Google's AI Mode just crossed one billion monthly users. Queries are at an all-time high. And according to Google's Marketing Live announcement, 75% of people report making faster, more confident decisions using AI Mode in Search. That's the good news.
The uncomfortable news: your 2024 playbook of keyword lists, manual bid adjustments, and A/B testing individual ad variations is now a liability. The system handles those tasks automatically. Your job has shifted from campaign operator to strategic input provider, and the CFO is going to ask what that means for CAC payback before you've finished your first pilot.
Let me walk you through what actually changed, what the math looks like, and how to structure a 90-day test that Finance will sign off on.
What Gemini Actually Controls Now
Google's I/O 2026 updates made Gemini 3.5 Flash the default model in AI Mode globally. The intelligent Search box, the biggest upgrade in 25 years according to Google, now anticipates user intent with AI-powered suggestions that go beyond autocomplete. Users can search across modalities: text, images, files, videos, Chrome tabs.
For advertisers, this means the auction you're bidding in looks nothing like 2024. Performance Max and AI Max for Search now automate targeting, bidding, and creative optimization across Search, YouTube, Display, Discover, Gmail, and Maps simultaneously. The AI sets bids in real-time based on conversion likelihood for each auction, identifies high-intent audiences without manual segmentation, and generates ad variations automatically.
Your inputs are what matter now: conversion tracking accuracy, first-party data quality, and creative asset depth. The outputs are largely out of your hands.
Two New Ad Formats Worth Modeling
Google is testing two Gemini-powered formats that change how your brand appears in conversational search results.
The first is Conversational Discovery ads. These show up when users are researching complex topics and want to know exactly how a product suits their situation. Each ad includes an independent AI explainer where Gemini evaluates and synthesizes information about your product, then displays that assessment alongside your ad. You don't control the explainer copy. Gemini does.
The second format is Highlighted Answers, which surface your product as a direct response to specific user questions within AI Mode. Marketing Dive's coverage notes these are designed to provide "relevant product details along with helpful guidance" rather than traditional ad copy.
Both formats require your Merchant Center data to be clean, your product attributes to be complete, and your conversion tracking to be airtight. The AI can only recommend what it can understand.
Ask Advisor: The Agentic Layer
Google's Ask Advisor, rolling out later this year, provides a unified entry point for in-product agents across Google Ads, Google Analytics, Google Marketing Platform, and Google Merchant Center. A haircare brand that wants to identify a new customer segment could tap Ask Advisor to pull product details from Merchant Center and launch a Google Ads campaign in one action.
"On the back end, our agents talk to one another and carry each other's content, creating a continuous thread of intelligence."
Dan Taylor, VP for global ads at Google
For marketing ops, this means fewer manual handoffs between platforms. For Finance, it means faster time-to-learning on new campaigns. For the CMO, it means you need to decide which human decisions still require human judgment and which ones you're comfortable delegating to the agent layer.
The Measurement Problem Nobody Wants to Model
Here's where the CFO conversation gets uncomfortable. As one LinkedIn analysis noted, Gemini doesn't just read web pages; it interprets intent, context, visuals, and emotional tone to decide what users see. You're not competing for slots anymore. You're competing for the story the model tells.
Traditional attribution breaks down when the user journey happens inside a conversational interface. The click you're measuring might be the fifth interaction in a session that started with a question you never bid on. The conversion you're claiming might have been influenced by an AI explainer that synthesized your competitor's reviews alongside your product specs.
Andrea Tortella's framing is useful here: "How do we measure share of answers, not just share of search?" That's the right question, and Google hasn't provided a clean answer yet.
What you can measure: incrementality. Run holdout tests by geography or audience segment. Compare conversion rates and CAC payback between AI Mode placements and traditional search. Build a sensitivity table that shows Finance what happens to unit economics if AI Mode conversions are 20% more expensive, 20% cheaper, or roughly equivalent to your current mix.
A 90-Day Pilot Structure
Week 1-2: Audit your Merchant Center data completeness and conversion tracking accuracy. If your product attributes are sparse or your conversion events are misconfigured, the AI will optimize toward the wrong outcomes.
Week 3-4: Enable AI Max for Search on one campaign with clear conversion goals and a defined holdout group. Set guardrails on asset types and brand safety exclusions.
Week 5-8: Let the system learn. Resist the urge to make manual adjustments during this period. Document what the AI changes and why.
Week 9-12: Compare incrementality between test and holdout groups. Calculate CAC payback for AI-driven conversions versus your baseline. Build the sensitivity table.
The goal isn't to prove AI Mode works. The goal is to understand the unit economics well enough to forecast what happens when you scale.
What Finance Needs to See
Your board deck should lead with three numbers: CAC payback period for AI Mode placements, incremental conversion lift versus holdout, and gross margin impact if AI-generated creative underperforms your control assets.
Show the assumptions. Show the sensitivities. Show what breaks if the AI explainer starts recommending your competitor.
The teams that win in 2026 won't be the ones who adopted Gemini ads fastest. They'll be the ones who modeled the downside scenarios before scaling, built measurement infrastructure that survives the shift from clicks to conversations, and earned Finance's trust by showing the math before asking for the budget.
Model or it didn't happen.
