Key Takeaways
- Flywheel, the commerce agency owned by Omnicom, announced at Cannes a GEO capability to get brand products into the AI assistant recommendations of Amazon, Walmart, and Target
- The capability combines AI auditing, content optimization, and performance measurement, and in one beauty-brand pilot it drove 56% portfolio growth
- The significance is that a major agency is now productizing "building product data that AI chooses" as a systematic service
Flywheel launches a tool to get brand products into AI assistant recommendations
A new tool helps marketers appear in AI-driven recommendations.
www.adweek.comOn June 23, 2026, in Cannes, France, Flywheel, the commerce agency owned by Omnicom, announced a new capability. It is a GEO (Generative Engine Optimization) capability designed to surface brand products in the AI-driven discovery of Amazon, Walmart, and Target.
When a shopper asks an AI assistant a question, the AI returns a recommendation narrowed to a few items. A product that does not make those few is treated as if it were not on the shelf. What Flywheel productized is precisely the support a brand needs to earn one of those few slots.
Flywheel CEO Alex McCord said this at the announcement.
As product discovery rapidly shifts to AI-driven shopping experiences, brands need to fundamentally rethink how they show up in these environments.Source: Alex McCord, Flywheel CEO
AI lead Mike O'Donnell put the urgency more directly: without action, products disappear from AI-generated recommendations and organic traffic declines. What is new about Flywheel's announcement is that it tackles this challenge not as ad-hoc tweaks but as a packaged service.
Why this announcement, and why now
The backdrop is that the search experience of retail platforms is itself being replaced by AI. Amazon's AI assistant Rufus (renamed Alexa for Shopping in May 2026) has around 300 million users and is reported to generate roughly $12 billion in incremental annualized sales. Walmart's Sparky is lifting average order value, and many app users have already tried it.
When a shopper asks "what is a safe sunscreen for kids," the AI returns an answer narrowed to a few items. The behavior of scanning a results page top to bottom gives way to a short conversation with AI. In that shift, how a brand gets its products seen becomes an urgent question.
The battleground here is the "discovery layer" where AI assembles recommendations. The idea of optimizing to be chosen by AI has already been discussed as AEO (AI Engine Optimization) and as the new shelf space of the AI shopping era. Flywheel's move turns that discussion into a concrete product from a major agency.
How GEO differs from SEO
In a word, GEO is optimization to be chosen by AI. Where SEO aimed to rank high on the search results page, GEO aims to get products into the recommendations that AI assistants generate. The opponent shifted from the search algorithm to AI reasoning.
| Dimension | Traditional SEO | GEO (Generative Engine Optimization) |
|---|---|---|
| Goal | Rank high on the search results page | Get selected by AI assistant recommendations |
| Target | Search engines like Google and Bing | AI assistants like Amazon Rufus, Walmart Sparky, ChatGPT |
| Evaluation basis | Keyword match, backlinks, SERP position | Contextual relevance, conversational language, shopper intent |
| What you optimize | Meta tags, titles, link structure | Product data, attribute fields, use-case descriptions |
| Success metric | Search rank, organic CTR | Exposure in AI recommendations, citations, sales impact |
The decisive line is the evaluation basis. In SEO, keyword matches and backlink counts drove rankings. What works in GEO is how precisely a product answers shopper intent. A study analyzing Walmart Sparky found that Sparky builds recommendations by mapping products to structured attribute fields such as size, material, use case, ingredients, and ratings. If the attribute fields are blank, no matter how strong the product's track record on Amazon, it will not appear in Sparky's answer.
What Flywheel's new capability actually does
What is interesting about Flywheel's approach is that it reverse-engineers the AI recommendation logic. It analyzes what product content AI evaluates and what cues it uses to pick recommendations, then works backward to shape product data. According to the announcement, the capability integrates three elements into a single workflow.
The first is AI auditing. It checks each retail platform's product detail page (PDP) against GEO best practices based on AI behavior, algorithms, and visibility signals, surfacing the missing signals. Information that human-facing copy tends to omit, such as gift use, target age, materials, and safety details, becomes visible here.
The second is content optimization. Based on the gaps found in the audit, it rewrites product titles, bullet points, and descriptions. The goal is not search rank but to fit the AI recommendation models by including conversational language, use cases, target audiences, and functional benefits. It is the opposite of keyword stuffing, reworking copy into text that answers the questions consumers actually ask.
The third is performance measurement. It continuously tracks how the work affected traffic, conversion, and sales. In one beauty-brand pilot, refining product descriptions to match consumer intent and AI recommendation models reportedly drove 56% portfolio growth and an 80% increase in clicks and site traffic.
This capability is not Flywheel's effort alone. Parent advertising arm Omnicom Media Group frames four pillars that decide whether a brand shows up or disappears in the AI search era: Consumers, Content, Code, and Credibility. Flywheel's product-data work sits within that larger GEO strategy.
What it means that an agency is "productizing" GEO
What should not be missed is that GEO is shifting from individual trial and error to a systematic service offered by major agencies. Until now, brands could only grope to check how AI sees them. When a provider like Flywheel offers auditing, optimization, and measurement as one workflow, brands can take on GEO in a repeatable way.
This entry by agencies is not limited to Flywheel. Just as Adobe opened up AI tools including Brand Visibility for free, support around visibility in the AI search era is growing into a market where multiple players compete. From the brand's side, the options expand while the relative risk of doing nothing grows larger.
What e-commerce operators should take away
Whether or not you use an agency's GEO support like Flywheel's, the basics e-commerce operators must cover are the same.
The top priority is thorough completion of attribute fields. AI assistants do not browse a page like a human; they read structured attributes mechanically to match intent. If even one field such as size, color, use case, ingredients, or material is empty, the product quietly drops out of consideration. The Sparky analysis is explicit that products with incomplete attributes do not appear in recommendations.
Next is how you write product descriptions. Instead of listing attributes like "waterproof lightweight outdoor jacket," shift to text that answers the context shoppers voice, such as "handles light rain on an April trip to Europe and fits in a carry-on." AI picks the product best suited to a specific situation, not the all-purpose one. For reviews, too, voices that include usage context, such as "if you usually wear M, go with L," serve as better grounding for AI to answer detailed questions than star counts.
Finally, there is cross-channel consistency of price and inventory. When data conflicts, AI treats the product as a risk and drops it from recommendations. GEO should be seen not as a one-time task but as ongoing work to keep updating product data as AI behavior changes.
Conclusion
Flywheel's GEO announcement is a signal that support around the discovery layer of AI shopping is shifting to systematic services from specialist agencies. In a world where shoppers ask AI and AI answers with a few items, whether you make those few drives sales. The work of shaping product data so AI can read, reason over, and choose it is no longer optional. It is worth checking, once more, how your own product data looks to AI.
