For 25 years, the keyword was the atom of paid search. You named the query you wanted to intercept, set a bid, competed for the click. Clean, legible, controllable. That model is now structurally broken. Not because search died, but because Google slid an AI intermediary between the query and the result, and that intermediary doesn't run on keywords.

Google AI Overviews appear on roughly 48% of all searches in 2026. On informational queries, the ones that used to drive awareness and early consideration, organic click-through rates have fallen anywhere from 15 to 61%. Advertisers can technically show ads inside those AI-generated answers, but they can't bid for that placement specifically, can't target the conversational intent directly, and can't pull clean performance data separating AI mode from standard results. You're inside the machine. You can't see the dials.

Who's still clicking, and why that answer changes everything

The clicks that survive the AI filter are not random. When an overview answers the question, casual browsers get what they came for and leave. The people who click through saw the summary and wanted more. These are decision-stage users: higher purchase readiness, clearer questions, less patience for vague landing pages. The volume is smaller. The per-click value is higher. That's the trade most advertisers haven't priced into their models yet.

The keyword was never really the bet

What the keyword model actually priced was intent. The keyword was just a coarse proxy for it. AI search makes intent itself legible to the platform. Google now understands what someone means, not just what they typed. A brand that surfaces because it matched a string is not the same as a brand that surfaces because the AI decided it's the right answer to the underlying question. One of those positions is fragile.

The brands that win in AI search aren't the ones with the highest bids. They're the ones the AI chooses to cite.

What actually moves the needle now

The new leverage points are semantic, not syntactic. Content that answers a question clearly and completely (structured, specific, authoritative) is more likely to appear in AI overview responses and more likely to accompany an ad placement when one shows. GEO (generative engine optimization) is not a rebrand of SEO. It's a different skill. Where SEO was about signals to a crawler, GEO is about answering the AI's question well enough that it quotes you. The two compound when done together. Neither alone is enough.

• Audit intent coverage, not keyword lists. Map the questions your buyer actually asks, not just the terms they type. The AI reads meaning. Your content structure should too.
• Build answer-first pages. The first sentence under every heading should fully resolve the question it poses. AI search cites sentence-level accuracy, not page-level relevance.
• Shift budget toward high-intent clusters. The volume will look lower. The conversion rates will be higher. Both things are true at the same time, and that's fine.

The measurement gap nobody's talking about

Most advertisers are still reporting keyword-level performance in a system where keywords are no longer the unit. The gap between what gets measured and what actually drives results is getting wider. The brands that close it first will have a material edge by the time the rest of the market catches up. Those are the ones rebuilding their measurement frameworks around intent clusters, AI-cited content, and decision-stage conversion signals. That window is open now. It won't stay that way.