Black Friday 2026 lands on Friday, November 27. By every measurement published in the last ten days, it will be the first holiday season in which AI agents are a real presence in the shopping funnel at scale: reading product pages, comparing prices, summarizing reviews, and steering decisions worth tens of billions of dollars. It will also be the first holiday season in which almost none of those agents actually buy anything. Both facts are true at the same time, and the tension between them is the defining commercial story of the season.
Four datasets released between November 4 and November 15 describe the same Black Friday from four different altitudes. Adobe’s Digital Insights holiday forecast sizes the season and the AI traffic surge. HUMAN Security’s October agentic traffic benchmark shows who the agents actually are. Talkdesk’s holiday survey and Deloitte’s consumer research map where human willingness stops. And the Coalition for Trusted Reviews’ new white paper, released November 9, documents the review pollution that sits directly underneath all of it. Read together, they describe a season where the bottleneck in agentic commerce has quietly moved. It is no longer payments, which were largely settled when OpenAI shipped ChatGPT Wallet on November 13. It is no longer distribution, which Meta and OpenAI now effectively own. The bottleneck is whether the product evidence an agent reads is worth trusting at all.
The traffic layer went agentic, fast
Start with the volume. Adobe forecasts US online holiday sales of $275.1 billion between November 1 and December 31, up 6.7 percent year over year, with e-commerce now capturing roughly one in every four dollars spent on retail goods. The striking number is not the total, it is the composition: AI-driven traffic to US retail sites grew 127 percent year over year in August 2026, and Adobe projects it will grow 130 percent for the holiday season overall and 141 percent on Thanksgiving Day specifically, outpacing growth from paid search and social referrals.
Salesforce’s holiday research sizes the same shift in dollars: AI is projected to drive $75.8 billion in Cyber Week spending, up 13 percent from 2025, and AI-driven recommendations are expected to influence more than one in five holiday orders. Traffic from AI tools to e-commerce sites has more than doubled year over year, up 104 percent. Almost 20 percent of the more than 2,500 shoppers Salesforce surveyed across the US, UK, Canada, and Australia now say AI is their first stop on the way to a store, and 74 percent trust the recommendations it gives them.
“AI is no longer a niche tool just for tech-savvy shoppers,” said Caila Schwartz, Head of Agentic Commerce Shopper Insights at Salesforce. “Everyday shoppers are asking AI what to buy, where to buy it, and whether it’s worth the money before they ever land on a retailer’s site.”
Who are the agents generating this traffic? According to HUMAN Security’s October State of Agentic Traffic benchmark, published November 9, agentic commerce is now effectively a two-company town. Meta and OpenAI together accounted for 89 percent of AI agent requests across the sites HUMAN protects in October, with Meta’s Muse alone responsible for 59 percent. Total agentic traffic in October was 2.6 times its September level and roughly eight times its August level. Muse, launched September 8 as a US-only personal agent inside Meta’s apps that can browse, fill forms, and check out, passed 10 million US downloads within 40 days according to Sensor Tower, and Meta said on its October 21 earnings call that more than 10 million people in the US and Canada had set up Muse agents, with about a quarter using one on any given day. OpenAI’s three agent products, ChatGPT Agent, Codex Browser, and the Codex Chrome extension, together hold about 30 percent of requests. Perplexity’s Comet, which led with 41 percent of agentic traffic in August, has fallen to 3.6 percent.
This is the traffic that will arrive at retailers during Cyber Week: concentrated, platform-owned, and growing at multiples that make the human web look flat.
The checkout layer stayed human
Now the ceiling. In HUMAN’s October data, 85 percent of agent requests went to product and search pages. Only 1.9 percent reached checkout. That figure roughly doubled in October alone, which tells you the direction of travel, but the base is tiny: agents are, overwhelmingly, researchers.
The surveys explain why. Talkdesk’s 2026 AI Holiday Shopping Survey, which polled 1,048 US holiday shoppers and 312 retail leaders in September, found 74 percent of shoppers plan to use AI assistants for shopping this season, while 89 percent of retailers are actively preparing for AI agents that shop on a customer’s behalf. But only 15 percent of shoppers said they would let an AI agent purchase products using their payment details even with a chance to review the transaction first. Without a review step, it falls to 10 percent.
Deloitte’s survey of 4,000 US consumers, conducted in August and September, puts the trust number even lower: only 16 percent say they trust AI agents to complete purchases for them, while 66 percent worry about how their data is used by AI assistants. Salesforce found the same pattern among people who actually use personal agents: just 19 percent have deployed one at checkout, only 9 percent give an agent unrestricted access to financial details, and 36 percent have set up a separate credit card specifically to firewall their agent from their real accounts.
“We are still a long way from most consumers letting an AI agent purchase on their behalf, especially for high-consideration goods,” said Michael Klein, head of retail, travel and hospitality product marketing at Talkdesk. “Retailers are eager to push the boundaries of agentic commerce, but consumers are making it clear that trust is the ultimate currency. Earning that trust requires clear guardrails, not just new technology.”
Payment rails are no longer the blocker. ChatGPT Wallet, launched November 13 with Visa, Mastercard, and American Express, tokenizes the card, requires biometric approval per purchase, and preserves dispute rights. The payment half of agentic trust is, functionally, solved. What consumers are withholding is not their card number. It is their belief that the agent will pick something good.
The trust transfer nobody priced in
Here is the number that should concentrate minds more than any other in the Talkdesk dataset: 49 percent of shoppers say they are willing to buy from unfamiliar brands based on AI recommendations. Salesforce found the mirror image: 30 percent of AI shoppers say these tools frequently introduce them to brands they did not know.
That is a massive transfer of trust, from brand reputation, from shelf familiarity, from a friend’s opinion, to a machine’s summary of a product page. And Adobe’s data shows what happens after the transfer: 69 percent of consumers say they are less likely to return an item they bought with the help of an AI assistant. Shoppers defer to the machine’s choice even after taking possession of the product. The recommendation does not just influence the purchase, it suppresses the correction mechanism, returns, that normally polices bad purchases.
Now ask what the machine actually read. The shopper worries about this too: in the Talkdesk survey, among the 74 percent demanding stronger AI rules, 55 percent worry about unauthorized data collection, 41 percent worry about paid product placement inside AI outputs, and 41 percent worry about being charged a different price than another shopper. Those fears are not hypothetical. The FTC unsealed allegations on November 10 that Amazon charged advertisers up to 250 percent above its own auction price on peak days, and peer-reviewed work on 2.6 million Amazon search results has shown sponsored results are costlier and worse-reviewed than the organic results they displace. AI agents summarizing that shelf do not launder it. They inherit every property of the ranking they read, including the parts that were bought, and they present it in a confident first-person answer where the sponsorship is invisible.
Black Friday’s fraud surface, now machine-read
Underneath the ranking sits the review corpus, and the season’s most quietly damning dataset was published on November 9 by the Coalition for Trusted Reviews, the industry body Amazon, Booking.com, and Tripadvisor formed in 2023. Its new polling of 6,000 consumers across the UK, US, and Italy found that 82 percent of US consumers consult online reviews at least sometimes before a purchase, and nearly three-quarters of US and UK consumers have abandoned a purchase specifically because they suspected the reviews were fake or misleading.
The coalition’s report details how manipulation has industrialized: third-party review brokers reuse identities, text, tactics, and network structures across multiple platforms, which makes fraud cheap for them and expensive for any single brand to detect. Some coalition members are litigating, including a joint Amazon and Better Business Bureau lawsuit against the operators of Skitsolutionbd.com, accused of selling fake reviews across both platforms. And a new ambiguity is compounding the old fraud: half of US consumers and 70 percent of UK consumers say they trust a review less if AI helped write it, yet Tripadvisor’s own forensics on the more than 214,000 AI-flagged reviews it removed in 2024 found most did not exhibit coordinated fraud patterns. Ordinary people are using AI to write genuine reviews. The fake and the authentic are converging textually, which means surface-level detection, human or machine, is dying.
“When reviews are fake or manipulated, the consequences go far beyond one misleading rating,” said Becky Foley, executive committee member of the Coalition for Trusted Reviews and VP of trust and safety at Tripadvisor. “Consumers risk making decisions based on information that isn’t genuine, while businesses that play by the rules face an uneven playing field. Over time, this undermines trust across the online marketplace.”
Put this on top of Black Friday economics and the risk compounds. Adobe’s discount curve runs 14 percent off in early November, 21 percent just before Thanksgiving, and 30 percent at Cyber Week, and BNPL is forecast to hit a record $21.3 billion for the season. Peak discounts, peak review velocity, peak urgency, and now a machine layer summarizing it all for a shopper who has already said, at 49 percent, that it will buy an unknown brand on the agent’s word. Retailers know their side is shaky too: 31 percent of retailers in the Talkdesk survey admit an AI tool took an incorrect action on a customer’s account in the past year, among retailers reporting AI errors, 29 percent lost a customer as a result, and 37 percent do not fully disclose when AI is part of the shopping experience even though 77 percent of shoppers want to know.
What the stack is still missing
The 2026 holiday architecture now has four solid layers: discovery (Muse, ChatGPT), payment (tokenized wallets with per-purchase biometrics), logistics (arrives-by-December-20 checks), and pricing (instant cross-retailer comparison). The missing layer is product evidence: an independent read on whether the thing the agent picked is actually good, computed from review quality rather than review quantity, immune to sponsored placement, and consistent enough that an agent can cite it in its reasoning.
That is precisely the gap GoBuy was built to fill. GoBuy filters manipulated reviews out, weights authentic ones up, and condenses the result into a Smart Score from 0 to 100 based on the quality of the evidence behind a product, not the volume of it. It deliberately shows only the top seven products in a category instead of thousands, because an agent, like a shopper, needs a shortlist with a spine, not another infinite shelf. Products that hold a Smart Score above 80 across 90 days earn the GoBuy Verified badge, the closest thing the stack has to a durability signal rather than a first-impression one. And because agents are the season’s fastest-growing audience, GoBuy exposes all of it over MCP at gobuy.ai/api/mcp, so Muse, dots, and every agent built on the protocol can consult an independent trust layer before recommending, not after the return window closes. For the human holding the phone, the Chrome extension injects the trust panel directly onto the Amazon product page.
Black Friday has always been a stress test of shopper judgment. In 2026, for the first time, the judgment is delegated to a machine reading a polluted corpus at 30 percent off. The winners of this holiday season will not be the retailers who shout loudest at the agent. They will be the ones whose products survive independent verification. Ask your agent to check the Smart Score before it checks out, or start at gobuy.ai. Builders wiring agents into the season can find the MCP docs at gobuy.ai/agent-docs.