Google AI Mode Switched From Search to Act
Last week I stopped reading about Google AI Mode and started using it. I fed it a batch of ad-banner briefs - sizes, copy, brand colors - and asked it to generate the creatives directly inside Search. Nine out of ten came back usable. Not "good for an AI," usable - the kind of thing I'd have briefed a designer for.
That's the moment it clicked for me. We've all been writing about AI Mode as a new place to get cited. We missed the bigger shift.
Stay in the loop
Get news and updates about GEO, AI search and new features. Unsubscribe anytime.
The true definition of an agent is an intelligent system that can reason, plan and act.
AI Mode stopped being a search engine and became an action engine. It doesn't just answer anymore. It does.
Search became act - here's what that actually means
If you read my I/O 2026 playbook, that was the "what's coming" version. This is the "it shipped, I tested it" version. Four capabilities turned AI Mode from a thing that retrieves into a thing that acts:
- It creates. Nano Banana 2 (Gemini 3.1 Flash Image) became the default image model inside Search/AI Mode in late February. It renders banners, social posts, and backgrounds at flexible aspect ratios - and crucially, the text rendering is finally good enough that the words on the banner come out right. That's the 9/10 I saw.
- It calls. Agentic booking expanded to home repair, beauty, and pet care, and for those categories Google will phone the business on the user's behalf to get availability and pricing. The buyer never visits your site - an agent calls your shop.
- It watches. Search Agents run quietly in the background tracking moving targets - an apartment in a radius, a tech drop, a ticket release - and ping the user the second something matches. Launching this summer for AI Pro and Ultra.
- It builds. AI Mode reads intent and codes a custom interactive widget or dashboard on the fly via Antigravity - a wedding tracker, a recovery routine, a multi-step planner - rendered right inside the results.
A retrieval surface rewards being citable. An action surface rewards being usable by an agent - findable, callable, transactable, with data clean enough that a machine can act on it without a human checking. Those are not the same job, and most stores are optimized for neither.
The retrieval game changed underneath too
Even before "act," the citation mechanics diverged from AI Overviews so hard that the old playbook doesn't carry over. The mechanism is query fan-out: AI Mode decomposes one query into up to 16 parallel sub-queries and pulls passages from separate sources for each. AI Overviews is single-pass synthesis from a handful of authoritative pages. Different machines.
The numbers make it undeniable. Ahrefs found that only 12% of AI-cited URLs rank in Google's top 10 for the original prompt, and the share of AI Overview citations coming from top-10 pages has fallen from 76% to 38% (Ahrefs). The pages AI cites are increasingly not the pages that rank - so the old ranking playbook tells you less and less about getting cited.
Knowing how to rank in AI Overviews and knowing how to rank in AI Mode are two different skill sets.
For your store, there are now two jobs - not one
Job 1: be something the acting agent can find, call, and transact
When I dug into our 50+ brand audit, the divergence between AI Overviews and AI Mode visibility was the clearest signal in the dataset - the brands winning AI Mode were the ones with clean machine-readable data, not the ones with the prettiest pages. Here's why that is now structural, not cosmetic:
| The agent wants to... | ...so your store needs |
|---|---|
| Recommend your product in a fanned-out answer | Complete Product schema + a Google Merchant Center feed |
| Show price, stock, reviews inline | Accurate GTINs, live inventory, aggregateRating in the feed |
| Call or book you for a service query | Correct business phone, hours, and service categories - consistent everywhere |
| Transact without a human | Clean returns/shipping data an agent can trust without a callback |
The gate is the feed. Google's Shopping Graph now indexes 60 billion product listings (Vidhya Srinivasan confirmed it at I/O on May 19), with 2 billion updating every hour. No Merchant Center feed, no entry into the Shopping Graph, no appearance in AI Mode shopping answers - full stop. A clean feed with accurate GTINs beats a keyword-stuffed product page every single time now. See our Product schema guide.
The May 6 update added Community Perspectives - Reddit and forum quotes - directly into AI Mode answers. So the acting agent is pulling product opinions from threads you don't own. That's the exact dynamic I covered in the Reddit piece, now baked into Google's surface.
Job 2: use the act layer yourself
This is the half nobody's talking about. The same capability that threatens your traffic is a free production tool sitting in your search bar. My banner test wasn't a gimmick - it's a real workflow:
- Draft ad and social creatives in AI Mode. Nano Banana 2 handles banners and vertical social at usable quality - in my run, 9 of 10 needed no rework. The 1 that failed was a dense pricing banner where small text broke. Brief it like you'd brief a designer: size, exact copy, brand hex codes.
- Generate product-page support imagery. Lifestyle backgrounds, size-comparison graphics, simple how-to visuals - the stuff you used to skip because design time was the bottleneck.
- Spin up landing-page mockups to test a campaign concept before you commit dev time, using the generative-UI building block.
Your competitors are still treating AI Mode purely as a threat to defend against. It's also the cheapest creative studio you have access to right now. Use it for production while you optimize for being found in it.
And you can finally measure it
The old excuse - "AI Mode traffic is untrackable" - is gone. AI Mode now has dedicated segments in Google Search Console's Search Type filter (impressions, clicks, CTR, separate from web). And GA4 added an "AI Assistant" default channel group that tags recognized assistant referrals automatically. Referral is still partially lossy on mobile and in-app browsers, but you have real numbers now. Pair it with our AI-search KPI guide.
What I'd ship this quarter
- Audit your AI Mode vs AI Overviews visibility separately. They barely overlap. Run your category queries in AI Mode specifically - don't assume your AIO wins carry over.
- Get the Merchant Center feed clean and complete. Accurate GTINs, live price and stock, aggregateRating. This is the gate to everything agentic in Google. No feed, no game.
- Fix your callable data. If you sell a service Google might book or phone for, your hours, phone, and service categories have to be right and consistent. The agent acts on what it reads.
- Set up Search Console AI Mode tracking and the GA4 AI Assistant channel today, so you have a baseline before peak season.
- Put AI Mode to work on creative. Run your next banner set through it. Worst case you learn its limits; likely case you save a design cycle.
The verdict
The rule I'd ship: stop thinking of AI Mode as a search result to rank in. It's an agent that finds, calls, transacts, and creates - and your store is now either data it can act on or noise it skips. Get the feed clean, get the callable data right, track it properly, and steal its creative engine for your own production while you're at it.
Want to see whether AI Mode can actually find and act on your store as it stands - feed, schema, callable data, the lot? Run GEOlikeaPro's Visibility Vitals checker and start there.
FAQ
How is Google AI Mode different from AI Overviews?
They're different machines with different mechanics. AI Mode uses query fan-out, decomposing one query into up to 16 parallel sub-queries and pulling passages from separate sources for each, with no organic links beneath it. AI Overviews is single-pass synthesis from a handful of authoritative pages. Ahrefs found that only 12% of AI-cited URLs rank in Google's top 10 for the original prompt, and the share of AI Overview citations from top-10 pages has dropped from 76% to 38% - so the pages AI cites are increasingly not the ones that rank, and the two require different optimization.
What does it mean that AI Mode 'acts' instead of just searching?
AI Mode now does tasks, not just answers. It generates images and banners via Nano Banana 2 directly in Search, calls businesses on the user's behalf for service categories like home repair, beauty and pet care, runs background Search Agents that track targets and alert you, and builds custom interactive tools and dashboards on the fly. For brands, that means being findable, callable and transactable by an agent matters as much as being citable.
Can I use Google AI Mode to create marketing banners?
Yes, and it works well. Nano Banana 2 (Gemini 3.1 Flash Image) is the default image model inside AI Mode and produces banners, vertical social posts and backgrounds at flexible aspect ratios with strong text rendering. In hands-on testing, roughly 9 of 10 banners came back usable with no rework - the failures tend to be dense small-text layouts. Brief it like a designer: exact dimensions, the precise copy, and brand hex codes.
Do I need a Google Merchant Center feed to appear in AI Mode?
Yes. The Merchant Center feed is the gate into Google's Shopping Graph, which now indexes 60 billion product listings (with 2 billion updating hourly) and grounds product mentions in AI Mode, AI Overviews and Gemini shopping answers. Without a feed, your products don't enter the Shopping Graph and can't appear. A clean feed with accurate GTINs, live price and stock now outperforms a keyword-optimized product page.
Can I track Google AI Mode traffic now?
Yes - the 'untrackable' era is over. Google Search Console's Search Type filter has dedicated AI Mode segments showing impressions, clicks and CTR, and GA4 added an 'AI Assistant' default channel group that automatically tags recognized assistant referrals. Referral data is still partially lossy on mobile and in-app browsers, but you now have real baseline numbers to work from.