A marketing team sees three signals in Merchant Center: “easy setup” is appearing frequently in conversational shopping queries, the brand’s AI share of voice is weak, and several priority SKUs are missing setup-related attributes. Does that automatically justify a new video shoot?
No. If the product fact already exists, the first fix may be the feed. If shoppers need to see a sequence, scale or operating condition, the gap may require footage. If the business cannot yet support the claim, the right action is product validation—not a more persuasive script.
Google’s AI performance insights turn conversational shopping activity into a useful planning input. For global brands that manufacture in China, the bigger opportunity is not another dashboard. It is a way to choose which questions deserve data work, which deserve a new visual proof, and which should not receive budget yet.
What the report actually measures
As of October 2026, Google Merchant Center’s AI performance insights are available for English-language queries to eligible accounts in Australia, Canada, India, New Zealand and the United States. The report covers organic shopping traffic from AI Mode and AI Overviews. Paid-ad traffic is excluded, and merchants review one product category at a time rather than one combined account-wide report.
The report includes:
- share of voice against a competitor set defined in Merchant Center;
- frequency signals for search types, terms, intents and attributes;
- the number of products showing for those signals;
- discovery, evaluation and ready-to-buy stages;
- top terms, popular attributes and top search intents.
Those metrics come with constraints. Merchants cannot manually choose the competitor set. A lack of sufficient competitor data can produce a 100% share-of-voice reading. Data is updated daily with a lag of several days. A visibility gap is therefore a place to investigate, not proof that a particular video will create sales.
Google said on September 16, 2026 that the report had become generally available in those five markets. Separately, OpenAI’s March 24 product-discovery update explains that merchants can provide structured product feeds and promotions through the Agentic Commerce Protocol so catalogs are represented more completely and accurately in ChatGPT.
These are separate company systems, not evidence of one universal industry standard. Together, however, they make one operational point hard to ignore: AI shopping visibility increasingly depends on maintained product information, while visual content still has to help a shopper understand what structured fields cannot show.
Separate a data gap from a proof gap
The same low-visibility signal can create three different production decisions.
The fact exists, but the catalog is incomplete
Material, dimensions, color, ports, included accessories and model compatibility may already be approved, yet absent from titles, descriptions or product attributes. That is a merchandising problem before it is a video problem.
Google’s optional conversational attributes include question-and-answer pairs, document links, related products, item-group titles and variant options. If the answer is already present in a manual or specification sheet, publish the correct data first and measure whether product coverage improves.
The fact is accurate, but shoppers need to see it
Some questions are visual by nature. A buyer may need to see how a projector behaves in ambient light, how a robot vacuum crosses a threshold, how a wearable fits beside glasses, or what a packed product looks like next to a carry-on.
In these cases, the video should not repeat the specification. It should show the action, scale, sequence and operating condition that make the specification understandable.
The brand does not have a defensible answer
High query frequency does not create permission to claim “works on every carpet,” “silent operation” or “all-weather protection.” If the engineering or compliance team cannot define the condition, the production brief should pause the claim. The priority is validation or a narrower statement, not a dramatic demonstration that outruns the product.
Build a SEARCH-to-SHOOT board across SKUs
Do not compare unrelated categories, markets and time windows in one ranking. Start with one category in one country, then give every candidate task seven fields:
| Field | Planning question |
|---|---|
| Demand frequency | Is this term, attribute or intent appearing consistently? |
| Visibility gap | Is the brand underrepresented versus the available competitor set? |
| Journey stage | Is the shopper discovering, evaluating or close to buying? |
| Data completeness | Do the feed, product page and documents already contain an approved answer? |
| Visual necessity | Must the shopper see an action, comparison, scale or condition? |
| SKU value | How important is the product’s margin, inventory position and market role? |
| Production effort | Can existing footage solve it, or does it require samples, talent, locations or testing? |
Teams can use internal 1–5 ratings rather than pretending to know an exact probability:
Shoot priority = demand frequency × visibility gap × journey-stage weight × visual necessity × SKU value ÷ total production effort
This is an ONCE planning model, not a Google metric or a conversion forecast. Data completeness works as a routing gate: when the answer is missing from the catalog, create a feed or page task first. When the answer is correct but visually weak, move it into production.
The board should produce four outcomes:
- Fix data; do not shoot.
- Re-edit approved footage around the question.
- Capture a missing proof shot.
- Hold the request because demand, SKU value or claim readiness is too low.
A shoot request must be specific enough to accept or reject
“Make an easy-setup video” is not a production brief. A deployable task identifies:
- the market, language, SKU and hardware revision;
- the term, attribute or buying intent behind the request;
- the approved product fact and its exceptions;
- the action, reference object, environment and sequence the camera must show;
- existing assets that can be reused;
- required versions for the product page, feed, paid media, support or sales enablement;
- the owner and trigger for future updates.
Delivery should include more than a finished edit. Keep searchable proof clips, accurate transcripts, SKU-to-question mappings, useful still frames, approved claim language, market scope and version notes. The ecommerce team can then update the page, the media team can build a cutdown, and a later product revision can retire only the footage that became inaccurate.
If your team already manages versioned product-video URLs, the next step is to connect those versions to visibility opportunities. ONCE’s guide to versioned product video for AI commerce explains why overwriting one URL destroys deployment history; the SEARCH-to-SHOOT board explains which gap deserves the next version in the first place.
Measure the cost of a resolved opportunity
For a multi-SKU portfolio, use a denominator that reflects business progress:
Cost per resolved high-priority gap = (insight review + fact verification + asset audit + production + feed/page coordination + version maintenance) ÷ high-priority tasks deployed with the data or visual gap actually resolved
Three aspect ratios and five subtitle exports do not become eight resolved opportunities when they answer one question. Conversely, one planned shoot can resolve several opportunities if each one has a specific SKU, approved fact, proof shot and deployment destination.
This model also changes the value of production partners. The job is not simply to execute a shot list. It is to help decide what deserves capture, cover several high-value proof needs while the right sample and specialists are available, and return footage in a structure the brand can keep using.
For a company with a China-linked supply chain, that planning has practical leverage. Engineering samples, factory experts and packaging variants are often easiest to access before global launch. A visibility-led brief can collect the missing proof while those resources are together, instead of commissioning fragmented reshoots after the questions surface overseas.
When to start—and when to wait
A category-level pilot makes sense when:
- AI performance insights show repeated attribute gaps across priority SKUs;
- installation, compatibility, size, materials or use conditions drive purchase confidence;
- the brand owns a large footage library but cannot retrieve clips by SKU and buyer question;
- the content calendar follows launch dates without using observed shopping demand.
Wait when the account or query language is outside the supported scope, the sample is too thin, or the problem is mainly catalog maintenance. Customer-support questions, return reasons, site search and distributor feedback can populate the same planning board until platform data is ready.
AI share of voice should not become a new vanity metric. Its best use is much simpler: helping a brand decide whether a missing answer needs to be written more clearly—or deserves to be filmed.
