Imagine consumers who never visit a company’s website, never walk down a retailer’s aisle, and never type a brand name into a search bar. Instead, they tell an AI agent they need a gentle shampoo for color-treated hair for under $15, and the agent does the rest: scanning thousands of webpages, reviews, paid ads, and influencer posts; weighing ingredients; cross-referencing reviews; and then placing the order. All of this takes place in seconds, without the consumer ever seeing a brand logo, packaging, or multimillion-dollar ad campaign.
Welcome to agentic commerce: a form of digital commerce in which AI agents autonomously handle discovery, comparison, purchase, and post-purchase based on the shopper’s goals, preferences, and constraints.
This is not a futuristic scenario. As consumers increasingly employ AI across the full purchasing journey, traditional search is declining rapidly and large language model (LLM) traffic is exploding on platforms such as Google’s Gemini, OpenAI’s ChatGPT, and Anthropic’s Claude. BCG’s December 2025 GenAI Consumer Survey of major consumer packaged goods (CPG) companies across the US reveals that 80% of consumers are already using AI, and nearly half of those are employing it to shop, in many cases the result of retailers building their own AI agents.
Among these AI shoppers, 96% use it for product discovery and 80% for consideration—the first two stages of the traditional purchasing funnel—while 24% are using it for purchasing. Fully autonomous purchasing remains less common, with just 4% of consumers saying they would let an AI agent complete a purchase alone, but we expect adoption across the purchasing journey to continue growing.
For CPG companies and their retail partners, this is not a channel shift. It’s a full funnel restructuring, from discovery to post-purchase. No stage of the consumer journey will be spared. And make no mistake, this is ultimately a growth question. As shopping activities move increasingly to AI agents, brands that are not surfaced will lose volume today and growth tomorrow. For example, an assessment conducted for a global CPG player’s suggests that 20% to 30% of its online revenue could be at risk as discovery shifts to AI search.
As shopping activities move increasingly to AI agents, brands that are not surfaced will lose volume today and growth tomorrow.
Yet the majority of CPG companies say they are not ready. While 41% have begun to develop some agentic AI capabilities, according to our BCG GenAI CPG Survey 2026, the remaining 59% report no meaningful capabilities in this area today.
To compete in this new world, CPG companies need to show up on the AI agent’s terms at every stage of the funnel. They should employ generative engine optimization (GEO), increasing their visibility to AI systems by making their rich data and optimized content more readily available. They should build their own agents that deliver personalized experiences and collect first-party data. And they should integrate themselves into the growing agentic commerce ecosystem, extending their reach via tech players, retailers, and specialists. This, in turn, will create more data, improving GEO and creating a virtuous cycle.
Each of these three plays is a growth lever: GEO wins visibility and share at discovery, owned agents deepen engagement and first-party data, and ecosystem integration extends reach. CPG companies piloting GEO are already experiencing a 1% to 5% incremental increase in revenues, our research shows, while those testing owned agents and ecosystem integration are driving 3% to 10% revenue growth due to deeper consumer engagement and expanded reach across tech and retail platforms.
Meanwhile, first-party data collected via owned agents is enabling better targeting, driving an estimated ROI uplift of up to 50% on digital media spending.
The Funnel Is Transforming from End to End
In another BCG survey, the BCG CCI Global Consumer Radar Survey 2025, found that roughly 40% to 50% of consumers today perform extensive research, evaluating several brands before they buy. As a result, shopping journeys are becoming much more complex as consumers bounce between seeking inspiration on social media, reading product reviews across the internet, and checking e-retail sites for product and pricing options. Meanwhile, they are increasingly choosing LLM-driven platforms and AI agents as key research destinations. (See Exhibit 1.)
Of shoppers who use AI, about 70% are using it to compare product attributes, while around 50% are running pre-purchase price comparisons. Their shopping decisions are therefore becoming more methodical and multibrand, with prices and attributes increasingly important. And their purchasing funnel is changing dramatically as a result.
- At the discovery stage, this funnel is shifting from streaming and scrolling (which demand search engine optimization strategies to reach consumers) to LLM-driven recommendations, with the agent researching on behalf of the shopper. Traditional traffic and engagement are shrinking across the board in response, with a drop of 10% in organic traffic already taking place in 2024 as compared to 87% growth in LLM traffic, according to our Radar survey. While AI-driven referrals accounted for less than 1% of visits to leading European retail and brand websites in late 2025, this figure could reach 5% by the end of 2026, and AI-driven traffic is expected to overtake organic search in the US as early as 2028. As this trend escalates, brands that are not surfaced by agents will effectively be unfindable to consumers who use AI to shop.
- In the consideration stage, AI agents are generating more structured comparisons across options and causing a reduction in the number of purely impulse-driven purchases. While brand citations tend to correlate with market share at the purchasing stage, there are real gaps at the consideration stage, so there is a clear option to gain advantage at this crucial step.
- At the purchasing stage, there is still significant disintermediation between discovery and purchase. However, interactions are being transformed as the funnel compresses, and discovery and purchase will likely eventually occur in a single interaction. In the meantime, this may make it difficult for direct-to-consumer CPG companies to ensure a positive post-purchase experience. When relevant, brands must therefore be careful to ensure that client services and automated replenishment capabilities are in place.
As [LLM-led search] escalates, brands that are not surfaced by agents will effectively be unfindable to consumers who use AI to shop.
Note that agentic commerce is affecting some areas more heavily than others, given that consumers in different sectors tend to rely on research to different extents. For example, 52% of consumers in the skincare sector tend to go straight to the brands they love, with only 42% relying on research to find the right product. In contrast, only 35% of consumers buying household appliances go directly to a preferred brand, with 59% being research-led.
The Shift Is Underway
With consumers already using AI heavily, other actors in the ecosystem are not waiting. AI labs (specialized units that develop and test applications and technologies related to AI) are working with major platforms such as ChatGPT, Gemini, and Claude to launch new agentic-commerce capabilities. Meanwhile, the rollout of new standards, such as Google’s Universal Commerce Protocol, is enabling end-to-end shopping journeys from discovery to checkout (if not yet to delivery).
Retailers are building their capabilities quickly, although unevenly. Many retailers still rely entirely on passive discovery via third-party AI engines and browsers, with limited control over data sources and recommendations. But some, such as Amazon and Walmart, have already built proprietary AI agents on their own platforms. These agents support the full customer journey with deep in-house data and full control over AI reasoning, recommendations, and consumer information. Others, such as Carrefour and Target, have launched dedicated apps within ChatGPT, with retailer-controlled data access enriched by client and loyalty history.
Beneath all this movement lies a structural tension. LLM-driven platforms have the users but often rely on incomplete, outdated, or inaccurate product data; brand sites have the expertise and content but generate little organic traffic without heavy acquisition spending; and retailers are building their own agents, often without sufficient brand input. CPG companies need to find ways to ensure that the right product information surfaces in the right place and at the right time if they are to compete in this new reality.
CPG companies need to find ways to ensure that the right product information surfaces in the right place and at the right time if they are to compete in this new reality.
CPG Companies Should Pursue Three Plays to Win in Agentic Commerce
The question is no longer whether CPG companies should respond, but how to respond. Three distinct plays are emerging, each with a different objective.
A Successful GEO strategy. To ensure that a brand is visible and recommended to—and by—all AI-driven discovery surfaces (that is, LLM-driven platforms and AI agents), companies need to optimize their brand content and product data for GEO. GEO is a must-have for all CPG companies if they wish to be visible in the places where clients search, and it is a significant opportunity in the short term.
This means creating AI-readable data, structured product claims, and R&D-based scientific content that represent the kind of depth competitors can’t replicate and are the starting point for visibility and recommendations across all these surfaces. BCG has already developed GEO scorecards for a number of businesses to support this effort.
By optimizing their GEO strategy, businesses will have the potential to maintain and even gain market share. To accomplish this, they should start by understanding what consumers want, using a consumer survey and first- and third-party data to build a taxonomy of consumer prompts. They should then adjust their product claims, and potentially their products, based on what shoppers are asking agents. This taxonomy will also allow them to track how they show up across LLM-driven platforms and AI agents.
In addition, they should use the findings to create AI-optimized content: restructuring their brand sites with AI-readable backbones, embedding scientific and R&D data into product content, and structuring that content as FAQs, use cases, and schema-marked sections (marked with structured data that tells LLMs and search engines what the page is about).
Next, they need to share proprietary data, such as product information, R&D claims, or consumer insights, with retailers and tech platforms to feed their models and boost brand recommendations. They should then run pilots, measure quickly, and scale what works. And they should make sure to build the appropriate governance, organization, and processes to operate at speed and embed their learnings globally.
Early results are encouraging. A global luxury fashion group prioritized 500 consumer prompts and launched about 10 pilots, starting with technical fixes such as structured data and metadata. Within eight weeks, its brand site’s citation rate in AI search more than doubled, and within six months LLM-sourced traffic more than quadrupled.
A global luxury fashion group prioritized 500 consumer prompts and launched about 10 pilots, starting with technical fixes such as structured data and metadata. Within eight weeks, its brand site’s citation rate in AI search more than doubled, and within six months LLM-sourced traffic more than quadrupled.
An Owned Agent. To deliver personalized recommendations and experiences, companies need to build an AI-powered proprietary brand agent using owned data and brand services—with or without embedded purchasing capability—with the clear objectives of owning the consumer relationship, collecting first-party data, and unlocking partnerships within the agentic commerce ecosystem.
To implement this play, companies need to begin by defining a distinctive value proposition that no competitor or platform can replicate, building on existing brand assets and services. They should use proprietary data as the core fuel, including product knowledge, R&D-based scientific content, consumer insights, and consumer shopping missions. They should then design their owned agent for interoperability from day one so that it can plug into external ecosystems. And they should optimize the agent for first-party data capture at every interaction.
A leading personal care company shows what this effort can deliver: shoppers who chat with its AI advisor convert to buyers about 25% more often than those who do not, and digital campaigns powered by first-party data from the agent generate twice the ROI of those relying on traditional first-party data.
Ecosystem Integration. While owned agents provide strong advantages, such as full control over brand recommendations and brand positioning, their reach is limited and costly, given that consumers must be drawn directly to the brand’s own site (versus searching on LLMs or using other ecosystem agents). Strong ecosystem integration is essential for extending that reach.
Luckily, a brand with proprietary data and a working agent has a unique value proposition to bring to the agentic commerce ecosystem: the assets and legitimacy to negotiate partnerships, whether upstream with tech platforms, integrating directly into LLM platforms such as ChatGPT and Gemini; connecting downstream directly to retailers’ agents, such as Walmart’s and Amazon’s, via APIs or agent-to-agent protocols at the time of purchase; or out to specialists and advocates. All of these drive extended consumer reach and generate interaction data that feeds back into the agent, closing the loop.
Access to retailer ecosystems, however, is not guaranteed; 32% of retailers provide direct data access only to selected agents, while 87% do not share customer data with agents at all. This makes proprietary data and owned-agent capabilities even more important when negotiating ecosystem partnerships.
To implement this play, businesses need to decide where integration will best extend their reach, such as with tech platforms and retailers. They should then focus on one or two strategic partnerships as learning pilots so they can build proven playbooks before scaling, crafting a win-win value proposition that articulates what each side brings and what each side gains. And they need to operate through a dedicated cross-functional team of business and technology staff with a clear mandate and top-management sponsorship.
Spinning the Flywheel of Agentic Commerce
CPG efforts are currently leaning heavily toward investing in GEO optimization (65%) and product data structuring (47%) over the coming year, according to our GenAI CPG survey, while others are thinking of investing in owned-agent development (41%) or agentic-commerce ecosystem integration with retailers or tech platforms, such as OpenAI or Google (35%).
But for CPG companies, winning the agentic commerce battle is not about picking one of these three plays and executing it well. Instead, all three are required, as each reinforces the next, building energy like a flywheel: superior data and well-structured content mean greater capabilities for owned agents; enhanced agents mean better chances of integrating with other agents in the ecosystem; broader ecosystem integration means greater consumer research and monetization potential; stronger partnerships generate more reach and richer data. And so the flywheel spins. (See Exhibit 2.)
In contrast, CPG companies that focus only on GEO will find it is not a differentiating factor in the medium term as adoption becomes widespread. As a result, a brand that only invests in GEO and data sharing as a defensive play can become commoditized.
Instead, businesses should also invest in owned agents and ecosystem integration as necessary offensive plays. Owned agents activate the ecosystem, negotiate partnerships, and collect proprietary data that powers the agentic solutions. A brand agent that recommends the right product and consumer profile doesn’t just provide abetter experience; every interaction generates first-party signals that feed back into content, innovation, and media strategy.
A brand agent that recommends the right product and consumer profile doesn’t just provide a better experience; every interaction generates first-party signals that feed back into content, innovation, and media strategy.
Consider two CPG companies facing the same shift to agentic commerce. Both accelerate their GEO and data sharing, restructure their product data, and make sure that LLMs can find them. But one stops there, and the other goes further. It develops its own agentic solutions and uses them to activate ecosystem partnerships.
A year later, both brands are visible, but only one collects first-party data at scale, has something of value to bring to ecosystem partners, and is actively shaping how agents recommend its products. They have the same starting point but very different outcomes.
The Value Equation Is Changing
Whichever plays that CPG companies choose to pursue, one thing is certain: the commercial model that has worked for decades will not survive the agentic shift unchanged, and CPG businesses will be forced to rethink their value equation.
Basket construction, pricing, media, distribution—each of these four pillars of commerce was built for a world in which the consumer made the decisions. When agents take over, the old rules stop applying. We recommend the following steps to support each of these pillars. (See Exhibit 3.)
- Basket Logic and Cross-sell. CPG companies need to get agents to move beyond single-product recommendations and build multi-item baskets within the same brand. To do so, they should create product interdependencies within brands through LLM- and agent-readable data, such as what’s done in the beauty category: routine-based approaches in which multiple products from the same brand are combined to deliver a specific outcome.
- Pricing and Promotions. When agents make traditional discounts invisible and compress margins through instant price comparison, CPG companies need to rethink their traditional pricing and promotion strategies. They should build real-time, structured promotional data feeds that agents can read and factor into recommendations. And they should invest in high-quality data and product claim authority, both of which would more likely drive additional volume than discounts would in an agentic environment.
- Media Strategy and Content Channels. AI agents ignore traditional ads; as a result, agentic media formats are emerging, and CPG companies need to adapt. These companies should shift their media investments more toward agent-influencing formats. They should also invest early in emerging media channels, such as sponsored prompts and in-agent ads, while ensuring transparency around sponsored versus organic results to retain consumer trust.
- Distribution Channel Strategy. Agents are reducing perceived retailer differentiation and creating new direct-to-consumer pathways. In response, CPG businesses need to rethink their distribution, seizing the direct-to-consumer opportunity created by agentic commerce, such as by creating brand agents with embedded purchasing capability. By doing so, they can capture some of the consumer relationships and margin currently intermediated by retailers, while protecting their sales from any less-mature retailers and increasing their differentiation across retail channels (such as via distinct assortments) to reduce price competition.
The Window for Action is Narrowing
GEO is the no-regrets move—but this defense mechanism alone won't win. Brands that aren't discoverable today won't generate revenue tomorrow. And the next frontier is already visible: AI touchpoints are beginning to play different roles across different purchase pathways, such as helping consumers research a product, sending stock-up purchase reminders, or providing scraping tools to help surface the best deals. As a result, value will shift from aggregate to precision reach—offering the right content, agent, or partner for each shopper mission.
CPG companies ready to go further should go on the offense, building their own agent and using it to unlock ecosystem partnerships and turn first-party data into a durable competitive advantage. Those that delay will not just miss the growth opportunity. They will negotiate ecosystem partnerships afterward from a position of weakness, with fewer partnership slots, reduced leverage, and a widening data gap.
Acknowledgments
The authors thank their BCG colleagues Sebastian Bak Adamek, Roelant Kalthof, Arthur Frossard Paulin, Jimmy Royston, and Nathan Shenck for their contributions to this publication.
