The 2026 holiday season is already underway, which is a strange thing to write on October 12. Amazon’s October Prime event ran October 6 to 7 and drove $9.9 billion in US online spend, up 9.2% year over year, with discounts peaking at 19% off list price. Adobe expects the full month of October to reach $95.8 billion online, up 8%: effectively a Black Friday’s worth of extra spending stacked in front of the season, as Adobe’s own forecast framing puts it. The turkey hasn’t been bought yet, but the retail year’s biggest quarter has started, and for the first time, the industry’s main forecasts are being written through an AI lens.

What they describe is two different holidays happening simultaneously. One is a record-breaking, discount-driven, mobile-first spending season that looks like every prior year, just bigger. The other is a structural change in how a large share of Americans now shop: research conducted inside AI chat, recommendations trusted more than reviews, and discovery that begins before any retailer’s website is ever loaded. The numbers from Adobe, Salesforce, and PYMNTS Intelligence, published over the past three weeks, are worth reading together, because the gaps between them describe the actual state of AI commerce better than any single forecast does.

What the forecasts actually say

Adobe’s holiday forecast, released September 28 and built on more than 1 trillion visits to US retail sites, 100 million SKUs, and 18 product categories, expects a record $275.1 billion in US online sales from November 1 through December 31, up 6.7% year over year. Ecommerce will account for one of every four dollars spent on retail goods this holiday. Cyber Week is projected to drive $47.5 billion, 17.3% of the entire season, with Cyber Monday alone crossing $15.1 billion and Black Friday reaching $12.9 billion. Electronics ($63.3 billion) and apparel ($51.3 billion) remain the tentpoles, grocery is the fastest-growing major category at 10.3% growth, and Buy Now Pay Later is expected to carry $21.3 billion of the season, including $1.09 billion on Cyber Monday alone.

Buried in the same release is the line that matters most for this column: Adobe expects AI-referred traffic to US retail sites to rise 130% year over year this holiday season, measured by shoppers clicking through from AI services, with the single biggest spike landing on Thanksgiving Day and strong surges on Black Friday (up 95%) and Cyber Monday (up 88%). That is not a forecast of a marginal channel. It is a forecast of a doubling-plus in a traffic source that barely registered two years ago, arriving precisely during the days when the most money changes hands.

Salesforce’s 2026 holiday predictions go further. The company projects that 20% of all holiday ecommerce traffic will originate from AI chat agents, a mix it explicitly breaks down into consumer-facing bots handling live queries, autonomous agents executing backend tasks, and competitor scrapers feeding algorithmic price-matching. Behind that projection sit adoption numbers that have stopped being speculative: 50% of shoppers now report using an AI assistant at some point in their buying journey, up 67% year over year, and reliance on AI assistants as the first stop of the shopping journey grew 200% between May 2025 and May 2026. Traditional search engine use and online marketplace use each fell 15% year over year, while what Salesforce calls “new channels,” AI assistants and social, surged 38%.

PYMNTS Intelligence’s September 2026 report, “Will the 2026 Shopping Season Go Agentic?”, puts a population count under the trend: nearly 132 million US adults have already used AI to assist in a retail purchase, 46% of shoppers specifically use AI to hunt for the best deal, and 58.4 million US adults now begin their online retail research with AI, a figure up 18% since June alone. More than half the US adult population has now shopped with an AI copilot. That is no longer an early-adopter behavior. It is the median behavior.

Discovery has been won. Checkout has not.

Here is where the two holidays diverge. The same PYMNTS report finds that 59% of AI-assisted purchases still conclude on Amazon. Checkout.com’s Agentic Commerce 2026 data puts agents at roughly 3% of actual merchant transactions. And Worldpay’s research finds only 14% of consumers trust AI to execute a purchase without manual verification.

Read those four numbers as one sentence: 132 million Americans use AI to shop, and almost none of them let it spend. The AI does the reading, the comparing, the deal-hunting. Then the human takes the recommendation, opens the familiar store, and presses buy. As Forkast’s analysis of the PYMNTS data put it, AI has won the discovery war and lost the checkout battle, functioning today as a force multiplier for the incumbents rather than a disruptor of them.

The interesting wrinkle is that the loyalty is conditional. PYMNTS finds 53% of consumers would leave a large marketplace if a brand’s own site offered a lower price. The gravity pulling 59% of AI-assisted purchases to Amazon is not love; it is default trust in a checkout that has never burned them. Which means the entity that solves verifiable trust outside the walled gardens has a real opening, because the shoppers have already told us the door is unlocked.

The number that should stop everyone: 41% skip the reviews

Now the finding that deserves far more attention than it has received. In its holiday research, Salesforce reports that 74% of shoppers say they trust the product recommendations they receive from AI chat. That alone would have been an extraordinary stat two years ago. But the sharper number sits in the brand-agent data: 41% of shoppers say a brand-owned AI assistant answering their questions makes them “much more confident” in a purchase, to the point, in Salesforce’s words, “where it replaces the need to read reviews entirely.” Another 36% report being somewhat more confident. Only 5% don’t trust the answers at all.

Sit with what “replaces the need to read reviews entirely” means. The review corpus was the internet’s answer to the used-car salesman: imperfect, spammy, gameable, but independent of the seller and voluminous enough that a motivated buyer could triangulate truth from a thousand strangers. It is now being swapped, mid-holiday-season, for a single conversational answer produced either by a general AI platform with its own ranking incentives or by a brand-owned assistant whose employer sells the product. The business case is real: Salesforce measured retailers using branded AI shopper agents seeing 59% higher holiday sales growth in 2025 than those without, 6.2% versus 3.9%, and predicts one in three ecommerce sites will have a site-specific shopper agent live by Cyber Week.

No independent evidence standard travels with that handoff. The assistant’s answer about a product is only as good as the data underneath it, and the data underneath it is, overwhelmingly, the review corpus plus the seller’s own product claims. The review corpus is contaminated at industrial scale: Trustpilot removed 4.5 million fake reviews tied to a single platform in 2024 alone, and Amazon itself reports blocking hundreds of millions of suspected fake reviews per year. When an AI summarizes reviews, it inherits the fakes with perfect confidence and no visible denominator. A shopper who skipped reading the reviews did not skip the fake reviews. They just stopped being able to tell them apart.

Adobe’s own consumer survey, 5,000 respondents fielded in July, shows what is at stake in both directions: 77% of shoppers who have used AI for online shopping report feeling more confident in their purchase, and 69% say they are less likely to return an item they bought with AI help. Genuine, well-sourced AI assistance measurably reduces doubt and returns. Confidence built on manipulated inputs produces the opposite: a purchase that arrives not as described, a gift that disappoints on December 24, and a return whose real cause, contaminated evidence, never shows up in any analytics dashboard. The confidence dividend and the manipulation risk are the same mechanism, pointed at different data.

Two holidays, one AI, split by income

Salesforce’s forecast also closes the loop on who is actually using AI this season, and it is not one audience. Consumer pessimism sits at 52% among low-income shoppers and 50% among middle-income shoppers, versus 36% for high-income earners. Thirty-five percent of all holiday shoppers are actively trading down to cheaper alternatives. In that environment, Salesforce finds the lower and middle parts of the income curve using AI defensively: cross-comparing prices, hunting deals, stretching a budget. The top of the curve uses the same technology for premium curation and personalized recommendations. Same tools, opposite intent.

The physical store is not going anywhere, either: 77% of shoppers still prefer physical stores for holiday shopping, ahead of online marketplaces at 69%, and 79% of in-store shoppers are on their phones in the aisles, with 12% consulting an AI assistant right at the shelf. Wall Street has noticed the same asymmetry from the other side. When CNBC surveyed what AI shopping means for retail stocks, Wells Fargo analyst Ike Boruchow pointed to the off-price “treasure hunt” at T.J. Maxx and Marshalls as the thing agents cannot replicate: “You really can’t find an agent to find treasure.” The retailers most exposed to AI intermediaries are the ones whose value proposition is a searchable, comparable, price-transparent catalog. That is most of ecommerce.

What to watch between now and Cyber Monday

A short watchlist for the next ten weeks, drawn from the forecasts above:

  • Black Friday’s AI surge. Adobe expects Black Friday AI-referred traffic up 95% year over year and Black Friday spend growing faster than Cyber Monday ($12.9 billion, up 9.2%). Watch whether AI-referred conversion holds, or whether the 82% cart abandonment Salesforce measures industry-wide gets worse as agent traffic that cannot check out floods in.
  • The agent-shortlist wars. With 20% of holiday traffic projected from AI chat agents and 74% of shoppers trusting chat recommendations, whoever influences the agent’s shortlist controls a bigger share of the season than any banner ad ever did. Ask of every recommendation you receive: chosen on what evidence?
  • Returns in January. If 69% of AI-assisted buyers really are less likely to return items, the AI-assisted share of the season will be visible in reverse logistics data by late January. If returns instead spike on AI-recommended gifts, that will be the evidence-quality problem surfacing on balance sheets.
  • The review-adjacent trust vacuum. One in three ecommerce sites will have a brand-owned shopper agent by Cyber Week, per Salesforce, and 41% of shoppers using them will skip reading reviews. That is a transfer of the verification function from an open, contestable corpus to a closed, self-interested one, happening mid-season, with no standard for what the agent’s answer must rest on.

The evidence layer this season actually needs

Strip away the record topline and the 2026 holiday season is running a live experiment: can AI-mediated product advice earn the trust that used to be spread across a thousand reviews, and what happens if the inputs are rotten? The answer does not depend on better models. It depends on better evidence reaching the model before the recommendation is made.

That is the specific gap GoBuy exists to fill, and the holiday calendar makes it concrete. A Smart Score from 0 to 100 computed from review quality after fake and incentivized reviews are filtered out, rather than from raw star volume a seller can purchase, is the difference between a recommendation that deserves the 74% trust stat and one that merely benefits from it. The GoBuy Verified badge, earned by products holding a filtered score above 80 across 90 days, is precisely the kind of durable signal a gift purchase needs: it cannot be manufactured by a December burst of purchased praise. Showing only the top seven products per category instead of ten thousand contaminated listings matches how the new shopper actually decides: they were only ever going to consider a handful. And because 59% of AI-assisted purchases still land on Amazon, the verification has to meet the shopper there: the GoBuy Chrome extension injects the trust panel directly onto Amazon product pages, and for the agents themselves, one structured call to the GoBuy MCP server at gobuy.ai/api/mcp returns filtered review intelligence and a Smart Score before any money moves.

The forecasts agree on one thing: this is the first holiday season where the shopping journey is AI-mediated at scale, and it will not be the last. What is still undecided is whether the AI layer becomes the most trusted advisor in retail history or the most efficient distribution mechanism for manipulated product data ever built. The difference is not the models. It is whether anyone verifies the evidence underneath. Check what you are about to buy this season at gobuy.ai, and if you build shopping agents, wire them to the evidence layer at gobuy.ai/agent-docs.