Can anyone explain how search revenue is still _growing_ in the age of LLMs?
My wife is an investor, and one of her portfolio areas is pharmaceuticals. A couple of portfolio companies have reported that it's becoming basically impossible to make any money off of a new product, because you need to advertise it to reach the customer, and Google will skim all the excess producer surplus off as you compete with other startups serving the same market.
It's basically the perfect business model. They own the path to the consumer, which means they own the economy.
I'd also recently hired someone out of Google Search, and they said that the only queries that "legacy" (non-AI-mode) search cares about are commercial-seeking queries, and the only metric they optimize for is ad conversions on those. It literally is thousands of people whose only job is to get you to click more ads.
You speak as if this is given. Sure, LLMs are gonna get effective but frontier ones are always going to be hosted.
Also, on-device LLMs are gonna have a 'cutoff' for training data. You cant ask a gpt-oss4 about "Who won Arsenal x Athetico Madrid game". It has to go to the internet, and do the 'search'. You certainly cant ask the local model "What was Google's Q1 2026 earnings".
The self-hosted / on-device LLMs are going to do a lot, but not all and the moment 'search' is involved, people will reach for Google.
--
Lots of commercial queries need to be 'fresh' ("Build me an itinerary for Paris"). While a self-hosted LLM can do it, you might not want to trust it, because its info might be stale.
I encourage you to go beyond group think (HN is guilty of that) and really evaluate your position is actually valid or not.
Non-commercial queries are basically a loss-leader for Google. They don't care about them, except that they keep people in the habit of Googling things whenever they need information, which is critical for the whole business model to work. That's why ChatGPT was such a threat: for a while it looked like people might get out of the habit of Googling things and instead just ask ChatGPT, which would've been disastrous. But they seem to have headed it off with Gemini and AI mode. I was just researching something new today, and while I felt like Claude gave slightly better responses, AI mode gave more convenient responses, with direct links and browser integration.
This is why commoditized LLMs aren't a threat. If it's commoditized, people just go to the one which already fits their habits, which is AI mode. And they don't charge anything anyway, so competing on price doesn't work.
New computing devices that don't default to Google Search are a threat. But that's why Google funded Chrome, and Android, and paid Apple billions to be the default search provider in Safari and iOS, and paid Mozilla billions to be the default search provider in Firefox. As long as the results for non-commercial queries aren't actively bad, it'll likely be hard to convince people to switch away from them, particularly given existing habits.
If I had to list the biggest threats to Google's stock price, I'd put them as:
1. A global macro downturn. Google's stock has been pretty macro-sensitive since COVID, because as the tollkeeper to the economy, their revenues directly depend on how many economic transactions there are. If say the Straight of Hormuz crisis results in stagflation, even if it's stagflation in Asia and Australia and Europe rather than the U.S., Google's going to feel it. You even saw that with the ~20% swoon in the early days of the crisis.
2. Dead Internet Theory. If people just give up on the Internet because it's boring or not good for their kids or filled with bots or just not useful, and return to their local communities, this is also the end of Google. There've been some moves in this direction (eg. we're raising our kids to socialize in person with neighbors instead of going online, and many teenagers today think the Internet is decidedly Not Fun), but it also needs to be much more widespread, and get around the fact that many niche products are only available online.
3. Internal decay. I'm an ex-Googler, still have a couple of (increasingly disenchanted) friends there, though many of my friends have retired or left in the last few years. It is a shitshow inside, with a mess of perverse incentives, sometimes incompetent executives, and employees who don't care and are just phoning it in while the stock price goes up. I'd still get questions on code I wrote in 2010, from teams in Bangalore who were just taking over the legacy search stack, who would ask me things because I was literally the only one left at the company from when the code was written and I'd be like "How the hell should I know? I left this project 15 years ago, left the company, haven't worked on Search since 2014, had several other positions, came back, several billion lines of code has been written on top of it since then, still don't work on Search nor do I actually write code anymore, no I can't answer whether this code is important or whether anything will break if you get rid of it." Particularly now that ~75% of the code at Google is written by AI, there's a decent chance that somebody or some AI will introduce a change that breaks the golden goose, it won't be caught until several million more lines of AI-generated code have been checked in on top of it, and that'll be the end of the fabulous machine known as Google. Reportedly this is what happened to Twitter DMs, they used AI to check in some code that broke the feature, nobody knew how to get it working again, and so they just unlaunched it.
It'd have to be a major black swan, though, because there are many layers of canaries and metrics checks and rollbacks and backups in place. I think the issue now is that an increasing number of Googlers don't realize why these are important, but Search was built for robustness.
You are mangling a well defined term of producer surplus that is widely accepted in economics with your own.
Normally, in an Econ 101 supply/demand model, you have an upward-sloping supply curve indicating how many units of some commodity a producer is willing to sell at a given hypothetical price, and you have a downward-sloping demand curve indicating how many units of some commodity a consumer is willing to buy at a given hypothetical price, and the point at which they meet is called the "market clearing price" where supply and demand are in balance. The area above the supply curve and below the market-clearing price is called the "producer surplus" [1], the amount of extra money the producer gets from the gains of trade. Very roughly, you can think of it as operating profit, though it's not quite the same thing because operating profit is a real tangible dollar amount while the supply curve is largely a hypothetical.
The way Google works is that it comes in for each individual transaction and effectively tells the advertiser "So you have a product or service that you'd like to sell? How much is it worth to you?" It runs a second-price VCG auction [2][3] to determine which of the available ads are shown to the user. Why second-price? Because an ordinary first-price auction incentivizes users to bid less than their true willingness to pay, because they know that if they win the bid, they will have overpaid. Same issue as buyer's remorse in real estate transactions: you know that if you won the bid, it was because nobody else thought the house was worth as much as you did. With a VCG auction you're only paying the marginal harm to the next bidder of winning the auction, so you have an incentive to bid your true valuation. And this is why I used the term "producer surplus" in the original post: it's to capture how Google effectively elicits from each advertiser the maximum amount that the transaction is worth to them, which if they're bidding rationally is the difference between their reservation price and the price they'll receive from the transaction, i.e. the producer surplus.
This also demonstrates where I've been playing fast and loose with the terminology. The price a VCG auction participant pays is not actually their bid, it's the marginal harm caused to other participants (basically the sum of all bids except the advertiser, minus the sum of bids of the other winners). But in a fairly competitive market, one with multiple producers who all face roughly the same cost structure, you'd expect that this quantity converges to the advertiser's actual bid, leading to a condition where Google has actually captured the entire producer surplus and maximized its revenue.
[1] https://www.intelligenteconomist.com/producer-surplus/
[2] https://en.wikipedia.org/wiki/Vickrey%E2%80%93Clarke%E2%80%9...
[3] Technically, it's a modified VCG auction, because the actual ranking score is a machine-learned function that also includes factors like "What's the likelihood the user is going to click on this ad?", "Is this ad spam or malware or illegal?", and "Is there brand damage to us from highlighting this ad?" But from an economics perspective, the only important part is that it's a second-price auction.
I keep thinking how often LLMs are used to “solve” problems we created.
“Meeting starts at 3” gets fed into an LLM to turn it into a 3 paragraph email, only to be summarized by a different LLM on the other email client end as “Meeting starts at 3”. What a waste.
You get completely different results if you search for "what is a sofa bed" instead of "sofa bed".
I say it's annoying, but it comes from real user behavior.
Maybe in the future something comparable will be invented and protected. No harm in dreaming a little I guess.
Now think about today. You search on Google and they run an auction to charge the vendor for getting their products discovered. You go to their website, but their website needs to be protected by CloudFlare from all the people who would take it offline. You buy the product with your Visa, which takes their ~1.6% merchant fee. They sell it with Stripe, which takes their 2.7% fee. You pay shipping and handling for UPS or FedEx to deliver it. Local, state, and federal governments all take tax out of this. Increasingly we're getting national-level shakedowns like Iran charging tolls for the Straight of Hormuz now.
Note that it was similar in olden times whenever one party could control the flow of goods between producer and customer. Sea merchants made huge profits in the 1500-1600s. Standard Oil got to be huge with its control of the distribution infrastructure for oil.
European governments have actually woken up to this with passage of laws like the DMA, but the American government has taken the other tack and decided that if life is a massive shakedown it's going to get in on the extortion business.
"sofa beds": Popular Products section (5x2 grid of chips) Reddit link: "Are sofa beds actually practical" Wayfair Reddit link: recommendations Wirecutter "Discussions and forums" section (Reddit and random crap) Ikea "In stores nearby" (5x1 of chips) "Deals on sleeper sofas" (5x2 of chips) "Things to know" section "Brands" (one row of chips) Furniture store Furniture store Furniture store "More products" (5x2 chips) Furniture store "People also search for"
None of these were marked as sponsored. I assume, but don't know, that the Popular Products chips are sponsored somehow.
"hard drive": Popular Products (5x2) Wikipedia Best Buy People Also Ask section Reddit: request for what brand is good Brand Picks for You (5x1) Things to Know section Amazon link PCMag reviews Reddit: "What exactly is HDD?" Brands section NYT review article What People Are Saying section (a weird mix of stuff) Wikipedia More Products section Store link People Also Search For section
None marked as sponsored.
Not that crazy for a vague, commercially-oriented search. Certainly not 100% ads.
1. Google has monopolized the current AdTech stack from all sides. Not only does it dominate the place where most lucrative ads are displayed (search), it dominates the tools used by publishers to sell ads, the tools used by advertisers to buy ads, and the exchange that matches them. And as the AdTech antitrust lawsuit found, they've been pulling some shenanigans in the auctions to pad their margins by eating into from the the publishers' margins.
2. The search UI is simultaneously limited real estate but also easy to stuff with ads, which is why it has been so valuable. (Google also monetizes a chunk of the display ad space across websites, but that is very clearly dead as plummeting traffic numbers show.)
3. The entire serving stack has been crazy hyper-optmized for this model.
However, with agents this whole stack gets bypassed. The conversational UX is very different. A user will not do N searches for a product with M ads on each SERP, resulting in NxM impressions and ~N clicks as they explore the product page and reviews to come to a conclusion.
They will instead have a conversation with an agent over N rounds of back and forth and only the final round will matter somewhat, ideally resulting in a final, very valuable click.
So now the ad impression volume (which is apparently more important than CPC these days?! https://www.marketingdive.com/news/google-parent-shifts-basi...) has gone down by a factor of NxM. Oh, eventually they will start displaying ads at each turn of the conversation, of course. But due to the user doing all the product exploration in the chat itself, it seems highly unlikely they'll click on those, and so those may not be very valuable at all.
Now layer on top of that the fact that LLMs are probably more expensive to serve than search or ads. So there's pressure on the top- as well as bottom-lines.
Furthermore, this could shift ad-selling and buying patterns, so all those shenanigans padding their margins may also vanish. (This will be a problem for Amazon too BTW.)
It is unclear that the value increase in that final click will compensate for that NxM display and margin-padding loss. Concerningly, even if N is 1, M these days is on the order of 10+.
Google will still be a ridiculously lucrative company, but its search + ad revenue could be a fraction of what it is today. That cloud revenue better pick up fast.
(googler, opinions are my own. I know nothing about this outside of external info)
I open Google Search. If I want an LLM, I click “AI Mode.”
I only bother to open Gemini or ChatGPT or Claude if it’s a “big research thing” which is almost never.
You just say "I want to find some headphones" and it makes you some recommendations. Or it helps you nail down what you're looking for first and then gives you options at various price points. I've found this useful when shopping for cars, computers, tourist activities, and much more.
Search engines do exist, and search engines do allow tool calls. Try find an AI chat provider that doesn't connect web search to the agent.
As far as the premise goes, people on this very site were saying it. The very parent is surprised search is growing.
Absent additional information, no one else can identify the false assumption underlies a strong belief that this should not be case. But something is flying in the face of the facts, and has been for a while. So, yeah, might want to take a look at those priors.
AI overview (organic for now?)
3 organic results (below the fold)
2 paid search results
Image results
1 organic result
If I don't run paid ads I don't get leads I rank organically for a lot of the keyphrases I'm bidding on (top 3 results)...organic 0-2 calls a week, paid ads 4-6 calls a week (niche business).
Another way to think about this is that LLM based search actually grows the entire pie for search and so multiple players can all be growing at the same time.
The real losers are publishers, blogs, forums etc. Instead of traffic going to them that traffic is being turned into more search queries and more LLM responses.
Personally I use AI chatbots to help me make every purchase decision I have to think about.
I just tried asking ChatGPT where to buy a backpack I'm looking for and it just... did a search. It would have been considerably faster to just do the search myself instead of wade through the slop about how "This backpack is hard to find in stock, you'll need to buy it directly from the brand yada yada yada".
Brand awareness is all that matters. Apple isnt counting on me clicking on the ad to buy the macbook
Right now lots of top tier online resources just become not usable due to amount of ads. It used to be a sign of a 3rd tier website.