What happened
Google is testing AI-generated product descriptions inside Shopping and product ads, a sighting Search Engine Roundtable surfaced on July 30. The test extends one already spotted on regular sponsored text results, so this isn't an isolated experiment — it's the same machinery reaching the ad format ecommerce actually lives on. Scope and rollout are unconfirmed; Google hasn't committed to shipping it.
The mechanics are what matter. Today, the description in your product feed is the copy — what you submit is what renders. If this test ships, that relationship inverts: your feed description becomes source material an AI rewrites, summarizes, or extends before the shopper ever sees it. You keep supplying the words. You stop controlling which ones appear.
Why this matters
An AI writing over sparse input doesn't leave blanks — it fills them. A vague two-line description gives the model room to improvise about materials, fit, compatibility, use cases. When it guesses wrong, the shopper doesn't blame Google; the wrong claim sits next to your product, your price, and your brand, and the return or the one-star review lands on you. Advertisers reacted to the earlier sponsored-results test with exactly this worry — text they didn't write shaping the click-through rates they pay for. Shopping was one of the last formats where your copy stayed literal.
It also fits the direction Google's ad products have been drifting for years: machine-assembled creative and campaign types that trade control for reach. Descriptions were one of the few surfaces where the merchant's exact words survived intact. Treat this test as notice that hand-placed copy in Shopping is on the clock, whether this specific experiment ships or not.
Skepticism is fair on two counts. Unconfirmed tests die quietly all the time, so restructuring your operation around a screenshot would be premature. And Google has an incentive to get this right — bad AI descriptions cost them ad revenue too. The move that makes sense is the one that pays off either way: make the feed so specific there's nothing left to invent.
What to do about it
Rewrite your weakest feed descriptions first
Sort your feed by description length and start with the shortest entries. Pack each with verifiable specifics — material, dimensions, compatibility, what's included, who it's for. Precise copy converts on its own today and constrains what an AI can invent tomorrow. Ambiguity is the only real exposure here.
Fill the structured attributes, not just the description
Color, size, material, gender, age group, GTIN, product highlights — populate every attribute Merchant Center offers. Structured fields are the least ambiguous input you can hand a rewriting system, and they improve Shopping matching right now regardless of what happens with this test.
Start a monthly ad-copy spot check
Search your top products, screenshot the Shopping ads, and compare the rendered copy against your feed. Today that confirms your words are intact; if AI descriptions roll out, the same habit becomes your accuracy monitor and your evidence if a generated claim misrepresents a product.
Watch for controls before you react
If this ships, look for the announcement and any opt-out or review setting in Merchant Center before changing strategy. Test-phase screenshots are not a rollout, and feed-accuracy work is the only response that isn't wasted if the whole thing disappears.