Home » Expert Insights » AI Agents and Holiday 2026: A Small Audience With an Outsized Footprint

AI Agents and Holiday 2026: A Small Audience With an Outsized Footprint

Article authors: Scott Dodge

Holiday 2026 is the first season AI agents are showing up in the data, and the numbers are moving fast. Meta launched Muse on September 8th, and six weeks later more than 10 million people in the US and Canada have set it up, per Meta’s Q3 earnings call. Since early August, agentic orders have grown more than 5x according to Forter, and just about every bot-management vendor has a chart going up and to the right to prove it.  

But context matters. Muse has roughly 2.5 million daily users, about a quarter of those who set it up. A breakout year for something small is still small.  

That leaves retailers somewhere between panic and shrugging. Agents are real, they’re already on your site, and they behave nothing like shoppers. But despite the volume of headlines, our recommendations for getting ready are lighter than you might expect.  

We pulled the latest data from HUMAN, DataDome, Forter, and eMarketer to separate signal from noise. Read on for our four biggest takeaways.  

Takeaway #1: Agents Are Growing Fast. But This Holiday, Plan for Footprint, Not Revenue.

Agents will stay a small slice of holiday sales, but they’re growing fast and generate far more traffic than their share of sales would suggest. 

The data below shows how big the footprint really is:  

  1. Muse accounts for about 72% of daily AI-agent requests, HUMAN observes, roughly 5.2x ChatGPT Agent’s daily volume.  
  2. Muse’s daily requests grew 4.4x in its first nine days, from 0.8M on September 22 to 3.7M on October 1, according to DataDome.  
  3. Agentic orders are up more than 5x since early August, per Forter.  

The growth is real, but so is the context. eMarketer expects AI platforms to drive about 1.5% of US retail ecommerce sales in 2026, in a forecast made before Muse launched. And roughly 148 million monthly GenAI search users make even 10M Muse setups a modest slice. Put those together and you get a small audience generating an outsized share of agent traffic.  

Size matters less than behavior, so the next three takeaways look at how agents shop, what Muse changes, and how to see them in your data.  

Takeaway #2: Agents Act at the Bottom of the Funnel. Win Them Upstream with Content and Product Data.

Agents complete the purchase, but they choose what to buy based on what they can find and read earlier. That makes your search fundamentals the main lever, with one new job this holiday: making sure your site holds up when an agent shops it.  

Consumers use AI in three ways, and the difference is how much control consumers hand over in each mode. In Find mode, AI answers questions and recommends solutions primarily in ChatGPT, Gemini, and Google AI Mode. In Decide mode, AI assistants compare options based on specs to help consumers select a path forward. Finally, Do mode is when AI agents browse, select, and check out for the shopper.   

Find, Decide, and Do are awareness, consideration, and conversion stages by another name. Agents are the Do mode, compressing all three modes into a single task. A chatbot tells you what to buy. An agent goes and buys it.  

Agents have their biggest impact lower down the funnel where shoppers delegate conversion to AI. But what an agent chooses is decided earlier, by what it can find and read. Here’s where to influence it:  

  • Find (Awareness): Quality on-site content and strong off-site authority signals.  
  • Decide (Consideration): Product data, review quality and sentiment, and clear availability.  
  • Do (Conversion): Navigation and UX agents can use, plus agentic commerce protocols like UCP and ACP.  

 The three modes overlap, and the same platforms often span all of them. What sets agents apart is who shows up on your site: Do is the only mode where AI, not a person, is the visitor. It happens in two ways:  

  1. On your site: Browser agents like Meta’s Muse, ChatGPT Agent, Claude Browser, and Perplexity Comet act like a shopper on your site. No opt-in required.  
  2. In the platform: The shopper buys inside the AI platform through a protocol like Google’s UCP or the Agentic Commerce Protocol (ACP) without ever visiting your site. But this path takes clean product data and integration. We covered where Google is taking the transaction layer after Google I/O.  

Browser Agents vs. Agentic Checkout: Prep Your Site First

The two paths ask different things of you. Protocol checkout only works if you’ve opted in and your product data is in shape. Browser-based, screen-scraping agents can shop anywhere and show up whether or not you’re ready, so your site has to hold up. Forter shows screen-scraping agents growing from 1% of agentic orders in late July to 70% by mid October, signaling that the way AI-driven shoppers reach checkout is shifting quickly.  

For this holiday, the near-term action is getting your own site ready for these agents. If you’ve been doing search well, none of this is new. Agents reward the same fundamentals that have always mattered: accurate product data, helpful content, and authority in the sources that AI trusts. For example, in our retail AI visibility analysis, nearly half of retail AI citations went to product recommendations or long-form content. What changes is the margin for error. A person can squint past a missing spec or a clunky checkout step. An agent often can’t.  

Takeaway #3: Shopping Is Muse’s Top Use. Set Your Agent Access Policy Now.

Shopping is the most common thing people do with Muse, while personal productivity and creativity round out the top three (email and calendar are the most common connectors).  

Meta’s CFO said over 25% of Muse weekly users shop with it, and Zuckerberg put the average purchase at over $200, including travel.  

What Muse Changes for Retailers

Four things shift for retailers:  

  1. Product data stakes go up. Visits skew toward high intent, so the Takeaway #2 fundamentals get tested: a missing detail, the wrong price, or unclear delivery dates will cost you sales. Audit product pages and checkout for the details an agent needs to choose your product. 
  2. Payments look different. Muse pays through Stripe’s Link wallet: instant checkout at the more than 1 million businesses that accept Link, and a single-use virtual card everywhere else, with the shopper approving the total in chat. PayPal has since been added. Loop in your fraud and payments teams, because these orders will show up as new card patterns.  
  3. Businesses will pay. Meta says Muse will stay free for most people and that it will charge a fee on the transactions it helps facilitate, which it expects businesses to pay. Build that into your channel economics.  
  4. Access is the open question. Amazon blocked Muse on September 20, citing its Conditions of Use, while Shopify partnered with Meta on agentic checkout through Shop Pay. Meta’s CFO expects the market to move away from agents navigating sites like a human toward permissioned, structured access, including a personal agent protocol Meta is building with partners like Walmart, Stripe, and Shopify. In the near term, she expects many experiences to stay AI-assisted rather than fully autonomous. Decide your agent access policy on purpose: allow, limit, or block, with security, fraud, and merchandising at the table.  

 

Takeaway #4: Agents Shop Your Site Unannounced. Start Tracking Them Before Cyber Week.

Agents don’t identify themselves. They arrive from cloud networks, rarely log in, and blend into your traffic, so the footprint from Takeaway #1 is hard to see, and measurement is the first thing to fix before peak. Say a shopper asks Meta’s Muse: “Find me a waterproof winter jacket under $200 that I can get by Friday.” Here’s what happens next:  

  1. Muse starts a browser. Each task runs in its own cloud virtual machine with a full browser. 
  2. It browses products. It runs searches and opens product pages to compare options. HUMAN found that about 84% of Muse requests hit product and search pages.  
  3. It adds to cart. It picks the item then completes forms and cart steps. In DataDome’s sample, 8.9% of Muse requests touched forms, login, payment, or cart, and Muse mostly did it without logging in.  
  4. It checks out. It asks the shopper to approve the total, then completes your normal checkout, often with a single-use virtual card.  

What does your site see? Not a shopper’s phone. In DataDome’s sample, 99.3% of Muse requests came from cloud and CDN networks, and it doesn’t announce itself.  

Segment Agent Traffic Before It Skews Your Holiday Data

Because agents blend into your traffic, funnel, and order data, they muddy conversion rates, attribution, and fraud signals right when holiday volume peaks. Treat any agent number as a floor, including the ones in this post. And keep in mind that HUMAN and DataDome both sell detection products, so lean on the pattern across sources more than any single figure. The practical move is to pair bot and agent verification tooling with your server logs, then segment by agent type, browse path, and orders.  

Focus on Operations: Analytics, Fraud, and Bot Rules

Our experts predict that AI-platform sales will stay a low-single-digit share of holiday ecommerce, with the risk to that estimate to the upside.  

The bigger near-term impact is operational. Beyond skewed data, agents create fraud exposure and can trip bot rules that turn away legitimate agent traffic. The four steps below cover each fix.  

Four Steps to Get Ready for AI Agents Before Cyber Week

We’re not recommending a large-scale overhaul. Here are four targeted steps that fit inside the program you already run:  

  1. Measure what you can. Track agent activity with bot and agent verification tooling plus server reporting, and segment by agent type, browse path, and orders.  
  2. Make sure your site is agent-friendly. Run an agentic browsing audit, confirm your bot management isn’t blocking legitimate agentic browsers, and reduce friction points that trip agents up.  
  3. Set your agent access policy. Choose deliberately which agents to allow, limit, or block, and build the expected transaction fees into your channel economics. 
  4. Keep building fundamentals and keep learning. Focus on product data enhancements, on-site content, and off-site trust signals. The space is moving fast, so stay close to platform and partner updates and build a testing roadmap to prioritize new platforms and capabilities.  

Agents Are Real, Early, and Worth Getting Ready For

The throughline across all four takeaways is the same: agents are a small audience with a big footprint, they shop your site differently than people do, and what they pick is shaped upstream by fundamentals you already know.  

So prepare for agents, but don’t panic about it. Work through the four steps above, and you’ll be ready whether agents stay a rounding error this holiday or grow faster than anyone expects.  

At Ovative, we’re benchmarking AI platform share, testing new placements, and helping brands build for what’s next. Our AI Activation Audits give you a clear view of where you stand across discovery, site configuration, and media performance. Contact us to get started. 

SHARE

ARTICLE AUTHOR

  • Scott Dodge

    Scott Dodge

    Sr. Director, SEO

    Fueled by curiosity, Scott has a sharp eye for scalable organic search growth. Known for turning search into a powerful engine for business growth, he's laser-focused on what moves the needle for clients.