This three-part series will chart the rise of the ‘two-tier internet’ and its impact on the future of commerce. In this second article, Nathalie Lethbridge, Founder, and Jess Carlin, Partner at advisory firm Atonik, move beyond infrastructure to examine behaviour. Drawing on data from the 2025 US holiday season, they explore how consumers are actually using AI agents under real commercial pressure.
A few weeks ago, we took a deep dive to describe the infrastructure being assembled for what we’re calling the two-tier internet: one layer designed for human browsing, another built for AI agents that interpret intent and execute transactions. We looked at the different visions for agentic commerce that are starting to emerge: OpenAI with their push for open commerce rails, Amazon’s walled garden approach (but allowing their agents to roam the rest of the internet freely!), Google’s more hybrid model, not to mention the Agentic AI Foundation’s work to create shared standards. That was the structural story: the systems and power dynamics reshaping how commerce works online.
A few weeks on, with the frenetic pace of holiday shopping behind us, what we now have is something that the infrastructure story alone can’t deliver: behavioural proof. From Black Friday through year-end, consumers encountered agent-mediated shopping under real pressure, with all the limited time, crowded marketplaces, rising prices, cognitive overload that the holiday shopping season brings. The data doesn’t suggest traditional shopping has collapsed. But it does show a shift: growing willingness to let AI systems mediate discovery, narrow choices, and shape decisions. Is this a bell weather for a rapid global shift in consumer retail behaviour?
Clearly agentic commerce and its capabilities will influence retail. The question is how fast the consumer will adopt this new retail ecosystem and how they will do so in different markets outside the US, such as Europe. What does that mean for businesses still optimising for search-driven, navigation-heavy discovery?
What the data coming through from the US for this past holiday season tells us
The last quarter of 2025 in the US became the first large-scale stress test of AI-led commerce under commercial pressure. AI-driven traffic to US retail sites jumped 805 percent compared to the previous year during Black Friday,, while Salesforce reported that AI influenced $14.2 billion in global online sales on Black Friday alone, including approximately $3 billion in the US. found that more than half of US shoppers expected AI to influence their holiday purchases.
The 2025 US holiday season became the first real stress test for AI-led commerce. AI-driven traffic to retail sites jumped 805 percent compared to the previous year during Black Friday, according to according to Adobe Analytics, while Salesforce reported that AI influenced $14.2 billion in global online sales on Black Friday, including $3 billion in the US alone. More than half of American shoppers expected AI to influence their holiday purchases.
These figures use different attribution models, so treat them as directional. But the pattern is clear: during the most commercially intense moment of the year, US consumers were willing to let AI systems sit between intent and transaction. That’s a huge shift. .
Holiday shopping is particularly revealing because it compresses everything: timelines, budgets, patience. Decision fatigue peaks, and tolerance for friction drops. Under that pressure, many consumers didn’t just tolerate AI assistance, they actively used it to make decisions for them, and reduce the mental load that finding the perfect gift can put you under. Conversational commerce didn’t gain a foothold because Americans suddenly decided that they trusted the algorithm. It gained traction because navigating fragmented retail platforms during the holidays is exhausting, and delegating mentally draining decisions to an AI felt like relief.
From Searching to Conversation
For the last two decades, traditional search trained consumers to think like machines. Your actual search request was compressed into keywords and filters: search, scroll, refine, repeat. You, as the user, did the heavy lifting..
Conversational commerce reverses that logic. Instead of hunting for products, consumers describe outcomes. Budget, timing, values, aesthetics, constraints are all explained at once, in conversational, natural language. The agent translates that description into inventory.
Conversational commerce flips that. Instead of hunting for products, you describe what you need. Budget, timing, values, aesthetics, constraints are all explained at once, in conversational, natural language: “I’m looking for a birthday present for my dad. He likes craft beer and woodworking, my budget is around £50, and I need it to arrive before next Tuesday.” The agent translates that description into inventory.
This isn’t just a new interface. It’s a different way of thinking for the consumer. Search assumes you know what you want and can express it in machine-readable terms. Conversational commerce assumes ambiguity. It interprets context and surfaces relevance without requiring you to master complex retail taxonomies or filters.
Holiday shopping made this behavioural shift visible. Consumers increasingly described needs rather than browsing categories, asking for “a gift that fits these constraints” rather than “products in this category.” Research suggests AI-driven discovery is replacing traditional search paths during high-pressure purchase moments. Whether this leads to better decisions or simply faster ones remains unclear. But the behaviour itself (describing rather than searching) is now clearly being adopted at scale.
What Happens When Shoppers Stop Choosing and Start Delegating
The biggest behavioural change we’re seeing isn’t speed: it’s delegation.
Post-holiday research from PYMNTS suggests consumers move through delegation in stages:first using AI to explore options, then to recommend a choice, eventually to handle the entire task. The holiday season appears to have accelerated that progression, particularly for functional, repeat, and time-sensitive purchases where decision fatigue is highest.
What’s driving that comfort is harder to isolate. Trust may be a factor, but so are exhaustion, price sensitivity, and cognitive overload.
The available evidence doesn’t support claims that consumers now “trust AI more than brands.” We do not believe this change of behaviour is motivated by increased trust in the algorithm. It’s a motivated by the convenience and growing willingness to accept curated answers over comparative exploration.
What’s driving that comfort is harder to pin down. Trust might be a factor, but so are exhaustion, price sensitivity, and cognitive overload. The evidence doesn’t support claims that consumers now “trust AI more than brands.” This isn’t about increased faith in algorithms. It’s about convenience, and growing willingness to accept a curated answer rather than doing the comparative work yourself.
That shift matters for brands. When agents decide what to surface, you’re no longer just competing for visibility. You’re competing for inclusion in the agent’s shortlist.
CES and NRF: If You Build It, Consumers Will Come
If the holiday season provided evidence from the actual consumer, CES and the National Retail Federation’s Big Show in January 2026 showed what the industry is thinking, and. the message was clear: platforms and retailers are building commerce systems as if agent-mediated shopping will become normal.
At NRF, Google and Walmart announced expanded shopping capabilities inside the Gemini AI chatbot, allowing users to browse, build carts, and complete purchases without leaving the conversational interface. Google also introduced the Universal Commerce Protocol, backed by Shopify, Etsy, Wayfair, Target, and Walmart, designed to let agents operate seamlessly across discovery, checkout, and post-purchase support. The emphasis on interoperability signals an assumption: agents, not users, will increasingly navigate between retailers.
At NRF, Google and Walmart announced announced expanded shopping capabilities inside the Gemini AI chatbot, all without leaving the conversation. Google also introduced the Universal Commerce Protocol, backed by Shopify and Walmart amongst others, designed to let agents operate seamlessly across discovery, checkout, and support. The focus on interoperability signals an assumption: agents, not people, will increasingly navigate between retailers.
It may not have been announced at CES or NRF, but in January, Microsoft also reported that journeys including Copilot resulted in 53% more purchases within 30 minutes, and were 194% more likely to convert. Microsoft paired these results with Copilot Checkout and brand agent announcements, signalling deeper platform commitment rather than feature experimentation.
Retailers across verticals are aligning with this new reality. JD Sports confirmed it would enable purchases directly through ChatGPT and Microsoft Copilot, while travel giants Expedia and Booking.com integrated conversational discovery and pricing into ChatGPT. Eighteen months ago, letting discovery and checkout move upstream of your own website would have been unheard of. This is a big shift.
All that being said, it’s worth acknowledging there is some self-fulfilling prophecy going on here. When platforms invest heavily in agent-mediated infrastructure, retailers face pressure to participate. When retailers participate, consumer behaviour adapts. When behaviour adapts, platforms justify further investment. Each move reinforces the next.
Why Europe Could Lead in Conversational Commerce
Conventional wisdom says European consumers, with stronger data protection and regulatory frameworks, will resist agent-led commerce longer than Americans, but recent analysis from Boston Consulting Group suggests this might not be the case..
European consumers already operate inside heavily intermediated retail environments. Marketplace mediation isn’t new for the European consumer, it’s the default. Strong consumer protections mean Europeans expect guardrails rather than unlimited choice. Trust in regulated institutions creates comfort with delegation that less regulated markets may lack.
Additionally, the market fragmentation found in Europe makes agents more valuable,, not less. Cross-border complexity, multiple languages and varying regulations all create friction that conversational interfaces can reduce. An agent that navigates French product regulations, German returns policies, and UK delivery options whilst speaking your language? That’s incredibly useful.
BCG reports that in Europe, AI search visits grew from 4 percent of organic visits in early 2024 to 8 percent in early 2025, with projections reaching 25 percent by end of 2026 and surpassing organic search by 2028. Even if LLM-driven traffic currently represents less than 1 percent of visits, BCG estimates it could reach 3–5 percent within a year, equating to €30–€50 million per €1 billion turnover for affected retailers.
And the data supports this. BCG reports AI search visits in Europe grew from 4 percent of organic traffic in early 2024 to 8 percent in early 2025, with projections reaching 25 percent by end of 2026. Even at modest adoption levels (3-5 percent of traffic), this could represent €30-50 million per €1 billion in retailer turnover. That’s not an insignificant number!
BCG warns that European retailers who delay adaptation risk losing discovery even if they retain distribution. The fragmentation that makes Europe challenging for traditional e-commerce may make it particularly well-suited for agent-mediated commerce, but will European regulatory frameworks shape how the model operates globally? One to watch but it is likely that the convenience of conversational commerce will also trigger rapid adoption in Europe. Brands take note.
What Brands Lose When Agents Mediate Discovery
For brands, the most uncomfortable reality of conversational commerce is simple: your touchpoints disappear. Websites, editorial content, design systems, narrative…all the surfaces that carry brand meaning vanish inside conversational interfaces. Products become attributes boiled down to the essentials: image, price, description.
Narrative control erodes alongside them. Agent summaries replace the brand voice you spent years building. Marketing language becomes noise unless backed by structured, verifiable data. Adjacent discovery, those happy accidents that drive incremental purchases, diminish when agents optimise narrowly for stated intent.
For businesses built on storytelling and emotional connection, this creates genuine tension. You’re competing on functional criteria whilst your differentiation lives in narrative. How do you embed meaning into product metadata without it becoming marketing noise the agent filters out? That question doesn’t have an obvious answer yet.
What Consumers Gain, and What They Lose
For consumers, this shift isn’t just a win or lose. What you gain in one area, you often trade away in another.
Efficiency improves as effort drops, but exploration narrows:
You get exactly what you asked for, which sounds ideal until you realise how rarely you’re surprised by something you didn’t know you wanted. Those accidental discoveries whilst browsing? Less likely when agents optimise for your stated request.
Personalisation increases, but preference get reinforced:
The agent surfaces more of what you’ve liked before. Sounds helpful, until you notice how rarely you encounter products that introduce something new. Your world becomes more relevant and simultaneously narrower.
Faster answers, less transparency
The agent says “this meets your needs,” which reduces decision fatigue for some whilst leaving others uneasy because they can’t see what other options were considered or why this particular product came out on top. Traditional shopping builds confidence through reviews, ratings and comparison.
Research from Accenture and Deloitte suggests consumers tolerate the lack of transparency from agents when outcomes stay consistent, friction drops, and trust feels institutionally backed. That may explain why adoption curves differ across markets, and why European consumers might prove more receptive than expected.
Is This Trust? Or Just Convenience?
It’s tempting to say consumers now “trust AI” to make shopping decisions, but the evidence doesn’t entirely support that.
What we can observe is behavior. The Microsoft conversion metrics we discussed earlier demonstrate this clearly, with 53% more purchases within 30 minutes when Copilot is involved, 194% higher likelihood of purchase when shopping intent is present. That’s decision delegation at scale.
But is this trust? Or just exhaustion? Are consumers genuinely confident in the agent’s judgment, or are they simply taking the path of least resistance when faced with overwhelming choice? There’s a meaningful difference between trusting an algorithm and accepting its answer because deciding yourself feels too hard. The available evidence doesn’t distinguish between them.
What is clear is that the signals agents use to judge credibility (data quality, verification, fulfillment reliability) are becoming the new gatekeepers of visibility. Not search rankings. Not paid placement. Agent logic now determines what’s “relevant” to consumer intent, and consumers have no visibility into how that gets calculated or weighted.
This is absolutely key for brands. If you’re not structured for agent discovery, you’re increasingly invisible.
What This Actually Means
The internet once presented options. Now it increasingly presents answers. For consumers navigating holiday shopping chaos, that feels like relief. For brands accustomed to controlling their presentation, it feels like loss of agency. For platforms, it’s a new centre of gravity in the battle for discovery.
The shift in consumer behaviour is clearly happening. The question isn’t whether to prepare it’s how.
Next week, we’ll move from behaviour to preparation. We’ll examine how brands can maintain visibility in agent-mediated discovery, what optimisation looks like when the rules have fundamentally changed, and why some businesses – particularly smaller ones – may be better positioned for this shift than conventional wisdom suggests. We’ll look at practical readiness, new competitive dynamics, and what commerce infrastructure looks like when conversation becomes the primary interface.
For brands and retailers, the strategic question is straightforward: how do you stay visible when agents mediate discovery, when structured data outweighs storytelling, and when the path to purchase no longer runs through your homepage?
Read more…
Part 1 – The Two-Tier Internet: When AI Agents Became the New Gatekeepers of Commerce
Part 3 – Strategy in a Two-Tier Internet: How You Actually Compete When Discovery Splits in Two
