Google has spent two decades perfecting the art of getting you to leave Google. Every blue link, every sponsored result, every Knowledge Panel was designed to send traffic somewhere else while keeping a toll on the exit. Now the company is spending billions to build the exact opposite: a search engine that answers your question so completely you never click away at all. This contradiction would be funny if it were not worth roughly two hundred billion dollars a year in advertising revenue.
This is not the death spiral some SEO blogs have been predicting since ChatGPT launched. Google Search still processes billions of queries monthly. What is changing is where those queries end. AI Overviews and AI Mode now summarise recipes, explain code, compare products, and draft emails directly on the results page. The search box has become the destination. For publishers and advertisers who built businesses around the old model of traffic handoffs, the shift is existential. For Google, this is an Innovator’s Dilemma played out at a scale no company has faced before.
The Infrastructure Bet No One Asked For
The physical reality of this pivot is staggering. Google’s capital expenditure has swung from CPU-heavy data centres built for crawling and indexing toward a GPU and TPU buying spree that would make a mid-size nation’s defence budget look modest. Training and serving models like Gemini at search scale consumes exponentially more compute than serving ten blue links ever did. The company has reportedly restructured multiple times, cutting legacy advertising and traditional search engineering roles while consolidating talent under Google DeepMind and core AI research divisions.
Running a single LLM query costs vastly more than returning a cached index result. Multiply that across billions of daily searches and you get a company quietly licensing exclusive data from Reddit and major publishers not to send them traffic, but to feed training pipelines. The web is being strip-mined for model weights, not page views. Users have noticed the side effects. Organic results feel more cluttered and less refined because the engineering priority has shifted. The ten-blue-links algorithm is now maintenance mode software. The talent is building answer engines.
The Revenue Trap It Built for Itself
The Innovator’s Dilemma bites here. OpenAI and Anthropic launched conversational products that hallucinated, offended, and disrupted entire industries because they had nothing to lose. Every ad dollar they might eventually capture would be net new. Google invented the Transformer architecture in 2017 and sat on it. Deploying an answer engine that reduced web clicks meant threatening the core monetisation model that funds everything else. The company that moved first on AI research hesitated to move first on AI product because it understood the economics too well.
The numbers reportedly bear this out. AI Mode users generate lower advertising revenue per query than traditional search users. This is not a bug. A user who gets a complete product comparison in a chat window does not need to click three sponsored retailer links. Google cannot simply launch a separate AI advertising platform without risking that its massive advertiser base shifts budgets away from highly profitable Search and Performance Max campaigns. It is remodeling the engine of a commercial airplane at cruising altitude, and the passengers are all holding quarterly earnings calls.
How Google Is Smuggling Ads Into Answers
The current solution is characteristically cautious and technically awkward. Rather than building a standalone AI ad platform, Google is forcing AI monetisation into existing AdWords architecture. Sponsored placements now appear inside AI Overviews and AI Mode, but advertisers cannot specifically target these placements or opt out of them. AI queries blend into standard cost-per-click auctions. Your traditional search campaign implicitly funds the AI engine whether you intended to or not.
This creates a strange subsidy. The legacy advertiser base, built on intent-driven clicks to external sites, now underwrites a product designed to reduce exactly those clicks. The technical challenges are equally contorted. Traditional search ads run millisecond-level auctions for static text links. LLMs generate unique, dynamic responses in real time. Injecting relevant, timely ads without adding latency or creating the appearance of heavy bias requires solving problems that did not exist in the previous architecture. Google fears degrading AI Mode UX with clumsy commercial insertions, yet needs the revenue to justify the infrastructure spend.
The Upper Funnel Fiction and the Real Race
Google’s public framing leans heavily on a distinction between research queries and purchase intent. AI Mode searches are characterised as upper funnel behaviour, users in broad exploration phases who need nurturing through multi-turn conversations before they are ready for sponsored product recommendations. The company wants to perfect conversational intent targeting before dropping high-value ads into chat flows.
This narrative serves multiple purposes. It justifies patience to shareholders already sceptical about AI capital expenditure. It explains why Google is moving slower than OpenAI, which launched advertiser systems recently and needs profitability paths to justify its valuation. It positions the current lower AI Mode revenue as a strategic choice rather than a structural problem.
Whether the distinction holds is another question. Users asking “best running shoes for flat feet” in a chat interface are expressing purchase intent whether Google labels them upper funnel or not. The delay in monetising that intent reflects genuine technical and UX constraints, but also a company terrified of accelerating the very cannibalisation it is already engineering. The risk is not just that OpenAI gets there first with advertiser tools. It is that advertisers themselves, particularly in performance-driven categories, begin allocating experimental budgets toward platforms that can demonstrate clearer attribution in conversational contexts.
What This Means for the Working Practitioner
For South African marketers and site owners, the operational implications are already arriving. Traditional SEO logic, optimising for crawlability and ranking signals, is giving ground to Answer Engine Optimisation and Generative Engine Optimisation. The goal is no longer exclusively to win the click but to be sourced in the answer, to have your brand or data surface in AI-generated summaries even when no visit occurs.
This requires different tactics. Structured data implementation becomes more critical because LLMs rely heavily on machine-readable context. Brand mention monitoring matters across platforms beyond your own site, since training data licensing deals mean your content may be summarised from third-party repositories. The traffic you do receive will likely skew toward high-intent, complex queries that AI cannot fully resolve, making conversion rate optimisation on remaining visits proportionally more valuable.
The advertising side is murkier. Google’s forced integration of AI placements into standard auctions means most practitioners are already participating in AI Mode monetisation without explicit control. Monitor campaign performance for unexplained efficiency shifts. Queries that historically converted well may degrade as AI Overviews intercept the intent earlier in the funnel. Budget reallocation toward formats with clearer placement transparency, whether within Google’s own ecosystem or toward competitors, deserves serious testing.
Google will likely accelerate its AI-first corporate identity pitch to shareholders because the alternative is explaining why capital expenditure doubled while ad revenue per query compressed. The company owns more proprietary data, more real-time behavioural signals, and more infrastructure than any competitor. Whether that advantage translates to winning the AI advertising race depends on whether it can solve the monetisation problem it created for itself without destroying the machine that created the advantage in the first place. The rest of the industry is just trying to build workable strategies around a company that no longer knows whether it wants you to stay or to go.
