AI-referred traffic to US retail sites is likely to grow more slowly this holiday season than Adobe’s own forecast implies. Signals point to Adobe’s season recap, expected in early January 2027, reporting November 1 to December 31 growth of roughly 90–115% year over year, short of the 130% it projected on September 28. The likely companion result is that the AI traffic which does arrive keeps converting at least 35% better than non-AI channels: fewer clicks than forecast, but better ones.
That is a narrower claim than it sounds. Nobody serious expects AI shopping to shrink, and a doubling of referral traffic in one season would still be a remarkable number for any channel. The prediction is about the gap between a click-based forecast and a market in which the AI assistants themselves are absorbing more of the research work that used to generate those clicks.
In short
- The prediction: Adobe’s 2026 holiday recap likely shows AI referral traffic to US retail sites up about 90–115% year over year, below its 130% forecast, with Cyber Monday growth likely below the 88% Adobe projected for that day.
- The companion call: the AI-referred conversion premium over non-AI traffic likely holds at 35% or more for the season, close to the 40% Adobe measured at the June Prime Day event.
- Signal 1: Adobe’s most recent AI-traffic forecast missed. It projected 103% growth for the June 2026 Prime Day event and measured 89%, while its September 28 holiday forecast was reissued on October 16 with every number unchanged.
- Signal 2: OpenAI’s October 1 shopping update (virtual try-on and saved Favorites) moves more of the evaluation step inside ChatGPT, which tends to remove exploratory click-outs before purchase.
- Signal 3: Profound’s tracking shows ChatGPT Shopping now pulls roughly 61–65% of product picks from merchant feeds carrying price and image data, while the number of distinct merchants referenced has fallen by about a fifth.
Why this matters now
The 2026 holiday season is the first in which AI assistants are a planned, budgeted channel rather than a curiosity. Adobe expects US online spending to reach $275.1 billion between November 1 and December 31, up 6.7%, with Cyber Week worth $47.5 billion and Cyber Monday alone $15.1 billion. Its forecast places AI traffic growth at 130% for the season, peaking on Thanksgiving (159%) and easing through Black Friday (95%) to Cyber Monday (88%).
Those percentages are not academic. Marketing teams set holiday targets for “AI visibility” programs off exactly these headline figures, agencies sell generative engine optimization retainers against them, and investors read them as a proxy for how quickly assistants are displacing search. If the click-based number undershoots while revenue holds, the wrong lesson would be that AI shopping stalled; the more likely reading is that it moved inside the assistant.
The calendar also matters. The season has been reshaped by earlier promotions, and the October Prime event now pulls demand forward; we argued in July that the 2026 holiday peak would flatten away from Black Friday. A flatter peak spreads AI research sessions across more weeks, which changes the year-over-year comparison on any single day.
Finally, this is a falsifiable call against a published benchmark. Adobe releases daily readings over Cyber Week and a season recap in January, so the outcome will be scoreable within roughly 75 days. That is a rare property for an AI-commerce prediction, most of which rest on survey intentions rather than measured behavior.
Signal 1: Adobe’s own AI-traffic forecasts have started to overshoot
Adobe has published AI-traffic forecasts before every major US shopping event since 2024, and the track record has recently changed direction. For the 2025 holiday season it projected 520% growth in AI referral traffic and measured 693%, a large beat driven by the first mass adoption of ChatGPT and Gemini for shopping research. For the four-day Prime Day event in June 2026, it projected 103% and measured 89%, its first notable miss on this metric.
The June miss is easy to overlook because total spending beat the forecast: Adobe recorded $26.4 billion against a projected $26.3 billion, up 9.3% year over year. In other words, the shopping happened, and AI-referred visitors converted 40% better than non-AI channels, according to Adobe’s final Prime Day figures. It was only the click count from assistants to retailer sites that fell short.
The holiday forecast arrived on September 28 and, on October 16, Adobe reissued it through Business Wire under a “correcting and replacing” notice. Per trade coverage of the reissue, the only change was a historical reference to when the October Prime event began; every forecast figure, including the 130% AI growth projection, was left untouched. A forecaster that had seen early October data running hot or cold could have revised at that point, and Adobe chose not to.
That non-revision cuts both ways. It may mean early October was tracking the model, or it may mean Adobe does not revise forecasts for small deviations. Either way, the record now shows the most recent comparable event (June) landing about 14 points under forecast, and the October Prime Day event, which opened on October 6 with Walmart and Target counter-programming, did not produce a published AI-traffic actual that would contradict the pattern.
| Event | Adobe AI-traffic forecast | Adobe measured | Gap | AI conversion vs non-AI |
|---|---|---|---|---|
| 2024 holiday season | Not comparable | About 1,300% | n/a | Below non-AI |
| 2025 holiday season | 520% | 693% | Beat by about 173 pts | 31% better |
| January to May 2026 | No forecast | 235% | n/a | About 42% better (Q1) |
| June 2026 Prime Day event | 103% | 89% | Missed by 14 pts | 40% better |
| 2026 holiday season | 130% | Pending (Jan 2027) | Predicted miss | Predicted 35% or better |
The table shows why one data point can matter. A forecaster extrapolating from a steep adoption curve tends to overshoot exactly when the curve begins to bend, and the June event is the first time the measured growth rate fell below the projection. The prior precedent points to a model calibrated on the 2025 surge carrying some of that momentum into a year where the base is far larger.
Why forecasters overshoot when an adoption curve bends
Forecasts of a young channel are usually built from recent growth rates applied to a seasonal shape. That method works while adoption is accelerating, and it tends to undershoot, which is what happened in 2025. Once the user base is large and usage per user begins to change, the same method tends to overshoot, because it carries forward a growth rate that the new base can no longer sustain.
The June 2026 result looks like the first sign of that transition. Adobe measured 235% growth for January to May, then 89% over the Prime Day event a few weeks later, a sharp step down within a single half-year. A holiday model anchored partly on the first-half rate would plausibly land above the eventual outcome, and 130% sits well above the most recent measured reading.
Signal 2: ChatGPT is pulling the evaluation step inside the chat
On October 1, OpenAI began rolling out two shopping features globally on web and mobile: virtual try-on for clothing and accessories, powered by its Images 2.5 model, and Favorites, which lets users save products to their ChatGPT Library and organize them into folders. A Try On button appears on eligible listings in shopping results, and users can upload a selfie or a full-body photo to see the item on themselves. OpenAI stores the reference photo for future sessions unless the user removes it.
Neither feature is a checkout. Both, however, replace work that shoppers previously did on retailer sites: opening product pages to look at imagery, comparing items across tabs, and returning later to a saved cart or wishlist. Every one of those steps used to register as an AI referral visit in Adobe’s click-based measurement, often several times per purchase.
The direction is consistent with what OpenAI said in March, when it stepped back from Instant Checkout. In a March 24 post, the company said merchants could use their own checkout experiences while ChatGPT focused on product discovery, and Walmart’s Daniel Danker told Modern Retail that in-chat purchases converted at rates three times lower than click-out purchases. The October update leans further into discovery and evaluation, which is the stage that generates the most exploratory, non-buying clicks.
The likely net effect is a shift in click composition rather than a collapse in volume. Shoppers who try on, shortlist and save inside ChatGPT are likely to click out later and less often, and with higher intent when they do. That is the mechanism behind the two-part prediction: growth in clicks undershoots, while the conversion premium on those clicks holds.
Other platforms are moving the same way. TikTok rolled out an AI shopping assistant with one-click checkout on October 5, and Google’s September 16 update gave Merchant Center users AI performance insights for AI Mode and AI Overviews. Each assistant that answers more of the shopper’s questions in place is, by construction, sending fewer speculative visits to the retailer’s product detail page.
Signal 3: feed-based product cards answer the question before the click
The third signal comes from Profound, which tracks how ChatGPT Shopping assembles its recommendations across a large set of monitored prompts. Its data, reported in early October, shows the share of product picks retrieved from integrated merchant feeds jumping from 8.26% to 61.54% on July 10, 2026, and reaching about 65% of tracked recommendations by September 3. OpenAI has not confirmed the change, and the data reflects monitored prompts rather than real shopper sessions.
Feed-retrieved products carry structured fields that page-crawled results often lack, including brand, price, availability and a checkout image. When a product card already displays the current price and stock status, the shopper no longer needs to click through to verify them. We covered which product feed fields decide inclusion in AI shopping agents earlier this year; the October data suggests those same fields now also decide whether a click is necessary at all.
Profound’s numbers also show concentration. References to the top 10 stores rose from 22.5% to 41.8% of recommendations, and the number of unique merchants referenced fell from 13,524 to 10,607, about 22% fewer. Among 687 tracked customers, 450 lost at least a third of their Shopping visibility in the shift, while 67 gained by the same margin.
Concentration matters for the forecast in two ways. Large retailers sit inside Adobe’s measurement panel, so a shift toward them may partly offset lost long-tail clicks; that is a genuine counter-signal. But concentration also means fewer comparison visits per session: a shopper offered three well-described products from familiar retailers clicks fewer tabs than one offered a dozen unfamiliar storefronts.
What the pattern suggests
Read together, the three signals describe a channel that is maturing rather than stalling. Adoption is still rising, which is why growth of 90% or more remains the base case. But the assistants are getting better at the work that used to generate clicks, and the year-over-year comparison is now against a season that was itself up 693%.
The base effect deserves its own line. Adobe’s January to May 2026 reading of 235% compared against early 2025, before the AI shopping surge; the June event at 89% compared against a mid-2025 base that had already grown sharply. The November to December 2026 comparison laps the largest base of all, the season in which AI referrals grew nearly eightfold, so the pattern suggests the growth rate is more likely to keep decelerating than to rebound to 130%.
| Signal | Source type | Date | What it shows | Direction for the 130% forecast |
|---|---|---|---|---|
| Adobe June forecast miss and unchanged holiday reissue | Analytics vendor data and press release | Sept 28 and Oct 16, 2026 (June actuals) | Latest comparable event 14 pts under forecast; holiday model not revised | Points to a miss |
| ChatGPT try-on and Favorites rollout | Platform product launch | Oct 1, 2026 | Evaluation and shortlisting move inside the assistant | Fewer exploratory click-outs |
| Feed-based ChatGPT Shopping picks | Third-party tracking data | Reported early Oct 2026 (data to Sept 3) | About 65% of picks carry structured price and image; 22% fewer merchants referenced | Fewer verification clicks, more concentrated |
| Base effect from the 2025 season | Adobe recap data | Jan 2026 recap | Comparison laps a 693% season | Mechanical drag on growth |
The scenario spread below reflects that reading. The base case is a miss in the 90–115% range; a narrow miss and a meet-or-beat outcome both remain plausible, and there is a real chance that Adobe changes its definition or does not publish a comparable season figure.
| Scenario | Season AI traffic growth (Adobe) | Conversion premium | Probability | What would drive it |
|---|---|---|---|---|
| Base case: fewer, better clicks | 90–115% | 35% or higher | About 55% | In-chat evaluation and feed cards absorb exploratory visits |
| Narrow miss | 115–129% | 30–40% | About 15% | Thanksgiving surge stronger than modeled, Cyber Monday soft |
| Meet or beat | 130% or higher | Any | About 20% | New user cohorts, Google cart transfer, branded agent launches |
| Unscoreable | No comparable figure published | n/a | About 10% | Definition change, panel change or metric dropped |
The day-level calls follow the same logic. Adobe’s forecast already assumes Cyber Monday growth (88%) well below Thanksgiving (159%), because Cyber Monday is the most intent-heavy day of the season. If assistants are absorbing research, the intent-heavy days are where the shortfall is likely to show first, so Cyber Monday growth below 88% is the most sensitive early checkpoint.
Why the conversion premium likely holds
The companion call rests on Adobe’s own conversion history. AI-referred visits converted 23% worse than non-AI channels at the 2025 Prime Day event, then 31% better across the 2025 holiday season, about 42% better in the first quarter of 2026 and 40% better at the June 2026 event. The trend has been steady improvement, and nothing in the October signals suggests a reversal.
Holiday seasons do narrow the gap somewhat, because deal-driven intent lifts conversion on every channel. In 2025, the AI premium reached 54% on Thanksgiving and 38% on Black Friday but averaged 31% for the full season. That is why the threshold for the companion call is 35% rather than the June level of 40%.
The click-composition argument pushes the other way. If exploratory visits (checking an image, verifying a price, comparing a third option) move inside the assistant, the visits that remain are disproportionately purchase visits. Removing low-intent sessions from the denominator lifts measured conversion even if the number of AI-assisted purchases is unchanged, so a traffic miss and a strong premium are likely to arrive together.
Wider context: agents, branded assistants and the measurement gap
Adobe is not the only firm publishing an AI forecast for this season, and the differences between the numbers are instructive. Salesforce predicted in July that 20% of all 2026 holiday e-commerce traffic would originate from AI chat agents, explicitly including autonomous agents running backend tasks and competitor scrapers feeding algorithmic price-matching. Adobe counts only human shoppers clicking a link from an AI service, a much smaller and cleaner population.
The two figures are not in conflict; they measure different things. As assistants do more on the shopper’s behalf, the machine-traffic share Salesforce describes is likely to rise while the human click-through share Adobe describes grows more slowly. Retailers reading both forecasts should expect their server logs to look busier than their analytics dashboards, and the gap is likely to widen through the season.
The checkout question has largely settled in the direction we expected. Our July analysis argued that agentic commerce would settle on retailer-controlled checkout by the 2026 holidays, and OpenAI’s March retreat from Instant Checkout plus Google’s September cart-transfer option for Universal Commerce Protocol merchants are consistent with that. Cart transfer is a counter-signal here, since it routes shoppers back to merchant sites and generates clicks Adobe can count.
Branded assistants complicate the picture further. Salesforce predicts that one in three e-commerce sites will run a site-specific shopper agent by Cyber Week, and retailers from Kohl’s (with an AI gift finder) to Gap have announced holiday AI features. Sessions with those assistants happen on the retailer’s own property and do not count as AI referral traffic at all, which is another reason the referral metric may understate how much AI shopping actually takes place.
There is also a consumer-side ceiling worth noting. Survey data points in different directions, from Zeta’s claim that more than eight in ten consumers plan to use AI this season to eMarketer’s citation of Similarweb data showing 37.1% planning to use it more. Wide variation in stated intent usually signals that heavy users have already adopted, and that incremental growth comes from lighter users who click less.
Implications for retailers, brands and platforms
For retailers, the practical point is to stop treating AI referral sessions as the KPI. If the base case holds, a retailer could see AI sessions grow by less than half of what an agency deck promised while AI-attributed revenue grows faster than sessions. The better scorecard pairs sessions with conversion rate, revenue per visit and the share of orders where an assistant appears anywhere in the path; our guide to measuring AI assistant referrals in GA4 and order data sets out how to build that view.
Feed quality is likely to matter more than page optimization this season. If roughly two-thirds of ChatGPT Shopping picks now come from merchant feeds, a retailer whose feed carries stale prices, missing images or vague availability is likely to be filtered out before any click occurs. The Profound concentration data suggests the penalty for a weak feed is not fewer clicks but disappearance from the consideration set.
For brands selling through retailers, the shift means more influence happens upstream of any measurable visit. Product imagery that renders well in virtual try-on, specification data that answers comparison questions and review content that assistants can summarize all shape the shortlist before a shopper reaches a product page. Brands that judge AI performance only on referral sessions are likely to undercount their own exposure.
For platforms and analytics vendors, the season is likely to expose the limits of click-based attribution. A metric that counts only link clicks from assistants will increasingly miss the evaluation work happening inside chat, and the pressure to add impression-style or assisted-conversion measures for AI surfaces is likely to grow after the January recap. Google’s new Merchant Center AI performance insights, which report share of voice on AI surfaces rather than clicks, point in that direction.
For investors, the risk is misreading a slower click growth number as a slowdown in AI commerce. The base case is consistent with assistants becoming more useful, not less, and with revenue per AI-referred visit continuing to rise. The more informative numbers in January will be conversion and revenue per visit, not raw traffic growth.
How to score this call
The prediction is designed to be checked against Adobe’s own published releases, with no interpretation required for the headline test. If Adobe’s season recap reports AI referral traffic growth below 130% for November 1 to December 31, the main call scores as a hit; within 90–115% it scores as a base-case hit. A figure of 130% or higher is a clear miss.
| Checkpoint | Expected date | What to look for | Base-case reading |
|---|---|---|---|
| Thanksgiving daily release | Nov 26–27, 2026 | AI traffic growth vs 159% forecast | Near or modestly below forecast |
| Black Friday daily release | Nov 27–28, 2026 | AI traffic growth vs 95% forecast | Below forecast |
| Cyber Monday and Cyber Week recap | Dec 1–2, 2026 | AI traffic growth vs 88% forecast; conversion premium | Below 88%; premium 35% or higher |
| Season recap | Early Jan 2027 | November-December AI traffic growth vs 130% forecast | 90–115% |
| First 2027 Adobe AI traffic report | Q1 2027 | Any change to the click-based definition | Definition unchanged; new assisted metrics likely discussed |
The companion call on conversion is scored the same way. If Adobe reports AI-referred traffic converting at least 35% better than non-AI traffic for the season, the “better clicks” half holds; a premium below 30% would suggest the click shortfall reflects weaker demand rather than more efficient research, and the thesis would need revisiting.
Caveats: what could go wrong
The strongest counter-signal is Adobe’s 2025 record. Its holiday forecast last year undershot the actual result by about 173 points, which shows how quickly a new cohort of users can overwhelm a model. If a large group of first-time AI shoppers arrives this season, perhaps pulled in by free-tier ChatGPT shopping features or Gemini integrations on Android, growth could clear 130% despite everything above.
Second, Google’s September changes push in the opposite direction from OpenAI’s. Cart transfer for UCP merchants sends shoppers from AI Mode and Gemini back to retailer sites, and Google’s reach on mobile search is large enough to add a meaningful number of countable clicks. If Google’s AI surfaces become the dominant holiday research tool, the click-based metric could look healthier than the ChatGPT-centric argument implies.
Third, the June miss may simply be noise. A four-day event is a small sample, and a 14-point miss on a forecast of 103% is within the range a forecaster could reasonably produce by chance. Adobe’s decision not to revise the holiday forecast on October 16 might reflect October data that was tracking the model well.
Fourth, the Profound data is a single vendor’s view of monitored prompts. OpenAI has not confirmed the retrieval change, and monitored prompts do not necessarily reflect what real shoppers saw. If the shift to feed-based cards is smaller in practice, the “verification click” mechanism would be weaker than described.
Finally, there is definitional risk. Adobe could change how it classifies AI referrals (for example, by adding agentic browsers or new assistants to the source list), which could lift the measured number independently of shopper behavior. That is why the scenario table assigns about 10% to an unscoreable outcome rather than pretending the benchmark is fixed.
FAQ
What exactly is being predicted?
That Adobe’s season recap for November 1 to December 31, 2026 likely reports AI referral traffic to US retail sites growing roughly 90–115% year over year, below its own 130% forecast. The companion call is that AI-referred traffic keeps converting at least 35% better than non-AI channels.
Does this mean AI shopping is slowing down?
Not in the base case. The argument is that assistants are doing more of the research inside the chat window, so each purchase generates fewer clicks to retailer sites. Growth of around 100% would still mean AI referrals roughly doubling in one season.
How does Adobe measure AI traffic?
Adobe counts visits to US retail sites that arrive when a shopper clicks a link from an AI-powered chat or browser service. Its holiday data draws on more than 1 trillion visits to US retail sites, 100 million SKUs and 18 product categories, according to the company. It does not count bot traffic or sessions with a retailer’s own on-site assistant.
Why would Cyber Monday be the first place a miss shows?
Cyber Monday is the most purchase-focused day of the season, so shoppers arriving from AI tools that day are most likely to have already done their research inside the assistant. Adobe’s forecast already assumes the slowest AI growth of the peak days (88%), and the pattern suggests the shortfall is most visible where intent is highest.
Didn’t Adobe beat its AI forecast last year?
Yes. It forecast 520% growth for the 2025 holiday season and measured 693%. That is the strongest argument against this prediction, but it came during the first mass-adoption wave, and the June 2026 Prime Day event (103% forecast, 89% actual) suggests the dynamic has changed as the base has grown.
Why does Salesforce say 20% of holiday traffic will come from AI agents?
Salesforce’s figure includes autonomous agents, backend task runners and competitor price scrapers, not just human shoppers following a link. Adobe’s metric covers human click-throughs only. The two numbers measure different populations and are likely to diverge further as assistants act more on shoppers’ behalf.
What should a retailer do differently because of this?
Judge AI performance on conversion rate, revenue per visit and assisted orders rather than on referral sessions alone. Prioritize merchant feed accuracy (price, availability, imagery), since feed-based product cards increasingly decide whether a product is shortlisted before any click happens.
What would prove this prediction wrong?
An Adobe season recap showing AI referral traffic up 130% or more would be a clear miss. A conversion premium below 30% would undermine the “better clicks” reasoning even if traffic missed, since it would suggest weaker demand rather than more efficient research.