Run a commercial query on Google right now—something like “best CRM for mid-market SaaS companies”—and watch what happens above the fold. If you’re in a cohort eligible for Google’s Search Generative Experience (SGE)—the AI-generated answer snapshots that appear above traditional results on an expanding set of queries—the first thing you see is a synthesized paragraph answering the question directly, pulled from three or four sources with citation links. Below that, you might see a carousel of product listings, then organic results, then—maybe—traditional text ads. The ads that used to occupy the top of the SERP are now competing for attention against a system that answers the query before the user ever needs to click.
This is not a UX problem. It’s an economic one. The ad auction that Google runs on every search query is, at its core, a market for human attention. Advertisers bid on the expected value of a user’s attention reaching their creative, clicking through, and converting. When an AI-generated answer consumes that attention before it reaches the ad slots, the expected value of those slots changes. And when expected value changes, rational advertisers should change their bids. The mechanism is quiet, but it’s structural: SGE doesn’t remove ad inventory, it degrades the attention budget that makes that inventory valuable.
The Attention Budget Model
Think of a SERP as a vertical stream of attention with a finite budget. Users arrive with some probability of scrolling to any given position, and that probability decays as they move down the page. Before SGE, the decay function looked roughly like this for a commercial query:
Position 1 (Ad): ██████████████████ ~85% reach Position 2 (Ad): ████████████████ ~70% reach Position 3 (Ad): ██████████████ ~55% reach Position 1 (Organic): ██████████████ ~50% reach Position 2 (Organic): ████████████ ~35% reach Position 3 (Organic): ██████████ ~22% reach
These numbers are approximate, drawn from click-through rate studies and eye-tracking research, but the shape is well-established: exponential decay with a long tail. Advertisers in Position 1 are paying for 85% of users seeing their ad. That’s why the top slot commands a premium CPC.
Now inject an SGE snapshot at the top. The snapshot typically occupies 600-900 vertical pixels—roughly the equivalent of two full ad positions plus an organic result. It demands cognitive engagement: reading a synthesized paragraph, scanning source links, possibly expanding a follow-up question. The attention cost isn’t just the screen real estate; it’s the cognitive switching cost of processing an answer before deciding whether to scroll further.
The revised reach curve looks something like this:
SGE Snapshot: ██████████████████ ~90% reach (top of page) Position 1 (Ad): ██████████ ~45% reach Position 2 (Ad): ███████ ~28% reach Position 3 (Ad): █████ ~16% reach Position 1 (Organic): █████ ~14% reach Position 2 (Organic): ██ ~8% reach
The top ad position lost roughly 40 percentage points of reach. The second ad position lost nearly half its audience. And organic results below the fold lost almost everything. This is the attention collapse that SGE creates, and it happens silently—no error log, no alert, no dashboard metric that flags it. The ads still load. The auction still runs. The impressions still count. But the human behavior behind those impressions has fundamentally changed.
How the Auction Internalizes the Attention Collapse
Google’s ad auction is a generalized second-price auction with quality score weighting. The effective cost-per-click an advertiser pays depends on the bid and quality score of the advertiser below them. But the rational bid—the maximum an advertiser should be willing to pay—is determined by expected value: the probability of a click multiplied by the value of a conversion multiplied by the probability of conversion given a click.
When SGE reduces the reach of ad positions, it affects the expected value calculation in two ways. First, the click-through rate for a given ad position drops because fewer users see it. If Position 1 previously had a 6% CTR and now has a 3% CTR because half the audience was captured by the SGE snapshot, the expected clicks per impression halve. Second, the users who do scroll past the SGE snapshot are self-selected for higher intent—they were not satisfied by the AI answer and want to dig deeper. This could actually increase conversion rate per click, partially offsetting the CTR loss.
The net effect depends on the query type. For informational queries where users are looking for a quick answer, the SGE snapshot likely satisfies them, and the residual audience reaching ads is small but high-intent. The EV drops because the volume collapse dominates. For transactional queries where users are ready to buy, the SGE snapshot may not satisfy the purchase intent—users still need to click through to a vendor—and the residual audience is less degraded. But even here, the attention collapse reduces the top-of-funnel awareness that display and text ads provide.
The quiet rewrite happens in the bidding layer. Smart bidding systems—Google’s automated bidding products like Target CPA and Maximize Conversion Value—use historical conversion data to set bids in real time. When SGE deploys on a query, the conversion rate from ad clicks on that query drops. The smart bidding system observes this over days and weeks and adjusts bids downward. The advertiser doesn’t need to understand SGE; the algorithm does the work. But the aggregate effect is a lower rational bid for the same ad position, which means lower auction clearing prices, which means lower revenue per search for Google on queries where SGE renders.
The Query Classification Gate
SGE does not render on every query. Google’s query classification system—part of the ranking infrastructure—determines whether a given search triggers an SGE snapshot, traditional SERP features, or both. This classification happens before the SERP is assembled, and it uses signals including query intent category, entity recognition, freshness requirements, and a commercial viability score that estimates whether the query is likely to generate ad revenue.
The leaked internal documents from the antitrust trial and the broader set of search quality guidelines reveal that Google classifies queries along multiple axes: navigational, informational, transactional, commercial investigation, and local. SGE deployment is not uniform across these categories. Observational data from SERP tracking tools shows that SGE renders most frequently on informational and commercial investigation queries—exactly the categories where ad density is lowest and ad CPCs are cheapest. It renders least frequently on high-commercial-intent transactional queries—exactly where ad revenue per search is highest.
This is not a coincidence. It’s a revenue protection strategy encoded in the query classification layer.
Consider the query “how does a heat pump work.” This is a pure informational query with minimal ad inventory. SGE can render a synthesized answer, satisfy the user, and Google loses almost no ad revenue because there were barely any ads to begin with. The cost of SGE generation—model inference, latency, server compute—is real but marginal at Google’s scale.
Now consider “buy iPhone 15 Pro.” This is a high-commercial-intent transactional query. Ads here command premium CPCs because conversion rates are high. If SGE rendered a synthesized answer here—say, a comparison of iPhone models with buying recommendations—it might satisfy the user’s need to click through to a retailer. The attention collapse would directly reduce ad clicks, and the revenue loss per search could be significant. So SGE doesn’t render here, or renders only in a minimal form that doesn’t cannibalize the ad slots.
The query classification system is the gate that determines which world you see: an AI-answer world or a traditional ad-supported SERP world. And the gate is tuned to protect the queries that generate the most ad revenue.
The Cascading Failure Analogy
What’s happening here is structurally analogous to a cascading failure in a distributed system. Google’s own Site Reliability Engineering team has documented this pattern extensively: when one component in a system absorbs load that was previously distributed across multiple components, the dependent components lose their input, and the system’s overall behavior changes in ways that weren’t designed for. The Google SRE book’s chapters on handling overload and addressing cascading failures describe how load redistribution in complex systems creates non-obvious failure modes that propagate to dependent components—a pattern that maps directly onto how SGE collapsing organic attention cascades into the ad auction’s economic layer.
In the SGE case, the “load” is user attention. The SGE snapshot absorbs attention that previously flowed to organic results and ads. The organic results lose traffic, which means publishers lose traffic, which means some publishers produce less content, which means the search index degrades over time. The ads lose impressions and clicks, which means advertisers adjust bids downward, which means auction revenue drops. The SERP’s economic equilibrium shifts, and the system hasn’t failed—it’s still serving results and running auctions—but the economics that justify the whole architecture have eroded.
The key insight from the SRE framework is that cascading failures often go unnoticed in monitoring because each component is technically functioning. The ad server is serving. The auction is clearing. The SGE model is generating. But the interaction between them has changed, and the system’s aggregate output—revenue per search, user satisfaction, publisher traffic—has shifted. This is why traditional adtech dashboards don’t catch the problem: they measure component-level metrics, not interaction-level effects.
What Advertisers Should Actually Measure
If you’re running search campaigns and SGE is rendering on your target queries, the standard metrics will mislead you. Here’s what to look at instead.
First, segment your Search Query Report by SGE presence. Google doesn’t directly flag which queries triggered SGE snapshots, but you can infer it by tracking SERP layout changes for your target keywords using a SERP monitoring tool. When SGE renders on a query where you have active ads, compare your impression share, CTR, and conversion rate before and after the SGE deployment. The delta is your attention tax.
Second, watch for smart bidding drift. If you’re using Target CPA and your actual CPA starts climbing on queries where SGE newly renders, the bidding system is responding to reduced conversion volume by raising bids to maintain target. This is the algorithm fighting the attention collapse, and it’s a sign that the query’s economics have changed. You may need to switch to manual bidding or adjust your target CPA to reflect the new reality.
Third, track the organic-to-paid traffic ratio for your domain. If SGE is cannibalizing organic traffic to your site, and you’re also running paid ads on the same queries, you may be paying for traffic you previously got for free. This is the double whammy: SGE reduces your organic reach, and you respond by increasing paid spend to maintain visibility, but the paid spend is now more expensive per click because the attention budget is thinner.
Fourth, monitor your Quality Score components. If CTR drops on a keyword because SGE is absorbing attention above your ad, Google’s Quality Score system will penalize you with a lower expected CTR component. This raises your effective CPC even if your bid stays the same, compounding the attention tax with a quality score tax. The system is designed to reward ads that get clicks, and when SGE prevents clicks, the advertiser pays twice.
Plotting the Interaction: Why This Needs Structural Mapping
The core challenge with understanding SGE’s impact is that the interaction between the answer engine, the organic results, and the ad auction is a multi-layer system with non-obvious dependencies. You can’t reason about it by looking at any single layer in isolation. You need to map the flow: query enters, classification routes it, SGE may or may not render, attention budget is allocated, organic results compete for residual attention, ads compete for what’s left, bidding system observes outcomes and adjusts. Each step feeds the next, and the feedback loops are delayed by days or weeks as smart bidding systems accumulate conversion data.
This is the kind of structural plotting that engineers in adjacent disciplines already do. In cybersecurity, the NIST Cybersecurity Framework formalizes the practice of mapping interactions between system layers—profiling how changes in one component propagate to dependent components. The NIST CSF’s structured profile and informative reference methodology exists precisely because complex systems fail at their interaction boundaries, not within individual components. The same principle applies here: SGE and the ad auction don’t fail independently, they fail together, and you need a structural map to see it.
That same discipline applies to narrative structure: before publishing, editors need a way to test events, claims, and consequences actually follow one another, which is where a novel plot generator that fits the project can function as a planning aid rather than a substitute for domain evidence.
For engineers and analysts who need to externalize this kind of structural thinking—diagramming attention flow through a SERP, tracing how a ranking change cascades to ad delivery—the discipline of plotting the relationships explicitly matters more than the diagram itself. A novel plot generator or any structural thinking aid can help enforce that discipline: the value isn’t in the output but in making yourself articulate dependencies you’d otherwise hold implicitly.
The Revenue Protection Incentive
Here’s the economic consequence that ties this together. Google has a direct, measurable incentive to limit SGE deployment on queries where ad revenue per search is high. The query classification system is the mechanism for doing this, and it operates at a scale that makes manual tuning impossible—billions of queries per day, classified in milliseconds, with SGE eligibility determined per-query in real time.
The classification system is trained on historical data: which queries generated ad revenue, which queries had high CPCs, which queries had high conversion rates. The model learns to suppress SGE on queries that look like high-revenue queries and deploy it on queries that look like low-revenue queries. This isn’t a conspiracy; it’s an optimization objective. If Google’s SGE deployment system is trained to maximize long-term revenue—which it almost certainly is, given that ad revenue is the company’s primary income source—then the system will naturally learn to protect high-value ad inventory from attention cannibalization.
The result is a bifurcated SERP. Informational queries get AI answers and minimal ads. Commercial queries get traditional ads and minimal AI answers. The web splits into two regimes: one where Google synthesizes content from publishers without sending them traffic, and one where Google auctions attention to advertisers at premium prices. Publishers lose traffic on the informational side; advertisers pay full price on the commercial side. Google captures value on both.
What Happens When Advertisers Catch On
The quiet part of this story is that advertisers haven’t fully internalized the SGE attention tax yet. Smart bidding systems are adjusting bids automatically, but the humans managing those campaigns are still looking at aggregate metrics—total conversions, total spend, ROAS—and not segmenting by SGE presence. As long as the aggregate numbers look acceptable, the per-query degradation goes unnoticed.
But as SGE expands—Google has been steadily increasing the query coverage of its AI overviews—more commercial investigation queries will trigger snapshots. These are the queries where advertisers do research-stage bidding: “best project management software,” “CRM comparison,” “marketing automation tools.” They’re not pure transactional queries, so they’re less protected by the revenue optimization. But they’re where a lot of upper-funnel ad spend goes. If SGE collapses attention on these queries, advertisers will see their research-stage campaigns degrade, and they’ll either pull spend or push it down to pure transactional queries where SGE doesn’t render.
This will concentrate ad spend on fewer, more expensive queries. The auction for “buy CRM software” will get more competitive because advertisers who used to bid on “best CRM software” are now competing for the same transactional inventory. CPCs on transactional queries rise. Google’s revenue per search on those queries rises. And the informational queries—where SGE renders and ad revenue is minimal—become a cost center: server compute for inference, with minimal ad offset.
The endgame is a search economy where Google’s AI answers are subsidized by the premium CPCs on a shrinking set of transactional queries, and publishers who used to get traffic from informational queries are cut out of the loop entirely. The attention budget has been rewritten, and the winners are the platform that controls the classification gate and the advertisers who can afford the remaining premium inventory.
For engineers building on top of this system—whether you’re a publisher optimizing for traffic, an advertiser optimizing for conversions, or a tool provider building measurement infrastructure—the takeaway is this: the SERP is no longer a single surface. It’s a branching system where query classification determines which economic regime you’re operating in, and the rules are different in each branch. Understanding which branch your queries land in is the first step to making rational decisions about where to invest attention—and money.