If your Search reporting already struggles to explain “why did this convert,” AI Mode is about to make that problem louder—and more expensive. Google is testing Gemini-powered ad placements inside AI Mode, a surface Google says now has over 1 billion monthly users.

If your Search reporting already struggles to explain “why did this convert,” AI Mode is about to make that problem louder—and more expensive. Google is testing Gemini-powered ad placements inside AI Mode, a surface Google says now has over 1 billion monthly users (per the provided coverage of Google’s AI Mode updates in 2026).

And the big shift isn’t a new bidding toggle. It’s that ads are moving closer to the answer itself: inside conversational flows, inside recommendation lists, and—via native checkout—closer to the transaction (per the provided summaries of Google’s AI Mode ad experiences and checkout work).

Here’s the one move that keeps ops teams sane while this rolls out: treat AI Mode placements like a new channel surface and run a holdout-based incrementality test before scaling budgets. Not because dashboards are “wrong,” but because they’ll be less interpretable when the “query” is a multi-turn conversation.

That’s the promise. The rest is the setup.

What changed: ads are showing up as part of the conversation

At Google Marketing Live in 2026, Google announced two Gemini-powered ad formats designed for AI Mode: Conversational Discovery ads and Highlighted Answers (as reported in the provided source article by Brooke Osmundson, Search Engine Journal staff).

Conversational Discovery ads are meant to respond to longer, exploratory prompts. In Google’s own example, a user asked how to make a home smell like “fancy spas or a rainy forest” with low-maintenance solutions (SEJ, Osmundson, May 2026). That’s not a clean keyword. It’s intent with adjectives, constraints, and vibe.

Highlighted Answers place ads inside AI Mode’s recommendation lists. Google’s example was a user researching language learning apps before a trip, where a relevant advertiser could appear inside the AI-generated recommendations (SEJ, Osmundson, May 2026).

Both formats include an AI explainer that synthesizes information about the product or service alongside the advertiser’s creative, and ads remain labeled as sponsored (SEJ, Osmundson, May 2026).

But there’s another layer. Search Engine Land’s reporting (per the provided brief) says Google is also testing AI-powered Shopping ads and rolling out Business Agent for Leads in open beta for U.S. advertisers—where users interact with a Gemini-powered brand agent trained on the advertiser’s website instead of a static lead form.

So the “ad unit” is starting to look less like a link and more like a mini product experience. That’s the pattern.

Why it matters now: your attribution model is about to get noisier

Google isn’t experimenting on a tiny surface. Again: Google says AI Mode has over 1 billion monthly users (per the provided 2026 AI Mode update coverage). That’s enough reach to change what “Search” means in weekly pipeline reviews.

At the same time, Google is pushing automation as the default layer in Google Ads—through AI Max for Search and upgrades to Performance Max (per the provided brief). Google also says advertisers created 3x more Gemini-generated assets in 2025, and that in Q4 alone Gemini was used to generate nearly 70 million creative assets for AI Max and Performance Max (per the provided brief).

That volume matters operationally. More variants means more creative fatigue risk, more brand QA, and more “which message actually drove qualified pipeline?” arguments. And when the click path gets compressed (native checkout inside AI Mode, Direct Offers surfaced inside responses), last-click gets even less useful as a story about incrementality.

Also: multiple sources in the provided results suggest classic SERP real estate is changing in AI-led experiences, sometimes with fewer predictable top-of-page placements. Ads may appear in AI Mode/AI Overviews under certain configurations (the brief cites examples like AI Max enabled, broad match, or compatible setup). Translation: eligibility and placement logic will be harder to reason about from the outside.

The uncomfortable truth: measurement and optimization won’t get simpler from here.

The one move: run an incrementality holdout for AI Mode eligibility

The goal isn’t to “prove AI works.” It’s to answer a narrower ops question: are AI Mode placements creating incremental conversions (or pipeline) beyond what existing Search would have captured anyway?

The hypothesis (make it falsifiable): If we enable AI Mode-eligible configurations for a controlled portion of Search traffic, then incremental conversions and downstream qualified pipeline will increase versus a holdout, because ads will appear inside AI Mode conversational journeys (Conversational Discovery, Highlighted Answers) where classic keyword-only coverage misses earlier-stage intent.

But the data tells a different story if it doesn’t work: you’ll see no lift, or you’ll see volume lift with worse lead quality—especially relevant if testing Business Agent for Leads (open beta in the U.S., per the provided brief) where conversation-based capture can change who raises a hand.

Run it this week: setup / launch / readout / next test

Setup (owners, scope): Marketing Ops owns tracking + experiment design. Paid Search owns campaign changes. RevOps/Sales Ops owns lead routing and lifecycle definitions. One shared doc. No heroes.

Audience: Start with one product line or one region where conversion volume is high enough to read a directional signal in 2–4 weeks. Avoid “everything everywhere.”

Budget range: Keep spend flat. Reallocate 10–20% of existing Search budget into the test cell so the CFO doesn’t have to approve a science project.

Timeline: 2 weeks to stabilize delivery + 2 weeks readout minimum. Longer if sales cycle is long; in that case, use leading indicators and accept the read is directional.

Tools: Google Ads + GA4 (or server-side events if available) + CRM (Salesforce/HubSpot) + a simple experiment tracker. Nothing fancy unless it changes data quality.

Launch: Split campaigns into two cells:

Readout (what to measure, and what not to over-interpret):

Next test: If there is lift, don’t scale budget first. Scale controls first: tighter asset QA, clearer value props on landing pages, and stricter lifecycle definitions so the new volume doesn’t poison the funnel.

The trade-off: more automation, less explainability

This is the deal Google is offering: more surfaces, more creative generation, more automation. In exchange, less visibility into the “why” behind performance.

Marketing Dive’s coverage (per the provided brief) points out that Google’s conversational AI experience in Ads can generate copy, keywords, and creative from a website URL, with early tests indicating higher-quality campaigns (improved Ad Strength) and more conversions when moving from “Poor” to “Excellent” ad quality. That’s real efficiency. It’s also a failure mode: teams can ship a lot of “pretty good” assets fast—and still miss positioning.

Practitioner commentary in the provided results makes the same point in plainer language: AI speeds up good marketing, but it doesn’t replace clear value props or authoritative content (Elevation B2B; Red Branch Media, per the provided brief). If the website is vague, the agent trained on it will be vague too.

When this is wrong: If the account is already measurement-mature, has stable query coverage, and the business can’t tolerate any short-term volatility in lead flow, a holdout test may be politically impossible. In that case, the “one move” becomes smaller: instrument new conversion events first (agent interactions, checkout steps, offer redemptions) and only then expand eligibility.

But for most teams, the practical path is the same. AI Mode is becoming a major Search surface.