigate easily and now that whole thing has to be redone to be able to do it for the agents because now agents would be doing a lot of those browsing and maybe deciding and maybe you know

22:26Filling up some of that information. So it's interesting that you know within 20 years we have reinvented the whole thing and we're going to start doing optimization on different different parameter altogether now.

22:38Absolutely.

22:40Yeah. So every 10 years I think there's a platform shift right. So from desktop to mobile to now AI. Yes.

22:48Oh yes. And yeah AI I think. Yeah. And our way of engaging with AI also today is primarily text. I know that there are now new form factors coming where you can just talk to the agent but I think those are those will become predominant. I think why text will also have to be the main thing. I think it'll that that form factor might also change very quickly.

23:14Yeah.

23:17Very cool. Like I think these were some of the major news that we had from last week. There were also there were other other critical things which happened in the in the ecosystem. For example, Nvidia had amazing earnings. That automatically means that the whole AI infrastructure buildout is going to continue well into 2027 means that all of the all of the hyperscalers already seems to be forecasting or seeing this whole AI compute AI demand going significantly higher. Hence this continued investment in infrastructure. So we're still looks like we're still in early days of the demand capture here. I think the demand is still many folds and like we're discussing if the compute goes really if the compute or the token utilization becomes really optimized I think the workloads will automatically increase people will want to do more and get more stuff done. so I think that would also be I think we are seeing some early signs of that demand bubble coming up already. But yeah, interesting to see that the ecosystem overall is still expanding very aggressively.

24:28Absolutely. Yeah, I think what it clearly proves that u that there is a huge huge push on obviously sovereign AI a lot of these nations are concerned about it. Second, I think there is a lot of enterprise push on terms of now cost optimizing their AI workload, right? So a lot of you know like Neoclouds are coming up which are offering open source model inference and that are being customuilt for you know various enterprise use cases right and and also inference right the inference time even on the frontier you know models I think the the inference market is actually beyond anybody's expectation Right? Because obviously Javans paradox is is is at play here. Right? So the cheaper things become more and more we are using it. But I think in my opinion the the sheer quality of the the frontier models right and the so I'm actually getting like people like me are getting more and more comfortable delegating more and more tasks. So and only because the models are now delivering on those tasks because of

25:45Their let's say improvement on the benchmarks their coherence their share ability to do you know long long tasks right and complex task is actually improving significantly right so I think this is may this is sort of giving a boost to the inference market and agents per se are going to be like a turbo booster on that right because agents is all in friends right so agents as we discussed earlier the delegation aspect of life for many people right it will free up so much more of their free time they would want to do more and when they do do pick up more projects they would definitely want more of agents to help them in so I think it's a it's a great virtual cycle of inference and from a search perspective or let's say AI visibility perspective I think or rather broadly marketing to agents is what is our mission of of this podcast is you know sort of you know to keep a track of how marketing to agents shape up I think now is the time when marketing to agent is unignorable right so anybody who was just sleeping on this this let's say you know like this topic now has to wake up in an urgency mode and you know

27:04Act on it right because the Google analytics reports that you are seeing in terms of number of visitors. Those are not actual visitors. Those are agents. So you better make sure that your content is well understood, is accessible and your website can let the users perform actions. Absolutely completely agree with you. Interestingly, we living in very interesting times with the radical changes coming every week.

27:34Yep.

27:35So good. Perfect. Yeah, I thank you for this amazing episode again and for sharing your insights. I think we'll see a lot more of these updates coming in as Erin comes closer. But yeah, thank you again Lith for spending the time today.

27:54Thanks Ashwani. Always always a great experience.

27:58Thank you.