AI ecommerce statistics 2026: traffic, conversion, and agentic commerce

AI-referred traffic to US retail sites grew 693% year over year during the 2025 holiday season, according to Adobe Analytics. It now converts above other channels in Adobe's US panel, though first-party academic data covering an earlier period disagrees. This page tracks both, with sources, samples and periods on every figure.

Consumer survey terms

The growth rates fall as the base grows. Adobe's own sequence runs from +4,700% in July 2025 to +1,200% in October and +693% across the holiday season, which is what a channel compounding off a small base looks like once the prior-year comparison stops being near zero. Adobe does not publish AI-referred traffic as a share of total retail visits, and this page does not estimate one.

Why the two results can both be right

1. 01 · Different comparators

   Adobe measures AI referrals against a blended baseline that contains paid social. Kaiser and Schulze compare ChatGPT referrals with each channel individually. Beating a blend that includes paid social is not the same test as beating organic search, and the two cannot be read as one.

2. 02 · Different periods

   Adobe's own earlier series agrees with the paper. AI traffic in the Adobe panel was 49% less likely to convert in January 2025, 38% less in April and 23% less in July, then crossed to +5% in September, +16% in October and +31% across the holiday season. The paper's twelve-month window largely predates the crossover.

3. 03 · Different data types

   A vendor analytics panel weighted toward large US enterprise retailers, against first-party transaction logs from 973 sites of mixed size.

4. 04 · Different populations and definitions

   Adobe is US retail only. The geography of the academic sample is not stated in the abstract and is pending confirmation from the full text; if it is European or mixed, the comparison with a US panel changes materially. Adobe counts AI referrals broadly across chat services and AI browsers, while the paper isolates organic ChatGPT referrals.

5. 05 · Selection effects

   Peak-season AI referrals arrive with gift intent and a deadline. Off-peak AI referrals skew toward research in complex categories, which is where the paper finds LLM outcomes strongest. The two datasets may be sampling different moments in the same funnel.

No source has yet measured both populations on the same definition over the same window. The direction of travel in the US panel data is upward and crossed zero in September 2025. Whether that holds outside peak season and outside the United States is unresolved, and this page will not resolve it by averaging the two.

Adobe survey, n = 5,000 US consumers, published August 2025

Adobe Holiday 2025 Consumer Survey, n = 1,000+ US respondents, published January 2026

The two Adobe surveys have different sample sizes and fielding dates. The 85% improved-experience figure (n = 5,000) and the 81% figure (n = 1,000+) are not a trend line.

Dated baseline: Capgemini Research Institute, fielded October to November 2024

Series sources: Salesforce, via Business Wire, 28 November 2023 · Salesforce, via Business Wire, 4 December 2024 · Salesforce, 5 December 2025

Salesforce forecast $73 billion and 22% for Cyber Week 2025 before the event and reported $67 billion and 20% after it. The forecast is not a result and is cited here only so the two are not confused. Salesforce, 2025

The peak-day conversion gaps are wider than the season average of +31%, which is consistent with the selection effect described in the conversion section: shoppers arriving from an AI service on Black Friday carry gift intent and a deadline. Adobe reports the season's total US online spend at $257.8 billion but does not publish AI-referred traffic as a share of it, and no share figure is estimated here.

Sources: ACP from OpenAI (24 March 2026) and Stripe (29 September 2025). UCP from Google's developer announcement (11 January 2026), its capability update (19 March 2026) and Merchant Center Help (retrieved 8 September 2026). AP2 from Google Cloud (16 September 2025) and Google (28 April 2026). A2A from the Linux Foundation (9 April 2026). MCP from Anthropic, the MCP project and the Linux Foundation (9 December 2025). MCP is included to show what it does not cover.

What happened to in-chat checkout

Stripe and OpenAI launched Instant Checkout in ChatGPT on 29 September 2025, US only, with US-based Etsy businesses at launch and Shopify merchants described as coming soon. The launch introduced the Shared Payment Token, scoped to a specific merchant and cart total. Stripe, 29 September 2025 On 24 March 2026 OpenAI wrote: "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery." OpenAI, 24 March 2026

The best available evidence on why comes from Walmart. Its EVP Daniel Danker told Wired, as reported by Modern Retail, that conversion rates were three times lower for selection sold directly inside the chatbot than for products requiring a click out to the merchant. Etsy told the same outlet it did not see large volume from Instant Checkout but does see valuable referral traffic. This is a named-executive disclosure from one retailer, graded C, and it is the only conversion comparison of in-chat checkout against redirect that has been made public. Modern Retail, 27 March 2026

Reported counts of merchants that went live on Instant Checkout vary across outlets and none has been confirmed by OpenAI or Shopify. [DATA NEEDED: confirmed live-merchant count]

What ACP is now

OpenAI describes ACP as "the connective layer between merchants and users throughout discovery," through which merchants share product feeds and promotions so their catalogs are represented in ChatGPT, with delivery paths that include third-party providers such as Salesforce and Stripe. OpenAI names Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot and Wayfair as integrated for discovery. Shopify merchants' product data is integrated through Shopify Catalog with no additional merchant work, and Shopify's Mani Fazeli is quoted saying buyers "can complete purchases on the merchant's online store through an in-app browser." Walmart has an in-ChatGPT app with account linking, loyalty and Walmart payments. The rollout covers free, Go, Plus and Pro users. OpenAI, 24 March 2026 The agentic commerce explainer covers the model in more depth.

What UCP is, and what is actually available

Google announced UCP at NRF on 11 January 2026 as an open-source standard co-developed with Shopify, Etsy, Wayfair, Target and Walmart, endorsed by more than 20 partners including Adyen, American Express, Best Buy, Flipkart, Macy's Inc, Mastercard, Stripe, The Home Depot, Visa and Zalando, and compatible with the Agent Payments Protocol, Agent2Agent and the Model Context Protocol as transports. Google Developers Blog, 11 January 2026 On 19 March 2026 Google added optional Cart and Catalog capabilities, the latter giving agents real-time pricing and inventory from retailer catalogs, and highlighted Identity Linking for loyalty benefits. Google, 19 March 2026

The specification and the checkout feature are different things. The specification is published and open. Google Merchant Center Help states that "the checkout feature enabled by UCP is available for select merchants at this time," that eligibility is limited to products in the United States, Canada and Australia, and that merchants express interest through a form. Retailers remain merchant of record. Google Merchant Center Help, Retrieved 8 September 2026

Where MCP fits

On 9 December 2025 Anthropic donated the Model Context Protocol to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. MCP joined goose and AGENTS.md as founding projects, and its maintainers retain autonomy over technical direction. At donation Anthropic reported 97 million+ monthly SDK downloads and 10,000 active servers. Anthropic, 9 December 2025 MCP is not a commerce protocol. It defines how a model reaches tools and data, which is why UCP lists it as one transport among several, and it does not define carts, checkout or payment credentials. The MCP server directory lists the ecommerce servers that exist on it.

The arc across both sponsors is the same: a native in-chat checkout was tried, it underperformed on the one public comparison, and both OpenAI and Google have converged on merchant-owned checkout with a protocol-mediated discovery and catalog layer in front of it.

1. 01 · Tag-based analytics reports human sessions by design

   Google Analytics 4 excludes traffic from known bots and spiders automatically, using Google research and the Interactive Advertising Bureau's International Spiders and Bots List, and the exclusion cannot be disabled or its volume seen. Any figure from a tag-based tool is therefore a human-session figure by construction. The requests that AI systems make to fetch product pages on a shopper's behalf, or to crawl a catalog, live only in server logs. Google Analytics Help, Retrieved 8 September 2026

2. 02 · User-agent strings are claims, not identity

   Google publishes a verification procedure because, in its words, site owners may be "concerned that spammers or other troublemakers are accessing your site while claiming to be from Google." The procedure is a reverse DNS lookup on the requesting IP, or a match against Google's published IP range files. Google Search Central, Retrieved 8 September 2026 OpenAI publishes separate IP address lists for OAI-SearchBot, GPTBot and ChatGPT-User for the same reason. OpenAI, Retrieved 8 September 2026 Allowlists and blocklists built on user-agent strings alone work from unverified identity data.

3. 03 · Attribution rules credit the last recorded touch

   Google Analytics 4 attribution models assign credit to touchpoints by rule or by a data-driven algorithm, and every model excludes direct visits from receiving credit unless the whole path is direct. Google Analytics Help, Retrieved 8 September 2026 A shopper who researches in an LLM without clicking through, then buys via a branded search or by typing the URL, leaves no AI touchpoint to credit. The purchase is attributed to search, or to whatever preceded the direct visit.

4. 04 · Some platforms expose no server logs at all

   The only unambiguous record of automated requests is the server or edge log, and not every hosted platform gives merchants one. Our server log monitoring guide documents what each major platform exposes; Shopify's online store, for example, offers no web server or edge log access, so merchants there cannot see crawler activity directly.

Ecommerce Guide reading: taken together, these mean most figures on this page probably understate AI influence rather than overstate it, because the research step that happens inside an assistant and never produces a click is invisible to every dataset cited here. The only measurement that captures automated activity with verified identity is server-side, matched against published IP ranges and confirmed with reverse DNS, which are the two methods Google documents.

Every statistic on this page links to the organization that produced the data, not to a report about it. Where a figure originates in a vendor's own release, earnings material, or a research paper, we link to that document. Where we have been unable to retrieve the underlying document, the figure has been removed rather than retained on trust. Each statistic carries its geography, the period measured, the sample or dataset size where disclosed, and the method used. Where two credible sources disagree, both are shown with their definitions, and we explain why they can differ rather than averaging them.

Published April 19, 2026. Last reviewed September 8, 2026, when every source below was opened and each figure checked against it. Any calculation performed by us is labeled "Ecommerce Guide calculation" or "Ecommerce Guide reading" in place. Maintained by the Ecommerce Guide editorial team; contact details are in the site footer.

Evidence grades

Grades describe the type of evidence, not the accuracy of the number. An A-grade panel can still be unrepresentative, and a C-grade company disclosure can be exactly right about the one company it describes.

A

Large first-party behavioral or transaction dataset: the Adobe Analytics panel, the Salesforce Shopping Index, the 973-site academic dataset.

B

Credible survey with disclosed methodology, sample and fielding period.

C

Company disclosure, earnings material, product documentation, or a limited vendor dataset (fewer than about 200 sites or brands).

D

Secondary report, unclear methodology, or a directional estimate or forecast.