The most important thing to know about advertising in Google AI Mode is that you do not write the ads, and the product data that does the work is usually not owned by your paid media team. Shoppers are moving their product research into AI Mode; most of them never click through to a site, and Google builds the ads inside that experience from your Merchant Center feed. Google has said AI Mode passed a billion monthly users this year, so this is where a growing share of buying decisions now start. At Marketing Live in May, Google gave advertisers the formats to appear inside it, and getting into them turns out to be less about bidding than about whether your feed is ready.
This is a different conversation from agentic checkout, where an AI agent does the buying for the shopper. That is a separate shift, and I have covered the feed and measurement side of it elsewhere. This is about the paid formats a human sees when they shop conversationally in AI Mode, and what you actually have to do to appear in them.
The New Gemini Ad Formats For AI Mode & Search
Google introduced four Gemini-powered ad formats at Marketing Live across AI Mode and Search:
1.Conversational Discovery Ads appear as the shopper describes what they are trying to solve, and Gemini writes a short explainer that connects your product to the specific question they asked. The ad carries a Sponsored label and reads less like a banner and more like part of the conversation.
2. Highlighted Answers puts your product inside AI Mode’s own recommendation set, alongside the organic suggestions, again with a Sponsored label and a Gemini-written note on why it fits.
3. Business Agent for Leads replaces the lead form with a chat agent inside the ad. It matters more for service and considered purchase businesses than for straight ecommerce, so I will leave it there.
4. AI-powered Shopping Ads are the considered purchase format, and they run on Search rather than inside AI Mode. When someone is researching a television, an appliance, or an espresso machine, Gemini writes a custom explainer for your product against that exact query, pulling the features that answer what they asked. Google has said this one rolls out more broadly later this year, with date and location to be confirmed.
Your Product Feed Is Now Your Ad Creative
In AI Mode, your feed does the job your ad copy used to do, so your best ad channel is only as good as the team maintaining your product data. Nobody writes ads for AI Mode. Google generates them on the fly from your product feed, which means the factor that decides whether you show up is not your paid media team’s bidding or targeting; it’s the quality of your product data. For most brands, that data is owned by merchandising or operations, not by the people running the ads. So, the first question is not what should we bid; it’s whether our feed is good enough to be chosen, and who is responsible for making it so.
That reverses where the work goes. For years, the creative lever in paid was producing more ad variants. In AI Mode, the lever moves upstream to your product data, because that is what the model reads. Vague data produces vague ads. Rich, specific data gives Gemini something to write with.
Specifically, that is the difference between a title that says “Blue Shirt” and one that says “Men’s Blue Oxford Slim Fit Shirt, 100% Cotton,” and the difference between a description that says “high quality for every need” and one that names the material, the use case, the fit, and what it is not right for. Google’s own guidance for the AI era points to length and specificity, rich titles, product descriptions with real detail rather than a single line, and complete attributes. A one-line description that reads fine in a standard Shopping ad gives a conversational model nothing to answer with.
This matters more now because the queries changed. Google says searches in AI Mode run about three times longer than traditional searches. Nobody types “running shoes” into a conversation. They type “a neutral running shoe with extra cushion for a heavy runner under 150 dollars that does not squeak on wet pavement.” A title and a one line description cannot answer that. The feed that can is the one that wins the placement.
The Feed Attributes That Matter Most
At Marketing Live, Google added a set of optional conversational attributes to Merchant Center, built specifically for AI surfaces. Several of them carry an explicit note in Google’s product data specification that they are primarily intended for conversational experiences such as AI Mode. If you do one thing after reading this, work these into your top spend SKUs.
The one worth the most attention is the question and answer attribute. It lets you submit FAQ style pairs as structured data, a plain language question matched to a plain language answer, which is exactly the format a conversational model wants to pull from. Google’s own example is a phone with “Does it have a headphone jack?” answered directly in the feed. Most stores already have this content sitting in product page FAQs, support tickets, and presale chat logs. Mining it into Q&A pairs maps your feed directly onto the questions shoppers ask AI assistants.
The rest of the set fills in context a standard feed never carried: related products so the model understands what goes with what, variant and item group fields so it understands your range, document links for spec sheets and guides, and a popularity signal so Google knows which of your SKUs actually sell. On the AI-powered Shopping side, Google has said the system leans on attributes like material, fit, and durability to write its per query explainer, so those are the fields to fill first for anything considered.
None of this changes what shoppers see on your existing listings. It is purely additive, submitted through a supplemental feed or the Merchant API, and it does not affect product approval. It only gives Gemini more to work with when it decides whether your product answers the question.
How To Get Eligible For AI Mode Ads
You cannot target these placements directly. There is no AI Mode campaign type to switch on, and you do not pick the placement the way you would pick YouTube or Display. Eligibility comes from running Google’s AI-powered targeting: Performance Max, AI Max for Search and Shopping, Shopping campaigns, and broad match or Dynamic Search Ads as they transition into AI Max. Smart Bidding is part of the requirement, not an optional layer on top.
Once you are eligible, Gemini decides when to surface an ad for you, assembling it per query from your assets and feed and showing it when your product is a strong match for what someone asked. So the lever is not targeting. It is being enrolled in the right campaigns and giving Gemini clean, specific data to build from.
Where Paid And Organic Compound In AI Mode
This is the part most paid teams miss, and it is where paid and organic start helping each other.
The paid ad appears inside the AI Mode answer as a clearly labeled sponsored unit, but buying it does not buy your way into the organic sources Gemini cites when it builds its recommendation. Those are earned through the same structured data, product information, and authority that drive organic visibility.
But this helps you more than it hurts. The clean, specific product data that earns an organic mention in AI Mode is the exact same data Gemini uses to write a strong Conversational Discovery ad. Paid and organic are not fighting over the surface. They draw from one shared foundation, which means the work you do on product data pays off twice. The brands that win AI Mode will be the ones whose data is good enough to show up both ways.
How To Measure This When There Is No Report Yet
The reporting is where this gets honest, and it is the first question leadership will ask: How do we know this is working? For now, the honest answer is that you partly cannot. There is no clean AI Mode view today, so you cannot open a tab and see what this surface drove, which is the single biggest reason to treat it as positioning rather than a channel you optimize weekly. Several of the formats are also still in U.S. testing, and you cannot block your ads from appearing in AI Mode. So, the message up the chain is that this is preparation for where search is going, not a reportable line item this quarter.
That gap is closing on the organic side. AI Performance Insights is already in a limited U.S. pilot in Merchant Center, with Australia, Canada, India, and New Zealand to follow. It reports share of voice across AI Mode, AI Overviews, and the Gemini app, plus the conversational terms surfacing your products and where your feed has attribute gaps. Read it as a visibility instrument, not as paid performance. There is still no dedicated report for AI Mode ads.
Until it lands, use proxies. AI Max already reports on the longer conversational queries feeding your campaigns, which is the closest read you have on how people are phrasing their way to you. Watch branded search as a lagging sign that discovery is working. And once the conversational term data arrives, treat it as voice of customer research, because the language Google shows you is the language to fold back into your product data, your landing pages, and your feed. When leadership wants a number, share of voice against your competitive set is the one to point to once it arrives, and rising branded search is the proxy to show them until it does.
Where To Start On The Feed
You do not have to reinvent the wheel. The work concentrates where your money already is, so start with your highest spend categories and your best sellers, not the whole catalog. Those are the SKUs Gemini is most likely to be asked about, and the ones where a richer feed pays back fastest.
From there, the sequence is simple. Tighten titles and descriptions on those SKUs first, since that is what every format reads. Pull your existing FAQ, review, and presale chat content into question and answer pairs, because that content already exists and maps straight onto how people ask. Fill the attributes that AI-powered Shopping leans on: material, fit, and durability, for anything considered. Then ship the conversational attributes as a supplemental feed so you are layering detail on top of your live listings without touching what shoppers already see.
Do that for your top SKUs, and you have covered the queries that carry revenue. Expand down the catalog as you have time, but the first pass is about getting your money makers ready to be answered, not perfect coverage on day one.
- Surviving The Impression Squeeze: How Agentic Commerce Is Changing Google Ads In 2026
- 70% Of Top Retailers Are Invisible To Agentic Commerce – Here’s Why
- Why Your Product Feed Is An SEO Asset (And Who Should Own It)
How do you advertise in Google AI Mode?
You do not buy AI Mode placements directly. You become eligible by running Performance Max and AI Max for Search and Shopping, and Google assembles an ad from your feed and assets when your product matches a shopper’s query. The work that decides whether you show up is the specificity of your product data, not a bid or a targeting setting.
What are the new Google AI Mode ad formats?
Google introduced four Gemini built formats at Marketing Live 2026. They are Conversational Discovery Ads, Highlighted Answers, AI-powered Shopping Ads, and Business Agent for Leads. The first three matter most for ecommerce, and each uses Gemini to write a per query explainer from your product data rather than serving a finished ad, though several are still in US testing.
Do I need to change my product feed for AI Mode?
Yes, and it is the highest value thing you can do. AI Mode ads are generated from your Merchant Center feed, so vague titles and thin descriptions produce weak ads. Google also added optional conversational attributes, including a question and answer field, built specifically to help its AI surfaces understand and recommend your products.
Can I track performance in Google AI Mode?
Not cleanly yet. There is no dedicated AI Mode report today, though Google is rolling out AI Performance Insights in Merchant Center with a share of voice metric (date to be confirmed). Until then, use AI Max’s conversational query reporting and branded search as proxy signals.
