Conversational Discovery Ads in AI Mode: How to Feed Gemini So It Builds Accurate Creative
At Google Marketing Live on May 20, 2026, Google introduced two new ad surfaces inside AI Mode: Conversational Discovery ads and Highlighted Answers. The disruptive part is not where the ads sit. It is who writes the creative now. The headline you used to agonize over in Google Ads is no longer yours to write. Gemini assembles it on the spot, per query.
Two numbers set the scale. AI Mode now has over 1 billion monthly users, and AI Overviews show up on about 14% of shopping queries, roughly 1 in 7 product searches. This is past the point of being a side experiment. Both formats are currently in US testing across mobile and desktop.
What the Two Formats Are and Where They Diverge from Classic Search and Shopping
Start with the mechanics. A Conversational Discovery ad fires when a shopper asks AI Mode a specific question, and Gemini builds an ad on the spot to answer that exact question. Ask “what sleeping bag for a three-night camp at minus five Celsius” and Gemini does not hand back a canned headline. It pulls attributes from your product data, the minus-10 temperature rating, the included compression sack, the 1.2kg weight, and writes a reply around those.
Highlighted Answers works differently. It does not compose a one-off response. Instead, high-quality ads become eligible to appear inside AI Mode recommendation lists. Think of it as Gemini listing a few options worth considering, and a strong ad earning a slot in that list. Both formats carry a “Sponsored” label, and Gemini writes a separate, independent explainer telling the shopper why this product fits.
The real break from classic Search and Shopping ads is when the creative gets made and who makes it. In a classic Shopping ad, the image, title, and price are fixed in your feed and shown verbatim once a keyword matches. Now Gemini assembles the message live around the shopper’s actual phrasing. The input shifts from one piece of copy you wrote in advance to the pile of structured attributes sitting in your feed. Rich attributes produce sharp creative. Thin or wrong attributes produce vague or off-base creative that still spends your budget.
Here is the side-by-side:
| Dimension | Classic Shopping ad | Conversational Discovery ad |
|---|---|---|
| Who writes the creative | You (fixed title, description, image) | Gemini (assembled live per query) |
| What input matters | Keyword match plus static copy | Structured product attributes in your feed |
| Where it shows | Search results, Shopping tab | Inside AI Mode replies and recommendation lists |
| Labeling | Sponsored | Sponsored plus an independent Gemini explainer |
| Your optimization lever | Edit copy, adjust bids, add keywords | Make the feed richer and more accurate |
The Mental Shift: You Feed Data, Gemini Writes the Ad
This is the awkward part, and the part that matters most. The old Google Ads job was writing copy, testing headlines, running A/B tests to find the line that clicked. With Gemini generating the creative, polishing that one headline stops paying off, because the sentence the shopper sees is not the one you wrote.
So where does the work go? Into the feed. The only way you shape what an ad looks like now is by making the product data you give Gemini richer and more accurate. The more material Gemini holds, the closer its creative lands to the shopper’s question. Gaps in the material mean Gemini either skips the answer or fills the hole with a guess, and the guess is usually wrong.
A concrete case. A shopper asks “is this vacuum washable and how often do I replace the filter.” If your feed never states “washable filter” and “replace every 6 months,” Gemini dodges the question or guesses from category norms. Guess wrong and the shopper returns the product. The refund and the bad review land on you. Put those two facts in the feed up front and Gemini answers correctly.
The frame has to flip completely. The question is no longer “how do I make this ad copy catchier.” It is “which questions will shoppers ask about my category, and does my feed hold the answers.” The first is a copywriting mindset. The second is a data mindset. This round rewards the second.
Which Feed Attributes and Brand-Tone Settings Drive Accurate Creative
In practice, go into Google Merchant Center and fill the feed across three buckets.
Conversational attributes first. Take the questions shoppers ask most about a product and write them into the feed as answers ahead of time. Machine washable or not, compatible with which model, battery life, warranty length, runs large or small. The stuff usually buried in PDP reviews now needs to be a structured field. List your high-frequency questions, walk the feed against that list, and fill the gaps.
Compatibility and substitutes next. Gemini fields a lot of “does this work with my device” and “is there an alternative” questions in conversation. Mark compatible models, matching accessories, and in-series substitutes clearly in the feed so Gemini can handle those follow-ups. Leave it out and the conversation stalls.
Accurate core specs last. Stop stuffing keywords into titles and descriptions. Write them straight from the real specs. Dimensions, materials, use case, key parameters. Get one wrong and Gemini repeats it wrong, stated to the shopper as fact, which does more damage than a stale classic ad ever did.
On brand tone, the new format adds controls for brand tone, messaging, and matching requirements. Where the format allows it, you can set the voice, professional and precise versus warm and casual, and constrain which messages must appear and which queries you want to match. This is not rewriting copy. It is drawing a box around Gemini so it works inside your brand’s edges and does not produce a line your brand would never say.
How AI Max for Shopping Fits and How to Test It
The companion to these two surfaces is AI Max for Shopping. At its core it is a one-click toggle that transforms your standard Merchant Center feed into conversational ad creatives, aimed at the long-tail, high-intent queries that standard Shopping campaigns miss.
Worth calling out on its own, because what standard Shopping ads miss is exactly the long-tail queries loaded with specific context and constraints. A search like “alcohol-free toner safe for sensitive skin during pregnancy” is hard for keyword matching to hit precisely, and this path catches it. Flip the toggle and your existing feed gets reused to cover that traffic.
For testing, run it this way. Pick a category where your feed data is already rich and the attributes are fairly complete, rather than flipping it on store-wide from day one. In a data-rich category, Gemini has the material to build good creative, so the test result actually means something. Start on a thin category and a bad result leaves you unable to tell whether the format failed or the feed was too sparse.
After it runs for a while, focus on the conversions coming from those long-tail, high-intent queries and compare against your existing Shopping campaigns. A lot of this traffic is incremental, queries the standard campaign never covered. So do not just watch total ROAS. Pull the newly covered queries out and evaluate them separately.
What to Watch
First, the Sponsored label and the Gemini explainer. Both formats clearly carry a Sponsored label plus an independent explainer Gemini writes itself. You do not control that explainer, and it is generated from your feed data, so a wrong feed makes even the supposedly neutral explainer wrong. Again, the root is data quality.
Second, this is US-only testing for now. Mobile and desktop both, but not a global rollout yet. If you sell outside the US, the useful move today is to enrich your feed across the three buckets above so you can run the moment it reaches your market, instead of scrambling to add data then.
Third, how to measure lift. Because the creative is built on the fly and the queries are long-tail, traditional keyword-attribution reporting will not line up cleanly. Treat this traffic as a new channel with its own report. Watch whether it brings queries and conversions you never covered before, rather than forcing it into the old Shopping comparison.
The decider this round was never how clever your ad copy reads. It is whether the product attributes in your feed are rich enough and accurate enough. The merchant with solid feed data gets Gemini to write creative that is accurate and on-point. The merchant still stuck stuffing keywords watches budget drain into vague, wrong creative, one query at a time.
Related Articles
Google AI Brief in Practice: Three Layers of Natural Language Control for AI Max
AI Brief lets advertisers steer Gemini inside AI Max using plain English across three layers: Messaging Guidelines, Matching Guidelines, and Audience Guidelines. Here is how to use each layer and what cross-border sellers should write in each field.
Google Veo 3 in Google Ads: Generate Video Ads From Product Images at No Cost
Since May 6, 2026, Veo 3 image-to-video is built into every active Google Ads account for free inside Asset Studio. Upload product photos, get a 10-second HD video with ambient audio. Here is the full workflow and what it actually produces.
Google I/O 2026's Multimodal Search Box: The Visual and Feed Assets E-Commerce Needs
At I/O 2026 Google shipped its biggest search box redesign in 25 years — it takes text, images, files, video, even Chrome tabs — and launched always-on Search Agents. AI Mode now has 1B+ monthly users. For e-commerce, product imagery, video, and structured feeds now directly decide whether you get found.