Does ChatGPT Actually Have A Higher Conversion Rate Than Google Traffic

So does ChatGPT actually have a higher conversion rate than Google traffic? Neil has actually done a study on this. So Neil, what have you seen?

So we see when you look at analytics and visitor-to-conversion rate, yes, it is higher. What analytics is not reporting is when you ask people their journey and when they go through ChatGPT to make a decision and then purchase, the amount of times they interact with the brand before they visit the website and do their research is much more higher. So instead of someone searching on Google, going to your website, doing research, then going to another link, doing research, then going to another listing on Google, doing research, and then deciding which site to buy, they're doing this research on ChatGPT. So you're not seeing the visitor count. When you look at true impressions to conversion rate, We don't think it's higher. I don't have exact data because they don't provide it, but we don't think it's higher because people are interacting with the brand way more often than they're actually clicking over to a website and then buying. Typically they're doing their research on ChatGPT, then when they're ready, they're clicking over and buying, but they've already done a lot of the research and they don't even have to go to the website to finish their research in many cases.

It's interesting because I saw these numbers this morning, but it took, ChatGPT about 4 years to get to a billion queries a day, and then Google took much longer, right? But the caveat is there wasn't an LLM in the past. The way now we're asking a lot more questions, so naturally there's going to be more queries, right? So it's not exactly apples to apples.

It was just like Threads was the fastest app to— was it a million or 100 million signups or something?

Yeah, 100 million in like a week or 2 weeks. But it's like, dude, no, you just piggybacked off the distribution you already had. And so, but in this case it's like, no, the search behavior actually changed. And so it's not apples apples to apples. But again, like we're not saying we don't love ChatGPT, we do. But, you know, Google's not exactly dead yet.

No, Google's not dead yet. And Sundar even, let's say ChatGPT is getting a billion queries a day, all right? Google's at 13.7 billion a day, and that was last year. During the Google I/O event, Sundar said search has already grown like more than 10%-ish or something like that because of AI.

No, their search, their, I don't know if we're talking about the same thing, but their coding velocity has increased 10%. So actual productivity in their organization, it's not just Like Meta's like Mark Zuckerberg's like, oh, we're 30 to 40% more code. Sundar's like, what's more productive for us is actually looking at engineering velocity. That's the 10% that I remember.

There was also, you could be right, but I thought I saw some stat that he was saying search. Oh cool. AI Overviews is driving over a 10% increase in usage for the type of query. So they're seeing a 10% plus increase in usage on Google.

And what do you see on X now? Oh, Oh my God, Google is stealing our traffic. Da da da da da. It's like, what are you going to do about it?

But let's just use rough numbers. If they're at 13.7 billion a day and you got a 10% increase, you're at another extra 1.37 billion searches a day.

Yeah.

Which is kind of crazy to think about. It's just like, oh, our AI just helped us increase our search volume by roughly the amount that ChatGPT gets on a daily basis.

This is also interesting. So Patrick stocks this product advisor tech. technical SEO guy, brand ambassador at Ahrefs. He wrote this post on how AI assistants are breaking web analytics and hurting their future. So the premise of this post, and it's a 6-minute read, but his whole premise is, look, search is moving from traditional search engines to AI assistants, but traffic from many of these sites isn't being tracked properly in your analytics. And it's not necessarily your fault. It's their fault, right? So he was just saying, look, you know, their LLM filter in Ahrefs web analytics, notice some common AI assistants missing from the list. They're in our filters, but we aren't seeing any data from their sites, right? So you see ChatGPT, Gemini, whatever, and it talks about the invisible traffic problem comes from these systems stripping the referral value. So I think this is a minor problem for right now, but just keep in mind that we're still very early days right now. Even the CEO of AMD yesterday said we're very early innings of AI right now. Yeah, so just keep in mind this stuff evolves. Like Google didn't become what it became like overnight.

No, and I do believe that, I do believe that when you look at Google and AI and everything that's happening, With ChatGPT, I think eventually they'll catch up to Google with a lot of the data and the analytics and the reporting functionality because they announced that they're doing $10 million or $10 billion in ARR and they are trying to— and they're trying to hit— or OpenAI, and they're trying to hit $120-something billion by 2029.

Dude, it's crazy because a couple of years ago it was like $2 and then $3 and then now it's like $10. And so it's compounding pretty quickly.

Yeah. I don't know if it went from $2 to $3. I think the numbers stick out for me.

I'm sorry, that number, that number comes from Anthropic. That's actually Anthropic, $2 to $3. That's recently.

Yeah, but ChatGPT's numbers is growing at a rapid pace. All they got to do is release more features, ads, and all this kind of stuff, and then boom, I actually think it'll grow much faster.

Yeah. So OpenAI's revenue through the years. Oh, here's an AI overview. Oh, this is a helpful AI overview. Look at that. 2023, $0.5 billion. 2024, $3.7 billion.

Okay.

2025 projected to reach $10 billion in annual recurring revenue, an 82% increase. They expect to generate $12.7 billion total revenue. For the year. And then 2029, to your point, OpenAI aims for $125 billion in revenue.

So I was talking, I was talking to a guy named Daniel George, and Daniel created this startup, Sequoia-backed or Sequoia Scout-backed. I don't know what you want to end up calling it. It's called Twinmind.ai. Here's an interesting stat for him. Okay. They have a lot of users. They're growing virally, just mainly through word of mouth. It's never forget anything. It's pretty much your second brain. That's their model. So it can listen on everything that you're doing, create AI notes, to-dos, answer answers before you even ask during meetings, lectures, conversations. It can pretty much help you with anything they're doing in multiple languages. It's free. When you look at this company, they're leveraging OpenAI's API and Gemini and some of these other players. Their cost per user, I believe, was roughly $30 a month when they first rolled out. Now their cost per user is around $3.

You're talking about just cost of goods sold, really?

Cost—

To service a user.

The cost for all the APIs.

Okay, yeah.

So it's gone down around 10x in like a year, year and a half.

Yeah.

What's the cost gonna be in—

That's 90%.

Yeah. What's the cost gonna be in another year or 2 years? 'Cause these models are getting more efficient. They're buying their own energy nuclear power plants and the chips are getting more efficient. So it's like, what's the real cost gonna be in another year or 2?

You know, every 2 generations, It basically goes down like 100x, something like that. Every 2 generations for these models. Yeah. Like what used to be really expensive 2 years ago is now like cents.

So let's go back here. What do you mean by 2 generations?

Like ChatGPT models. Like it's like, so right now we have O3 Pro, right? And then you had O3 and then you had 4.1 and you had 4.0, right? But then you had ChatGPT-3 before, right? Those, those, the cost for tokens used to be way more expensive. And to your point, they're dropping like 90%, 99%, right? And then you have DeepSeek come out, this 99% drop, right, from these tokens. And so to your point, it's becoming a lot cheaper. And they call this, remember we talked about Jevons paradox, the more something gets used, it actually proliferates even more. And so the costs are coming down. You're gonna see a lot more queries, a lot more energy being used, a lot more compute being used.

And to go back, when you're saying every 2 generations, you know, like there's 4, there's 4 Mini, is that considered, do you consider that another generation?

Like 4, 3, from going from 3 to 4 would be one generation. Okay, so you're not considering Considering all the like sub-variations. Yeah, yeah, no.

Okay, that's still, you know, crazy to think about. I don't know when they're coming out with 5. ChatGPT.

They say they're holding off a little bit, I remember.

Oh, mid to late 2025.

Yeah, so they're holding off a little bit. Like they're pretty close already, but Sam Altman said like they're going to wait a little bit because it's that amazing.

ChatGPT-5 is expected to be a significant upgrade, potentially surpassing the capabilities of 4.5.

Have you used O3 Pro?

I don't know which version I'm using. I just know I paid $200 a month.

Yeah, yeah, so you have O3 Pro. So they just came out with this like 2 days ago, right? And what it does is, O3 is a really good reasoning model, but O3 Pro, here's what I did. I was like, okay, tell me what are the highest leverage things that I can do with this model that I cannot do with O3 or any other deep research, whatever other model out there, right? And then it basically talks crap to me in the beginning. It's like, you ask very, like, the things you ask me are like how to do things, right? It's like, no, here are the highest, like 12 leverage, highest leverage prompts. And, uh, that's supposed to be your kids? No, they're yelling.

They're calling for me.

Oh, they know you're here.

Uh-huh.

Okay.

So going back to 3 versus 4, when you're talking about— I thought when we're just using the interface on the website, we're just using the latest 4 model or whatever it has.

No. So by default, you're on 4, right? But I keep switching it to O3 Pro. I wanted to see how good this is. Let me read you some of the prompts, right? So here's what it said. It's hard truth. Unless you push the model beyond straightforward, tell me what to do requests, You'll keep getting answers you could have stitched together from me, O3, plus a few hours of Google, ChatGPT, and Excel. The O3 Pro tier only earns its keep when you force it to synthesize large, messy, multi-domain inputs, run deep quantitative or algorithmic reasoning, and deliver personalized decision-grade output that you struggle to reproduce manually. So let me give you 2 prompts and then we can, we can move on. So the exact question to ask: ingest these 5 years of Single Grain and Carrot P&Ls plus my personal spending and create a Monte Carlo forecast runway of a runway B, probability of hitting $200 million ARR by 2030 under 3 hiring plans. Show the top 3 levers that shift the distribution greater than or equal to 10 percentage points. I dropped this in, it thought for 20 minutes, and then it gave me an answer. I'm going to give you one more. So write Python that optimizes my LinkedIn ABM pricing tiers using historical close rate data and price elasticity curves. Recommend the 3 highest EV price points and expected churn impact. So highest EV would be expected value, and then you would ingest data here, you would preload data from your, like, your close won, lost data, your pricing and usage metrics. And so this stuff just getting a lot better. You're a data guy. Imagine all— so I just take all of these prompts and I just, I have them all running at the same time thinking about these 12 things and they're all like thinking for 15, 20 minutes, taking all the data you have. You have— it's an even smarter strategist in your pocket now. Yeah.

And I think it's just going to get better and better. I'm just curious to see what 5 is going to do.

Yeah, we'll see.

But it's already really good.

Yeah.

I do believe it hallucinates.

For sure.

And I have that in one of the topic sheets 'cause I was just in Calgary and I asked everyone in the audience how many times or how many of you guys have used ChatGPT or any AI model and seen that just the answers are just really off. And I forgot what percentage in the room raised their hand, but it was something significant like 40-ish plus percent.

Yeah, and I think the rest are lying. I mean, it's— but it's like humans, we're wrong all the time.

Yeah, but even the ones that are hallucinating, I asked them, You know it's hallucinating. How many of you still use it to do research, figure out what products to buy and get help? And almost all of them still raise their hand because they're like, we know it hallucinates. You double-check it. The value it provides outweighs the times it hallucinates. Right.

And the only things, the only 2 things that are certain in life are death and taxes. So, you know, nothing else is certain. Even if you have a genius computer, who cares? Like it's just never that 100% right.

That is true. Death and taxes are certain, but depending on where you live, because some countries have no taxes.

Yeah. But they'll find a way to tax you some way, shape, or form. Something else in your life— inflation will tax you. Something else in your personal life might tax you. You never know. You're always taxed in some way, shape, or form.

Now, did you see the inflation numbers that came out?

No, no. What was it? What was it? Was it down?

It was down.

Yeah.

Less than expected.

But see, now the narrative is like, oh, well, wait till the tariffs kick in and wait until this and that.

Dude, that is what they actually said. They're like, well, we feel that the tariffs will kick in in the next few months and then go from there.

I don't know. The world's unstable, man. It's like, so, um, My, our mutual friends Syed and Yaniv came in yesterday. Once they left, I'm like, holy God, so country got attacked, right? Iran got attacked. I'm like, what is going on? So yeah, anyway, not a politics show.

No, not a politics show.

If you're in B2B, here's the deal. LinkedIn account-based marketing sucks, but it doesn't have to. With Carrot, it allows you to create personalized ads towards your customers and also personalized landing pages that align with the ads. So you're going to have higher conversions and influence pipeline reporting to prove that your ABM works. So if you want to learn more, just go to karrot.ai and you can join the waitlist. Again, it's karrot.ai and we'll see you on the other side. Here's another one that's really interesting to me. So we— I guess I'll experience share here. So you weren't here the other day, but I was going to have you come to this, this founders dinner where we went to eat hot pot at Hai Di Lao and with Syed. Yeah, Syed was there and then we had some other founders and we were just talking about AI. So one guy brought his head of AI, I brought my CTO, and the same thread, I don't know if this thread's happening with you right now. It's like we don't really, I don't think anybody takes any joy in having to cut people, but we are looking for ways to spread AI throughout our organization. So obviously you, the founders will trumpet it from the top, but that's not enough, right? You have to look for people who are willing and If you're willing, like, I will teach you, right? But you're gonna have people in your org, like, especially when you have, you know, thousands of people, you're gonna have people who are unwilling and resistant, right? And that's the problem. And so our whole thing is, and Syed was actually the one that brought this up, it's like you want to find, you want to get people to the aha moment where like you start using ChatGPT or Cursor, you realize you're all of a sudden you're 10 to 15 times faster. You just need to get people to that point and then they can activate on their own, right? But I think the problem is like, You know, the people that are kind of fighting against this, it's like, dude, it's like, it's the equivalent of saying like, oh, don't use power tools to build a home. It doesn't make any sense, right? But I guess my question for you is like, how are you guys spreading AI fluency throughout your organization, or are you even thinking about it?

We don't think about it from that way of spreading AI fluency. We have a mandate. We tell people to use AI whenever it makes sense. It's a little tricky because quite a few of our enterprise customers don't allow us to use AI.

Mm-hmm.

So just some of them, we're not allowed to even use it contractually. And we have, if we want to use it, we got to let them know what tools, get authorization, even if their data's not shared. And then there's also liability issues that we have to worry about on our end if we use it, even if it is allowed by them with their notification. But we have quite a few contracts that are just saying, You cannot use it. So on our end, the way we think about AI internally is if you can do things more efficient, by all means, use it when you're legally allowed to use it. I know that sounds really crazy to think that way, but at size and scale, you also start thinking about wealth preservation versus wealth generation, because for us, it's more important to keep what we have. Yes, I want to grow, but We first optimize for keeping what we have and not losing it. The way we really think about things is we have a team that's dedicated on disruption, and their goal is to just disrupt every single department through AI or any other forms of technology or efficiency or processes. And we found it, it's easier to have a team that focuses on disruption than it is to get every single department to try to disrupt themselves. That's just what we found internally. So, you know, that central team has been doing some really cool work. And then when they find some huge wins, we roll it wherever we can out legally.

So what, this is interesting. So Yaniv and I were talking yesterday and then we went to a friend of ours and they do well as a company. And he was just saying, you know what, like I'm just hiring a lot of people who are 17, 18, 19 years old, right? And These people are very motivated. They're coming out of college, right? And some of these people sometimes they'll ask, they're so motivated with it because they're just trained on this stuff. They don't know any other world where there is no AI, right? And they're just like, hey, can I sleep on the couch? Like they go to office and can I sleep? I'm like, dude, that's amazing, right? And so they're fully bought into this and they're kind of driving this world forward. And in his mind, it's like, and in my mind too, I'm like, a lot of people I talk to from UCLA over here, I'm like, I'd like to recruit more. And so, because again, they come equipped with this. And so again, I have no issue. And what we talked about at dinner is like, we have no issue teaching people who are willing. It is the issue where if the people who are unwilling, that means they don't align with— they don't necessarily align with maybe the future of where we're going. And that's going to become more and more of a thing. And so I think my opinion is the founder has to keep trumpeting it from the top, but you need champions within the organization and people who identify other people and kind of spread this Positive virus throughout the organization.

Yeah. Yeah. If you don't have a culture where people want to embrace it, I think it's going to be hard. There was a guy that we hired a long time ago. He didn't end up working out. He worked for one of the largest ad agencies out there, publicly traded, massive, call it 100,000+ employees. And his job, this was before all the LLM stuff that was really coming out. His job was to automate as many departments as possible and reduce headcount. So when he would go to the departments and try to help them automate and reduce headcount, no one wanted to help him out because it was just going to cause—

this guy's going to cause chaos.

Yeah. So you got to really have a culture that wants to improve and they want to embrace this stuff, or I just think it's really hard to get things implemented. Like we have a culture and a lot of people do want to embrace their technology. We're just restricted in a lot of ways legally, or legally is probably the wrong word, contractually with many of our clients not allowing us to use this new modern technology.

It's not even just that. I mean, you also have to make sure that from a security standpoint, things like we're talking like actual security in your company too. Like you can't, it's not like if you're starting a company, you can kind of do whatever you want instantly. It's not like that. And that's what we were talking about yesterday too. It was coming from a preservation angle, but how do you push this on people who are willing? Again, so I think, by the way, if you're listening to this and you work for a company, You know, you're hearing about all these companies. Okay, this company cut 300 people over here. This company saved $8 million in payroll. Google is paying people to leave. Microsoft just cut 6,000 people, right? And so I think you don't— don't worry about AI slashing headcount. It's more so if you can adopt a willing mindset and once you hit that aha moment where you've become 10x or 100x faster, that's all you really need to do. You just need to flip that mindset and you'll be okay.

Yep.

Yeah. So, but yeah, it's kind of crazy. Like, what was the deal with Google? What, what, what will they pay you? Do you know the terms?

No, but I saw Google recently made a buyout offer for employees. Yeah. Google buyout. Uh, let's see. Google offers buyout for employees in search across, or to employees across the company, including search. Let's see. The buyouts are part of company's effort to reduce headcount, which Google has continued to do in waves since laying off 12,000 employees in 2023. If, uh, some teams are mandating office, uh, returns for remote workers who live within 50 miles, Of an office, which is a lot of traffic depending if it's in Northern California or where it's at.

Or LA. LA's pretty hard. Yeah. 50 miles is like 2 hours in LA. Yeah.

But it's their way of saying, come to the office or quit.

Or else. Yeah. Yeah.

Yeah. They're not, it's not sharing how much they're offering, but I bet you it's really nice.

Yeah.

From a severance standpoint, they can afford it. Yeah. And it makes sense. You just start buying off people saying like, all right, you can leave.

Well, that's it for today, guys. Please don't forget to rate, review, subscribe, and we'll see you tomorrow.