How AI Search Engines Are Replacing Traditional SERP Ad Inventory

Google built a two-decade business on a simple trade: someone types a question, Google hands back ten blue links plus a few labeled ads, and advertisers pay for a spot near the top. That trade is coming apart. People are still searching, but a growing share of them get their answer before they ever reach the link layer. AI Overviews now show up in 57% of Google results. More than half the time, Google's own product shrinks or erases the exact click path advertisers spent twenty years building campaigns around.

Gartner, usually one of the more conservative analyst shops in this space, expects a 25% drop in traditional search engine traffic by 2026. A November 2025 study found something sharper: AI Overviews cut organic click-through rate meaningfully on informational queries. I've seen a lot of "channel decline" charts over the years, and most of them show erosion, a slow bleed. This looks different: inventory disappearing in specific categories while everyone watches it happen in real time.

Here's the wrinkle. Clickstream data from SparkToro and Datos shows overall search volume held flat between April 2024 and June 2025, and it actually rose some months. Meanwhile the share of Americans using AI tools ten-plus times a month jumped from 3% in January 2023 to 21% by June 2025. Seven times over, in under two and a half years. AI use is climbing fast and search volume isn't falling; the only way both are true is if AI is soaking up queries that never would have existed as searches in the first place, not just stealing old ones.

None of this should comfort anyone selling ads, and I don't think it does, once you sit with it. The queries migrating to AI interfaces are the ones that used to pay the bills: product research, side-by-side comparisons, "what should I buy." McKinsey's AI Discovery Survey from August 2025, nearly 1,930 respondents, found 44% of AI-search users now call it their preferred source for this kind of research, against 31% who still say traditional search. That preference gap already flipped, and it flipped fast.

Total query volume looks fine on a chart, and that reading is technically accurate. But the harder truth is in what's leaving the building: the intent that converts, the exact queries ad dollars were built to catch, moving into formats where nobody's built the auction yet.

The economic stakes that make this an urgent decision, not a planning exercise

McKinsey puts $750 billion in U.S. revenue on track to move through AI-powered search by 2028. That's the size of the pot getting redealt, and it's worth sitting with that number for a second before moving on. Brands that do nothing could see traditional search traffic fall 20 to 50 percent, per McKinsey's own range. That's a wide spread, wide enough to tell you this hits categories unevenly, which is exactly the kind of detail a tidy forecast tends to smooth over.

Compare that to what's already on the table today. Google Search pulled in $66.89 billion in ad revenue in Q1 2025 alone. One quarter. That's the number now exposed to structural rewiring, and Google knows it better than anyone; its own projection has U.S. AI-driven search ad spend climbing from just over $1 billion in 2025 to nearly $26 billion by 2029. A company doesn't build that kind of internal forecast unless it's racing to convert its own market before someone else does it first.

Sit those two numbers side by side: $66.89 billion in a single quarter of legacy revenue, against an AI ad market still measured in single-digit billions a year. That gap is where the tension lives, and whoever moves into it first stands to compound the advantage in a way that's hard to claw back later. This is a timing problem, plainly stated: the window between early mover and late adopter is closing, and it's closing faster than a normal annual planning cycle can react to.

Platform by platform: how the major AI search surfaces are building ad inventory right now

Start with Google, since it's building on infrastructure that already exists. AI Overviews hit 2 billion monthly users as of July 2025, spanning more than 200 countries and 40-plus languages. AI Mode passed 100 million monthly active users in the U.S. and India, and daily queries per user in the U.S. have doubled since launch. In May 2025, Google widened Search and Shopping ads inside AI Overviews and began testing ads inside AI Mode for desktop users in the U.S. Same auction, new real estate, dropped into an attention environment that behaves nothing like the old results page.

Seer Interactive tracked a large sample of organic and paid impressions from June 2024 through September 2025. Paid CTR without an AI Overview present fell from 19.1% to 13.04%, a 32% drop. But when a brand got cited inside the AI Overview itself, paid CTR ran 91% higher than when it wasn't cited at all. On Google now, being inside the answer beats being next to it. That's the line worth remembering out of this whole section.

OpenAI is playing a different game from a different starting line entirely. ChatGPT counted 400 million weekly users in early 2025, and at Cannes Lions 2026, OpenAI's Head of Global Ads Solutions put commercial intent at roughly 20% of ChatGPT queries. Criteo became the first ad tech partner in OpenAI's advertising pilot in March 2026; by June 2026, more than 2,000 brands were buying through Criteo across the U.S., Australia, Canada, New Zealand, the UK, Japan, and South Korea. The pilot crossed $100 million in annualized revenue within six weeks. Axios reported OpenAI projecting $2.5 billion in 2026 ad revenue. The inventory's real, but access still runs through one gate, and that matters more than the growth numbers suggest on their own.

Microsoft has the most mature runway for advertisers already sitting inside its ecosystem, and this is the platform I'd point a skeptical CFO to first. Copilot shows 73% higher click-through rates and 16% stronger conversion than traditional search, with customer journeys running 33% shorter, per Microsoft Advertising's own August 2025 numbers. ROAS jumps thirteenfold when someone uses Copilot before running a traditional search. That's an odd number to sit with, honestly, and the best explanation I've heard is that Copilot works upstream of the purchase decision, more like a filter than a search engine.

Perplexity took the narrowest path of the group: sponsored follow-up questions, where a brand pays to surface first among the AI's suggested next steps. As of October 2025 that program stayed closed to new advertisers, live for early partners only. Perplexity burned through $65 million in 2024 against just $34 million in revenue. For them, monetization is survival, full stop. Stack that against Google's $66.89 billion in a single quarter and the scale gap doesn't need further comment.

Underneath all four sits a layer nobody's bothered to name yet: thousands of AI-native apps, vertical assistants, and AI-powered site search tools generating real, high-intent conversational traffic with no ad infrastructure built for any of it. DSPs built specifically for conversational AI, like Thrad, can read a prompt for intent signals in real time and slot in a contextual ad within milliseconds, reaching that long tail at a scale keyword systems were never designed for. This monetization layer was built for the format from the ground up, the way a legacy ad stack bolted onto a chat window rarely is when it's just there to fill a Q3 revenue line.

Why the mechanics of conversational ad inventory are fundamentally different from keyword auctions

In a keyword auction, advertisers bid on a word or phrase matched against a query string. In an LLM, the unit is intent, spelled out in full sentences, sitting inside context. Someone typing "I'm renovating my kitchen and trying to decide between a gas and induction range, what do I actually need to know" hands over more signal in one sentence than any keyword ever could. Category, decision stage, the specific gap in their knowledge, all sitting right there for whoever's reading it.

Researchers are trying to formalize this, and it's slower going than the marketing decks suggest. Work published in ACM SIGecom Exchanges in March 2025 looked at auction designs meant to guarantee higher bidders get more prominent placement inside LLM-generated answers, plus retrieval-augmented generation auctions that weigh relevance against bid size when deciding what makes it into a response. The same research flags the hard part plainly: a banner ad looks identical every time it loads, but an LLM can phrase or place an ad differently on every single query. Click-through rate gets much harder to learn and stabilize at scale when the thing you're measuring never holds still long enough to measure twice the same way.

Targeting shifts with it, necessarily. The emerging LLM demand-side platforms run on conversational intent, topic, and publisher category, with zero dependency on third-party cookies, which is either a relief or a headache depending on which side of the privacy debate you've been standing on. The formats split roughly three ways: contextual recommendations priced by click or impression, in-chat sponsored placements priced by click or action, and native display priced by impression or click. All of them live inside the flow of the answer rather than parked beside it.

Here's the deeper shift underneath all of it, and it's the part I keep coming back to: since the LLM already answers the question, there's no click path left for a traditional pull ad to sit in. Native and programmatic formats built for this layer are filling the space that pull search used to own, because that space is the only space left.

A cost pressure is pushing this along, too, from the supply side, and it's easy to forget this isn't optional for the platforms themselves. Running an LLM at scale is expensive. ChatGPT's compute bill reportedly runs around $700,000 a day, and subscriptions don't cover the free tier. Ad revenue is closer to a structural requirement for these companies than an experiment, the thing that keeps the lights on for everyone who isn't paying $20 a month.

What the early performance data actually shows, and what it doesn't yet prove

Criteo's aggregated data found users referred from LLM platforms like ChatGPT converting at about 1.5 times the rate of other referral channels. Criteo continued to report strong AI-referred conversion performance across retail categories as the pilot expanded. At Cannes Lions 2026, Criteo put ChatGPT ad performance at 2 to 3 times the click-through rate and a 4x spend lift across more than 2,000 active brand campaigns. Those numbers line up with the conversion data instead of contradicting it, which is at least internally consistent.

Microsoft's Copilot figures point somewhere slightly different, and worth pausing on. That 73% CTR lift and 16% conversion lift, paired with journeys 33% shorter than traditional search, doesn't read like a top-of-funnel bump to me. It reads like the entire funnel getting compressed and sped up at the same time, which is a different phenomenon than "more clicks."

Then there's the number that should keep everyone honest, and I think it's the most important one in this whole piece. Separate analysis of a broader brand set found incremental gains that were positive but modest, with ad spend essentially flat. Positive, real, but modest. That suggests the channel is additive for most advertisers right now rather than transformative, and it's a genuinely different picture than the Criteo numbers sitting two paragraphs above. Both are true at once, and neither cancels the other out.

None of this is settled, and I'd be suspicious of anyone who tells you it is. Sample sizes are still small, results vary enormously by category, and most of the campaign data cited here runs through a single tech partner inside a pilot program. The CTR and conversion figures will move as inventory scales and more advertisers bid for the same placements, probably compressing some of these early advantages. Direction is consistently positive across every data set here. Whoever treats today's numbers as a permanent baseline, in either direction, is going to be proven wrong within a year, maybe less.

What this means for how advertisers should allocate attention and budget right now

The vacuum left by the old results page is real, but it's not evenly spread across platforms, and treating it as one undifferentiated "AI search" bucket is a mistake I'd steer any budget owner away from. Google AI Mode and AI Overviews sit at the highest scale today, 2 billion monthly users, running through Google Ads infrastructure advertisers already know how to use. That's the lowest-friction door available. Walk through it first.

The 91% paid CTR lift Seer Interactive measured for brands cited inside AI Overviews changes the calculus on organic strategy too, and this is the part a lot of paid-search teams still haven't internalized. Getting cited inside the AI response directly lifts the paid ad sitting right next to it, which means SEO and paid media just got a lot harder to plan separately. ChatGPT's roughly 20% commercial-intent share across 400 million weekly users is a large enough pool of high-intent moments to justify testing spend there now, pilot stage or not.

Copilot's funnel data, the shorter journeys and the thirteenfold ROAS lift when it runs ahead of a traditional search, argues for treating Copilot touchpoints as upper-funnel qualifiers rather than last-click conversion assets. Don't force it into a last-click model just because that's the model everyone's comfortable with.

The biggest opening might not sit inside any of the named platforms at all, and this is where I'd put speculative testing dollars if I had to choose. It's the layer underneath: thousands of vertical AI tools and chat interfaces where intent runs high, competition for placement is close to nothing, and no legacy auction has set a price floor yet.

Creative needs to change to match the format, not just the channel. When the targeting unit is a full sentence of stated intent instead of a keyword, ads that interrupt the answer perform worse than ads that flow with it. Native and contextual formats win here for a plain reason: they meet the user inside the need they already spelled out, instead of jumping in front of it uninvited.

Framing this as AI against search, two budgets fighting over the same dollar, misses what the data's actually showing. Traditional search volume holding steady increasingly just means it's the leftover after high-intent queries move elsewhere, and spend should follow that reality rather than the flat top-line chart. The window for cheap entry into this inventory is open right now. It won't stay open long; Google's own forecast of near $26 billion in AI-driven search ad spend by 2029 says plainly enough how fast the price of these placements is about to climb.