Google AI Mode Ads Put the Product Feed in Charge
Search Engine Journal reports that Google AI Mode ecommerce ads are generated from Merchant Center product feeds, which shifts the work from ad copywriting toward catalog quality, structured attributes, and feed operations.
TL;DR: If Google AI Mode ads are generated from Merchant Center feeds, ecommerce teams should treat product data as performance creative, not back-office plumbing.
What changes if the feed becomes the ad?
Tony Adam at Search Engine Journal reports in “How To Advertise In Google AI Mode For Ecommerce” that Google generates AI Mode ads directly from ecommerce product feeds, not from advertiser-written ad copy.
That is a small sentence with a large operational consequence.
Classic search ads gave marketers a clean layer between the product catalog and the buyer. You could write copy, test angles, choose landing pages, and shape the pitch. Product data mattered, but the ad was still a separate artifact.
In this AI Mode framing, the catalog becomes the raw material for the ad. The model, or Google’s ad system around it, decides what to show based on Merchant Center data. Search Engine Journal says specific Merchant Center attributes influence whether products get chosen, although the material here does not include Google’s own documentation or a complete attribute list. That distinction matters. This is reported guidance, not a first-party spec.
Still, the direction is clear enough: AI search ads compress search, recommendation, and product listing into one generated answer-like surface. The advertiser’s “creative” is no longer just headlines and descriptions inside Google Ads. It is the product title, category, image, availability, price fields, identifiers, variants, and the general cleanliness of the feed.
What should ecommerce teams fix first?
I would start with feed truth.
Not feed volume. Not clever keyword stuffing. Truth.
If AI Mode is choosing products from Merchant Center data, then the worst failure mode is not a weak headline. It is a mismatch between what the feed says and what the shopper actually wants. Bad variant data. Vague product names. Missing attributes. Images that do not show the product clearly. Outdated availability. Category drift. Duplicate products that compete with each other.
This is catalog operations, but with revenue pressure attached.
The practical workflow looks boring, which is usually a good sign. Pull your top revenue products and your highest-margin products. Compare the Merchant Center data against the actual product page. Then compare both against how customers ask for the item in search, chat, reviews, and support tickets. Anywhere those three views diverge, fix the feed before you complain about AI ad performance.
The other shift is measurement. If AI Mode reduces the advertiser’s direct control over ad wording, teams will need cleaner before-and-after tracking around feed changes. Change one family of attributes at a time where possible. Watch impressions, click quality, conversion rate, return rate, and assisted sales. Do not just optimize for being selected. Being selected for the wrong intent is expensive.
What is still unproven?
The missing piece is Google’s first-party detail.
Search Engine Journal’s report is useful, but operators should not treat every implied mechanism as settled policy until Google publishes the relevant docs or advertiser controls inside its own help center and product UI. Pricing, eligibility, reporting breakdowns, control surfaces, and attribute weighting should come from Google, not inferred from trade coverage.
There is also a strategic risk here. The more the ad is generated from a feed, the more advertisers may converge on the same optimization pattern. Everyone cleans titles. Everyone improves images. Everyone fills attributes. Good. Necessary. But not durable by itself.
The harder advantage is merchandising clarity. Do you actually know which products fit which intents? Do your bundles make sense? Are your product pages specific enough for an AI system to understand the difference between similar SKUs? Can your inventory and pricing systems keep up with what your ads imply?
For builders, the move is to build a feed QA loop before chasing AI Mode tricks. Create a weekly audit that flags missing, stale, inconsistent, or generic product data across your most important SKUs. Pair that with query mining from search terms, site search, reviews, and support chats. Then update the catalog in controlled batches and measure downstream behavior. The catch most teams miss: in AI-mediated commerce, the feed is not just data for Google. It is your sales pitch, your targeting, and your product truth in one place.