How AI Search Engines Are Replacing Traditional SERP Ad Inventory

For two decades, the trade behind search advertising has stayed the same: a person types a question, scans a page of links, and clicks one that happens to be a paid placement. Every dollar in that market rests on that sequence actually playing out. AI Overviews are breaking the sequence on the exact queries that paid for it. When a large language model writes the answer directly into the results page, there's nothing left for a paid link to interrupt. The user got what they came for before reaching the layer where ads live, so the ad has no surface left to occupy. This isn't a marginal drag on click-through rates. Seer Interactive's research, summarized in a Search Influence post, found that paid click-through rate falls on queries where an AI Overview appeared, and by September 2025 the gap against queries with no Overview had reached seven points. The queries losing that click path are product research, comparison shopping, and purchase-intent searches, the exact inventory that commanded the highest bids in the old auction. What's breaking is narrow and specific: a particular ad format, built for a particular click sequence, is losing the sequence that made it valuable.

Why surface-level revenue stability masks the erosion happening underneath

Search advertising revenue still looks fine at the aggregate level, but that stability is what makes the underlying shift easy to miss. Google's own Q2 2026 SEC filing shows Search and other revenues rising year over year, but Google Network revenues fell, and a decline in AdSense drove most of that. That split matters because AdSense is the mechanism by which independent publishers get paid for the SERP-adjacent inventory that built the open web's ad economy. So when Google's top-line search number holds steady while its network revenue drops, independent publishers are absorbing a loss that Google's own properties don't show.

Query volume tells a similar story. SparkToro and Datos clickstream data show total search volume holding flat or rising between April 2024 and June 2025. Read quickly, that looks like evidence nothing has changed. The same data shows AI absorbing precisely the query types that used to convert into revenue, even as the overall volume chart stays level. Ridge Marketing's analysis of data from Deloitte, Gartner, AP-NORC, and Pew confirms that traditional search still handles a large majority of global search traffic today. The shift hasn't shown up in the aggregate numbers yet, which makes it easiest to dismiss as noise precisely now.

Google isn't waiting for the gap to close on its own. The company's internal forecast has U.S. AI-driven search ad spend climbing from a modest base in 2025 to nearly $26 billion by 2029. That number is Google racing to rebuild, on its own properties, the revenue stream it is simultaneously eroding from third-party inventory. A company does not forecast tens of billions of dollars in new ad spend inside a product category unless it already understands which older category that spend is replacing.

Query behavior shifts and intent types moved into conversational interfaces

None of the demand disappearing from SERP inventory has actually vanished. You get that answer inside conversational sessions instead, in a format that the keyword-based infrastructure behind most ad platforms was never built to read. Ridge Marketing describes a split-path pattern taking shape: AI tools now handle exploratory research and quick factual answers, while traditional search still carries the weight of decision-making and purchases. That split matters because the commercial queries moving to AI tools are the exploratory-to-comparative stage of the buying journey, the research and comparison work that happens just before someone is ready to buy.

The preference data backs this up. McKinsey's AI Discovery Survey found that 44% of AI-powered search users now name AI tools as their primary and preferred source for product research, a share that already outpaces those who still default to traditional search. That's a flip in stated preference that happened inside about a year.

What's changed is not just where people search, but how the search itself unfolds. A multi-query path once ran across several separate searches, moving from vague concern to product awareness to comparison to purchase readiness, but now it plays out inside one conversational thread. A user might never type the phrase "project management software" into a search box. Instead, that same person might describe, across one extended conversation, the trouble their team has coordinating deadlines and collaborating remotely. Every detail in that conversation is a high-intent signal pointing toward a software purchase, and none of it is visible to a keyword-based ad platform built to match against discrete search terms. The intent hasn't gone missing. It has moved into a format richer than any keyword ever captured, sitting in a layer that traditional demand-side platforms cannot reach.

Scale backs up why this matters. Ridge Marketing reports that ChatGPT's weekly active user base grew sharply between February 2025 and January 2026, and the platform now processes billions of queries a day. That makes the conversational layer a real destination for attention. Even on Google's own turf, the pattern holds: LLM Billboard reports that AI Overviews reached a global monthly user base spanning countries and territories worldwide as of July 2025. Attention is moving into conversation whether that conversation happens on a dedicated chatbot or inside Google's own results page.

Ad inventory reconstructed inside conversational interfaces

Rebuilding ad inventory for a conversational interface is not a matter of moving keyword auctions onto a new screen. The targeting input itself is different in kind: ads are matched to the live meaning of a conversation as it unfolds, not to a static search term typed once and gone. That difference in input produces a difference in measurement, and the mechanics now emerging across platforms show both halves of that shift.

There's no keyword bidding in this model. The primary lever advertisers control is a freeform, natural-language context description, capped at 280 characters and written at the ad group level, and it describes the kinds of conversations where a given ad belongs. That description gets matched against a live conversation in progress, not against a search query typed into a box. Academic work backs up how unsettled the auction mechanics still are. Research published at NeurIPS 2024, and discussed further in ACM SIGecom Exchanges, examined auction designs built specifically for large language model ad environments, including retrieval-augmented generation auctions that weigh relevance against bid size, along with mechanisms meant to guarantee that higher bidders get more prominent placement inside an LLM-generated answer. That research also names the hard measurement problem at the center of this shift: a banner ad renders identically every time it loads, but a large language model can phrase or place an ad differently on every single query it answers, which makes click-through rate far harder to learn and stabilize at any scale.

Platforms are solving this differently depending on what they already own. Google is extending its existing auction infrastructure into the new surface: Search Influence reports that ads inside AI Overviews and AI Mode draw on Performance Max, Shopping, broad match Search campaigns, and AI Max for Search campaigns already running today. The auction machinery is familiar, but it now has to perform inside an attention environment that behaves nothing like the old results page.

Microsoft has built something closer to a native conversational product. Its "ad voice" feature creates a bridge between the AI's answer and the sponsored message, matching tone and context to the user's ongoing conversation, and its contextual conversation ads can weigh an entire session rather than just the most recent query, tracking a user as they refine questions and move deeper into a decision.

OpenAI's mechanics inside ChatGPT carry both real reach and a real constraint. Ads target free and Go subscribers aged 18 and up in the US, free users can opt out in exchange for stricter daily conversation limits, and advertisers can choose a Reach objective billed on CPM, a Clicks objective billed on CPC, or a Conversions objective billed on oCPC. The constraint that matters most for planning: anyone on a paid ChatGPT plan, Plus, Pro, Business, Enterprise, or Edu, including seats an employer is paying for, will never see an ad. Ads only run on the free and Go tiers. For a media buyer trying to reach business decision-makers through ChatGPT, that rules out the exact population most likely to be on a paid seat, and it has to be treated as a hard limit when any campaign plan assumes ChatGPT reach.

A platform able to read conversational context across multiple AI surfaces, rather than sitting inside one surface's walled inventory, can match intent signals that no keyword system was ever built to see, and can offer reach that no single-surface network can match on its own.

Early performance data from conversational ad placements

The early numbers coming out of conversational ad placements look genuinely strong, and they also come with a clear limit on how much weight they can bear. Microsoft Copilot's self-reported figures show substantially higher click-through and conversion rates than traditional search ads produce, but those numbers are early-stage and platform-reported, and no outside party has verified them. That distinction matters because a platform has every incentive to report its own product in the best possible light, and the industry has no independent benchmark yet to check those numbers against.

A fair counterpoint complicates the picture further, in a useful way. Measured's analysis of a large cohort of brands found that Google's AI Overviews did not weaken paid search performance at all. Incremental revenue and incremental orders actually rose across that cohort through the first quarter of 2026. That finding doesn't contradict the structural argument running through this piece. Short-term paid search resilience and a long-term migration of intent into conversational interfaces can both be true simultaneously: a brand's paid search campaigns can keep performing well in the near term even as the underlying query mix shifts permanently toward conversational formats over a longer horizon.

Measurement, not performance, is the problem: advertisers cannot yet see what users asked or attribute results to it. Zéa Hirsh, performance media supervisor at Collective Measures, told Adweek that advertisers lack search- or prompt-level reporting, so they cannot see what users were actually asking when an ad got served inside a conversation. Conversational intent carries far more signal than a search term ever did, but the infrastructure needed to read that signal and attribute results to it does not exist at scale yet. The new inventory shows genuinely strong early performance data. It does not yet prove conversational ads outperform search ads at scale, because the tools that would let anyone prove that haven't caught up with the medium.

The intent signal conversational prompts produce

A prompt inside an AI conversation carries more than a search box ever could. It shows not just what someone wants, but where they stand in a decision, what limits they're working within, and what objections they're still turning over, a depth of signal that no keyword or demographic profile has ever delivered. A keyword has always worked as a proxy, a rough stand-in for whatever a person actually wants. A prompt states intent directly, in plain language, carrying constraints, preferences, and the user's stage in their decision all at once.

Watch how that signal builds across a single conversational thread. A user might open with a general concern, move into asking for comparisons between options, name a specific budget, describe a workflow problem in detail, and then ask for a recommendation, with every one of those steps visible in sequence to whatever system is reading the conversation. If a search platform matches against a single typed phrase, it catches none of that progression. A system reading the full conversation catches all of it, in order, as it happens.

That gap is the real measure of what's at stake in this shift. The click path that funded two decades of search advertising is narrowing on the exact queries that made it valuable, and the revenue sitting on top of that click path is already splitting between the companies absorbing displaced demand and the publishers losing it. None of that demand has disappeared. It has moved into a format carrying more information about intent than a keyword auction was ever built to use, and the platforms, advertisers, and measurement systems that learn to read it first will be the ones that end up owning it.