I’m sure I’m an outlier, but ChatGPT has cut my Google searches in half, no question. I no longer search for “answers”: definitions, explanations, code examples, recipes, or household tips… all of those were SEOed to hell.
It’s not because of how good AI is (it struggles) but rather how bad search had become. No one from the search provider to the website you land on wants you to have easy access to an answer because there’s no profit motive.
With paid AI services, providing better answers faster IS the profit model. I think something like Claude + Kagi (paid search) would be killer combo.
I’ve actually been kind of nervous about seeing more & more free AI services because that compute is expensive & I’d hate to see the market dynamics that ruined search happen all over again. I want to pay for services with my $ rather than my attention & my data.
Claude 3 + Kagi is the exact combo I use. Claude has been my go to since discovering it didn't have the same level of curation / censorship models like ChatGPT and Gemini have.
I picked up Kagi the day they released their $10/mo plan and I don't see myself dropping it unless their quality degrades or price increases substantially. The search result I want is typically in the first few results, while Google was just ads the first few results, SEO spam, and then MAYBE the result I want near the end of the page.
Kagi users with the Ultimate plan can access most of the major LLMs via the !chat bang too (OpenAI, Gemini, Claude, and Mistral). Regular plans have access to the cheaper models only. Then there’s the FastGPT that runs when the search query contains a question mark
Really makes it easy to use when it’s all accessed from the address bar.
Providing better answers faster was the original advantage that Google had over the likes of Lycos, Yahoo, AltaVista, etc.
We're in the "early Google" days of AI when it's actually a nicer, better experience at least for some queries. When they figure out how to monitize it, it will become the same ad-optimized spam cesspool that we have with traditional search and web pages today.
> With paid AI services, providing better answers faster IS the profit model
I hope this is the case, but aren’t today’s paid services all heavily subsidized by investment dollars, and brand new, and being used to find traction and validate MVPs? I suspect today’s prices and today’s business model isn’t sustainable.
What hope is there that AI won’t end up being much better and more clever at inserting ads in between captive audiences and the things they’re looking for than Google currently is? I get the feeling advertisers are salivating over all the new possible ways AI can sell things, way more than us consumers are excited about having search without ads…
I wish everyone wanted to pay for services with money rather than attention, but that’s a small minority of people for any given service, and we are totally hooked on a whole lot of free stuff. Ad-driven business models haven’t been declining, they’ve been growing, we seem to be stuck with them for the time being.
There is a certain type of search query I no longer use Google for, but I think that is probably the least common type of query Google handles. Directory searches, weather, sports scores, news, showtimes and shopping make up the bulk of Google searches, and AI chat seems to offer no to negative benefit for these types of searches. Maybe once there is reliable AI agents that will change, but the data seems to show AI is not eating into Google search market share.
Came here to write the same thing. Real search queries have gone down about a half for me. I also largely agree on the reason. Hallucinations happen, but a specific question + answer gets me still faster where I want to be. SEO and StackOverflow karma optimisations have basically made the usefulness of my typical search results go down in the last 5 years or so. Besides, many Google search "features" feel like useless patronisation. For instance, if I search for a thing that is likely on wikipedia, with the english title, Google now shows me the corresponding german wikipedia page first. Its just that I dont need the translated stuff to be able to understand it, and, even more important, english wikipedia articles are usually a little more detailed. Thats why I searched for the english title in the first place damn it, Google. But yeah, I get why you do this. Its just, that I am not the audience you are trying to make happy... So, since there is now an alternative, you basically lost me. I guess the big picture is, they actually dont care about experts leaving. Because, you know, we are a minority.
I use LLMs to answer certain questions, but those are often questions that I wouldn't have bothered using a search engine for in the past, rather I would have asked a colleague or just thought through the question on my own. And when I try to ask an LLM questions that search engines are good at, I'm most often disappointed.
In other words, it's not clear to me LLMs are going to eat into the market share of search engines, rather than just providing a tool with largely orthogonal use cases. But we'll see how the tech develops from here.
I strongly disagree. As a simple example, just this week I was looking for ice breaker questions for a work team event. I started with Google and was wading through a myriad of pages stuffed with ads and noise. I happened to have Claude open for an unrelated work experiment and thought to ask Claude for ice breaker questions. It provided 10 good questions and I selected the first two. It’s just a matter of time until we retrain our brains to first use LLMs before Google and then Google’s usage is going to drop like a rock. LLMs for many use cases is simply better, providing better results with far less noise.
That's pretty much the ideal use case for LLMs, as a tool for creative brainstorming where the question of accuracy is irrelevant. I've used them like that frequently for help in writing/worldbuilding.
But I think it's safe to say that the vast majority of Google searches are looking for factual information of some kind. I can see LLMs as an interface on top of search, but not replacing it.
As a person with reading comprehension skills, I'm still not sure what a use case is for LLMs for me. Everything I try feels like outsourcing a homework problem to a middle schooler.
There is so much more you can pick up by reading a text yourself, even quickly, than comes through when an LLM summarizes a text or answers a complicated question. The way the text is written, the language chosen, the punctuation, the sources chosen, etc.
> While we wouldn’t go so far as to claim AI search is a fad — because it’s not — the data does hint at a potentially hidden bubble that needs to be investigated further.
Okay, so it’s potentially a bubble, but is axiomatically not a fad? Eh?
Corporate Web is SEO Ad hell but it's got nothing on the future that awaits us.
Let me present AI Ad tech 2030
All commercial websites contain meta tags with vector embedding links. Vector embedding services are standard offerings in the big Cloud. Everyone selling something uses them.
All commercial LLM products consume meta data and RAG the global vector embedding index. LLM-SEO is big business as companies fight to game the LLMs with their content.
Generative AI Ad tech inserts Ads into relevant LLM output and is trained on the conversion success of successful Link follows and subsequent purchasing decisions.
This is one set of speculations trying to counter another set of speculations. Here is my speculation to add to the pile. Google tolerates but does not particularly like sending traffic to external websites. To the extent that Google can scrape the useful content from websites (and other sources owned by third party rights owners) and provide it to users without those users leaving a Google property, they will do so. LLM chatbots make it easier to scrape this content while gaining plausible deniability when it comes to copyright infringement by creation of derivative works.
So, this does not have to be driven by user behavior. The user wants plausibly good enough responses to queries. The user typically wants to pay as little as possible in terms of time and money for those responses. To the extent that Google and other Google-likes can satisfy those users without sending traffic to or paying third party site operators and rights owners, they will do so.
Think about it in terms of a speculative video search tool. The user searches Youtube for an answer about a type of tax credit. Above the fold, Google offers to render on the user's machine an LLM generated video answer to their tax question. The video agent will also answer follow up questions without refreshing the page. Below the fold, Google places the Youtube videos. To the extent that users pick the AI result, Google does not need to pay for the third party videos or the traffic costs. At the end of the AI generated video, Google places an inobtrusive paid ad for TumboTax.
> Using the next two charts to illustrate this point
Good on you for using graphs to make your point, but if you don't label your axes or caption your figures I haven't the faintest clue what I'm looking at.
"...conventional wisdom that AI, while good at providing fast answers, is terrible at referring users to where they want to go"
I don't understand why we'd care about referral metrics in this context. Do people use chatbots do "go places"? And the leading chatbots are not designed to refer people off-site, or even to attribute the source material at all, so it should not be surprising that they don't generate a lot of referrals.
I really appreciate that DuckDuckGo has managed to get 2.5% of US search traffic. Frankly that is impressive. It now equals that of yahoo, while obviously a tiny fraction of the overall market it’s likely enough to sustain it
Regarding generative AI for search. Brave and Google already do this to a large extent. I don’t see myself switching.
That said, these AI models are far better when you’re looking for code and recipes, etc so perhaps I’m speaking too soon.
It’s not because of how good AI is (it struggles) but rather how bad search had become. No one from the search provider to the website you land on wants you to have easy access to an answer because there’s no profit motive.
With paid AI services, providing better answers faster IS the profit model. I think something like Claude + Kagi (paid search) would be killer combo.
I’ve actually been kind of nervous about seeing more & more free AI services because that compute is expensive & I’d hate to see the market dynamics that ruined search happen all over again. I want to pay for services with my $ rather than my attention & my data.
I picked up Kagi the day they released their $10/mo plan and I don't see myself dropping it unless their quality degrades or price increases substantially. The search result I want is typically in the first few results, while Google was just ads the first few results, SEO spam, and then MAYBE the result I want near the end of the page.
Really makes it easy to use when it’s all accessed from the address bar.
We're in the "early Google" days of AI when it's actually a nicer, better experience at least for some queries. When they figure out how to monitize it, it will become the same ad-optimized spam cesspool that we have with traditional search and web pages today.
I hope this is the case, but aren’t today’s paid services all heavily subsidized by investment dollars, and brand new, and being used to find traction and validate MVPs? I suspect today’s prices and today’s business model isn’t sustainable.
What hope is there that AI won’t end up being much better and more clever at inserting ads in between captive audiences and the things they’re looking for than Google currently is? I get the feeling advertisers are salivating over all the new possible ways AI can sell things, way more than us consumers are excited about having search without ads…
I wish everyone wanted to pay for services with money rather than attention, but that’s a small minority of people for any given service, and we are totally hooked on a whole lot of free stuff. Ad-driven business models haven’t been declining, they’ve been growing, we seem to be stuck with them for the time being.
In other words, it's not clear to me LLMs are going to eat into the market share of search engines, rather than just providing a tool with largely orthogonal use cases. But we'll see how the tech develops from here.
But I think it's safe to say that the vast majority of Google searches are looking for factual information of some kind. I can see LLMs as an interface on top of search, but not replacing it.
There is so much more you can pick up by reading a text yourself, even quickly, than comes through when an LLM summarizes a text or answers a complicated question. The way the text is written, the language chosen, the punctuation, the sources chosen, etc.
Okay, so it’s potentially a bubble, but is axiomatically not a fad? Eh?
Let me present AI Ad tech 2030
All commercial websites contain meta tags with vector embedding links. Vector embedding services are standard offerings in the big Cloud. Everyone selling something uses them.
All commercial LLM products consume meta data and RAG the global vector embedding index. LLM-SEO is big business as companies fight to game the LLMs with their content.
Generative AI Ad tech inserts Ads into relevant LLM output and is trained on the conversion success of successful Link follows and subsequent purchasing decisions.
May I present Alpha-Ad.
It's on the way folks.
So, this does not have to be driven by user behavior. The user wants plausibly good enough responses to queries. The user typically wants to pay as little as possible in terms of time and money for those responses. To the extent that Google and other Google-likes can satisfy those users without sending traffic to or paying third party site operators and rights owners, they will do so.
Think about it in terms of a speculative video search tool. The user searches Youtube for an answer about a type of tax credit. Above the fold, Google offers to render on the user's machine an LLM generated video answer to their tax question. The video agent will also answer follow up questions without refreshing the page. Below the fold, Google places the Youtube videos. To the extent that users pick the AI result, Google does not need to pay for the third party videos or the traffic costs. At the end of the AI generated video, Google places an inobtrusive paid ad for TumboTax.
Good on you for using graphs to make your point, but if you don't label your axes or caption your figures I haven't the faintest clue what I'm looking at.
I’m adept at using quotes and “+” and “-“ for keywords on queries. For about 20 years, if something existed, I could find it easily.
What amazes me now is not so much the sponsored/SEO crap listings at the top — it is there is nothing else as one works their way to the “bottom”.
Too often, things that should be easy to find just don’t show up. I feel like they have entirely given up on indexing.
It’s a sad state of affairs when Bing does better…
If true, they get to look like visionaries. If not, nobody will remember anyway. It’s a no-brainer.
I don't understand why we'd care about referral metrics in this context. Do people use chatbots do "go places"? And the leading chatbots are not designed to refer people off-site, or even to attribute the source material at all, so it should not be surprising that they don't generate a lot of referrals.
Regarding generative AI for search. Brave and Google already do this to a large extent. I don’t see myself switching.
That said, these AI models are far better when you’re looking for code and recipes, etc so perhaps I’m speaking too soon.