When a video appears in a retailer's search results, category pages, or product detail pages, it is not merely a social ad placed closer to checkout. The shopper is already comparing products. Price, specifications, reviews, availability, and competing SKUs sit nearby. A dramatic opening may still earn attention, but attention is no longer the whole job.
The production unit has to change. Instead of buying one hero edit plus aspect-ratio cutdowns, brands need a shelf-video system that connects each asset to a product, a retail placement, and a measurement question.
What has actually changed
On April 23, 2025, Criteo introduced Onsite Video as part of its retail media mix. The company positioned the format on retailer websites, near moments of purchase decision, and alongside onsite display and Sponsored Products. Criteo also reported a 5.6x lift in new-to-brand customers when Onsite Video was combined with Display and Sponsored Products, plus a 280% CTR increase in an Albertsons beta test for shoppers exposed to video and Sponsored Products. Those are vendor-reported results from specific combinations and tests—not an industry benchmark, and not proof that video alone caused the outcome.
The more useful production signal sits in Criteo's operating documentation. Its Onsite Video guidance says video creatives are added to Onsite Display campaigns, not to Onsite Sponsored Products. In the creative workflow, advertisers select a retailer before the available page types and formats appear. Another build guide notes that the retailer can decide whether closed captions are shown. The container is not neutral: retailer, page, format, product information, and creative all shape the delivered ad.
The network is still expanding. On May 19, 2026, Criteo announced an onsite retail media partnership with Lidl in Germany. The release specifically described Sponsored Product Ads on Lidl.de; it did not say that Onsite Video was live there. That distinction matters. Expansion of onsite retail media does not mean every format is available at every retailer.
Amazon provides a separate example of the same structural pattern. Sponsored Brands video may appear in shopping results and on detail pages, and it can lead to a Store or product detail page depending on campaign goals. Amazon also requires the video to be relevant to the advertised product or brand and to the landing context. Criteo and Amazon are different products, but both bind video to a retail environment and a purchase path.
Why the social cutdown often fails on the shelf
Social video is usually designed to interrupt a feed. It can delay product explanation with a character, a reveal, a cultural reference, or a mood. Shelf video faces a narrower question: what does a shopper still need to understand before choosing?
Three mismatches recur.
The story is memorable, but the SKU is not verifiable
A lifestyle film may communicate aspiration without showing the port layout, size relationship, setup sequence, material response, or real operating state. On a retail shelf, those details are not low-level information. They are unresolved purchase friction.
Audio carries the entire explanation
Playback behavior varies by device, page, and retailer implementation. Criteo's documentation explicitly says retailers may control whether closed captions display. Production should therefore avoid assuming that voiceover will be heard—or that baked-in captions are the only answer. The core proposition must survive muted viewing, while clean picture, dialogue, mix stems, caption files, and text-safe masters remain available for adaptation.
One edit tries to answer every purchase question
A shopper scanning search results, a shopper comparing a category, and a shopper studying a detail page are missing different information. Packing five selling points into one short edit makes the asset harder to read and makes the result harder to diagnose.
The ONCE three-layer shelf-video model
This framework is not a platform specification. It is a shared production structure for the brand, production partner, and media team.
Layer 1: Product truth
Lock the actual SKU, color, included accessories, and claims before scripting. Capture physical details and actions that matter to the purchase. Keep a traceable basis for dimensions, compatibility, performance, and demonstrated outcomes.
If generative tools are used, any change that could alter product structure, materials, ports, quantity, bundle contents, or performance should return to product-owner review. This layer should become a durable visual master: clean product actions, critical details, use context, and necessary limitations.
Layer 2: Purchase decisions
Each module should answer one shopper question:
- Who is the product for, and in what situation?
- What distinguishes it from a nearby model?
- How difficult is setup, operation, or cleaning?
- What does one key feature look like in use?
- What limitation should be understood before purchase?
The modules do not all need to become separate ads, but they should have clean entry and exit points. Search can use a fast recognition edit, category pages can use comparison modules, and detail pages can support fuller explanation without reshooting the entire production.
Layer 3: Retail adaptation
Once the media plan is known, map retailer, page type, placement, aspect ratio, duration, caption handling, linked SKU, destination, CTA, market language, and approval status. Criteo's workflow surfaces page types and formats only after the retailer is selected. Production should therefore adapt against a media–retailer–product combination, not against a generic export list.
If the retailer has not been selected, commission Layers 1 and 2 with flexible framing and protected safe areas. Do not pretend that a platform-ready deliverable exists before the platform variables are known.
Three questions determine whether a separate production is justified
Not every brand needs to reshoot immediately. Use purchase proximity, explanation load, and measurement readiness.
1. How close is the impression to a product choice?
If the video serves broad awareness and lands on editorial brand content, a compliant social master may be enough. If it appears in search, category, or detail-page environments and points to a specific SKU, a shelf-specific version deserves priority.
2. How much must motion explain?
Simple products with obvious differences can reuse existing shots after adaptation. Products involving installation, compatibility, motion, size relationships, or several similar variants need dedicated decision modules. The higher the explanation load, the more likely a cutdown removes the most valuable information.
3. Can the media report identify the conditions around each asset?
If reporting cannot separate placement, audience, SKU, or creative version, producing dozens of micro-variants is premature. Start with a few stable modules and asset IDs, then establish how the retailer or platform reports viewable exposure, viewing, clicks, attributed outcomes, and incrementality.
If purchase proximity or explanation load is high—and asset-level readback is available—a dedicated shelf-video package is usually justified. If all three conditions are weak, a careful adaptation is the more rational first step.
A production contract should specify more than “one video”
A reusable package should define at least five deliverables:
- Product-truth masters: accurate SKU actions, details, limitations, and a shot-to-claim record.
- Decision modules: separable segments for use, comparison, setup, feature explanation, and limitations.
- Separated picture and sound assets: clean picture, music and voice stems, caption files, and required safe-area masters.
- Retail adaptation matrix: retailer and placement mapped to ratio, duration, SKU, destination, CTA, language, and approval.
- Asset and test IDs: a stable link between the media report, the script promise, the shot combination, and the version date.
The cost conversation should follow the same structure:
Shelf-video cost = product-truth capture + decision-module production + retail adaptation + rights and claim management + test and version maintenance.
This is a budgeting model, not a market price claim. Product-truth capture is the work that should be least frequently repeated. Retail adaptations and test variants need explicit unit pricing and revision limits.
Measure the question before celebrating the number
The January 2024 IAB/MRC Retail Media Measurement Guidelines cover onsite, offsite, online, and in-store activity. For video, they encourage completion and interaction reporting where applicable, separate disclosure of forced duration, and duration weighting to improve comparisons across different creative lengths. They also emphasize placement-level viewability and warn that two ads with the same view rate may perform differently because of targeting, relevance, creative, attention, or engagement.
The framework also separates attributed outcomes from incrementality. In November 2025, IAB and IAB Europe published Guidelines for Incremental Measurement in Commerce Media, covering experiments, model-based counterfactuals, econometric models, and hybrid approaches.
For a brand team, the practical reading is straightforward. CTR can indicate whether a shelf expression earned further interest. Attributed sales reflect an association under the platform's rules. Incrementality tries to answer whether the outcome would have happened without the media. These are different questions.
Creative tests should change one primary variable at a time and retain retailer, placement, audience, SKU, bid, length, and exposure conditions. Otherwise, “Video A won” may only mean that Video A received a better shelf position.
The brief to prepare now
Onsite retail media video is neither a completely new film genre nor one more checkbox in a social export menu. It is product explanation embedded in a digital shelf. It must be accurate, add information relevant to the current purchase decision, and remain legible to the retailer's placement and measurement system.
When commissioning new cross-border product video and multi-platform assets, start with a one-page map of media, retailer, page type, SKU, and shopper question—not a list of 16:9, 9:16, and 1:1 deliverables. That map reveals which shots can be reused, which require new production, and which versions are actually worth testing.
