What are AI-powered Shopping ads?

AI-powered Shopping ads are a new format launched at GML 2026 where Gemini generates an explainer alongside the product card. The explainer describes why the product matches the shopper's query, using the Merchant Center feed as source material.

Does Gemini rewrite my product title?

Gemini can rewrite titles for AI Max for Shopping campaigns. For AI-powered Shopping ad explainers, it composes new descriptive copy that summarizes why the product matches a query, drawing from the description, attributes, and product_detail fields.

Will Gemini make claims my product cannot back up?

Gemini paraphrases the structured data and description you provide. The risk of unsupported claims drops when the description is factual and specific. Vague marketing copy increases the risk because the model has to interpret rather than report.

Should I write descriptions for shoppers or for Gemini?

For both. Factual, specific descriptions read well to shoppers and give Gemini accurate material to paraphrase. The two audiences want the same thing: concrete information instead of generic marketing claims.

Product Descriptions Gemini Will Use in AI Shopping Ads

At Google Marketing Live 2026, Google announced AI-powered Shopping ads where Gemini writes a per-query explainer alongside the product card. The explainer paraphrases the Merchant Center description, attributes, and structured details to tell the shopper why this product matches their search.

That single feature changes what a good product description looks like. It is no longer copy a shopper might skim on the way to the product page. It is the source material for ad copy that no human at your company will ever review before it runs.

This guide is how to write product descriptions Gemini will actually use well.

What Gemini reads when it writes an ad

Gemini does not invent product facts. It paraphrases what the feed provides. Four feed fields supply most of the raw material.

description: the longest free-text field, used for context and tone
product_highlight: bullet-style claims, used as quotable facts
product_detail: structured key-value data, used for specific specs
soft attributes (material, fit, pattern, color, size_type, age_group, gender): used for query matching and detail callouts

A description that reads well to a shopper but is light on facts produces a vague ad explainer. A description that reads like a spec sheet produces useful explainers but feels lifeless on the product page. The trick is writing copy that does both.

Three failure patterns in current descriptions

Most retail descriptions fall into one of three patterns that Gemini cannot work with.

The marketing-claim wall

“Premium quality fabric for the modern professional. Crafted with care from the finest materials. Designed to elevate your everyday wardrobe.”

Gemini reads this and has no facts to paraphrase. The ad explainer will either repeat the marketing claims word-for-word (low quality) or skip the description entirely and fall back to attributes (better but underused content).

The platform default

“Black t-shirt. Available in sizes S-XXL. Free shipping.”

The description is technically accurate but adds nothing the attributes do not already cover. Gemini has nothing distinctive to paraphrase, so explainers across competing listings start to read identically.

The wall-of-text spec dump

A 600-word paragraph cramming every detail into one block. Gemini can extract facts, but the description fails as merchant-site copy and confuses the model on what matters most.

A description template Gemini and shoppers both like

A reusable structure for product descriptions that work in both contexts.

Line 1: a concrete one-sentence summary

What the product is, in factual terms, with the one or two distinguishing details that matter most.

Example. “Slim-fit men’s chinos in 100 percent stretch cotton with a flat-front cut and a hidden coin pocket.”

Lines 2 to 4: structured fact bullets

Three to five short bullets covering material, construction, fit, and use case. Each bullet is a fact, not a claim.

Example.

98 percent cotton, 2 percent elastane for stretch and recovery
Tailored slim fit through the thigh, straight from knee to ankle
Reinforced belt loops and a hidden interior coin pocket
Machine washable on cold, hang dry recommended

Line 5: an honest use-case sentence

When and where the product makes sense. One sentence, no superlatives.

Example. “Designed for the office or weekend wear, with enough stretch for a full day of meetings or travel.”

A description in this structure gives Gemini specific facts to paraphrase, gives shoppers scannable content, and works across surfaces from Search to AI Mode to merchant-site product pages.

What to populate in product_highlight and product_detail

Description content is half the work. Structured fields do the other half.

product_highlight

Three to five short, factual highlights. Each one is one short phrase, not a sentence.

“Stretch cotton fabric”
“Slim through thigh, straight through leg”
“Reinforced belt loops”
“Hidden interior coin pocket”
“Machine washable”

Gemini quotes these directly in conversational ad formats.

product_detail

Structured key-value pairs that Gemini can present as comparison data.

material: 98 percent cotton, 2 percent elastane
fit: slim
inseam: 32 inches
care: machine washable, hang dry
closure: zipper fly with hook-and-bar closure

This is the field most retailers underfill. Filling it well is the single biggest lift you can make for AI-powered Shopping ads.

What to avoid

Three habits actively work against Gemini producing useful copy.

Superlatives without proof

“The most comfortable.” “The best fit.” “Unmatched quality.” Gemini will either skip these or reproduce them, and reproduction creates compliance risk if the claim is not verifiable.

Borrowing competitor language

Description copy lifted from a competitor or distributor often references their context, not yours. Gemini may surface that mismatch in the ad explainer.

Stuffing keywords

Old-school keyword stuffing made sense when match was keyword-based. Under Gemini-powered formats, stuffed copy makes the description harder to paraphrase cleanly. The model reads natural language better than packed phrases.

How to retrofit an existing catalog

A practical sequence for a catalog of more than 1,000 products.

Step 1: prioritize by revenue

Start with the top 20 percent of products by trailing 30-day revenue. These see the most impressions and the biggest payoff from rewrites.

Step 2: rewrite description and product_highlight together

Treat the two fields as a pair. The description provides context. The highlights provide quotable facts. Update them in the same pass.

Step 3: backfill product_detail systematically

Use a structured prompt and your product data source to populate product_detail across the priority set. This is the field most catalogs are missing entirely.

Step 4: monitor ad explainers after launch

Once AI-powered Shopping ads are running for your account, pull a sample of live ad explainers per week. Look for paraphrasing errors, missing facts, or unsupported claims. Adjust source copy to fix the pattern, not the individual ad.

What success looks like

A description Gemini uses well produces an ad explainer that.

Names the specific feature that matches the shopper’s query.
Quotes a structured detail like material, fit, or capacity.
Reads like a knowledgeable friend, not a press release.

That outcome is more likely when the source description is short, factual, and structured. It is less likely when the source is long, vague, or claim-heavy.

Where to go next

If you want the broader AI Max context, the AI Max for Shopping feed attributes guide explains how soft attributes drive conversational matching. For the full GML 2026 picture, see the Shopping recap. For systematic catalog cleanup, the Google merchant feed field audit walks through the audit pattern.

AI-powered Shopping ads turn the product description into the input for ad copy that will run thousands of times. Writing descriptions Gemini can paraphrase well is no longer copywriting hygiene. It is performance work.