Research · November 22, 2026
The rise of agentic commerce
Shoppers are handing research, and sometimes the purchase itself, to AI agents. The traffic is still small, but it's growing fast, it converts better than other visits, and the platforms are racing to own it.
AI traffic to US retail sites, holiday 2025 against holiday 2024.
Adobe, Jan 2026conversion for AI-referred visits than other traffic, March 2026.
Adobe, Apr 2026AI bot visits per human visit on publisher sites, Q4 2025, up from 1 in 200 a year earlier.
TollBit, Apr 2026US online spending by agentic shoppers forecast for 2030.
Morgan Stanley, Dec 2025Shoppers already use it
39% of US consumers have used AI to shop, in Adobe's survey of more than 5,000 (Adobe), and 47% of US shoppers use an AI tool for at least one shopping task (Visa and Morning Consult). Amazon says 250 million customers used its Rufus assistant in 2025, and that those users were 60% more likely to buy (Q3 2025 earnings call). Similarweb counts 770.7 million AI referral visits a month worldwide, up 117.4% in a year (Similarweb).
AI shoppers went from browsing to buying
In early 2025, a shopper who arrived from an AI assistant was much less likely to buy than other visitors. By the 2025 holidays that had flipped, and by March 2026 AI-referred visits converted 42% better.
Conversion rate of AI-referred visits to US retail sites, compared with all other traffic. Adobe Analytics: Aug 2025, Jan 2026, Apr 2026.
Agents read far more than they click
On publisher sites in TollBit's network, AI bot visits went from about 1 in 200 visits at the start of 2025 to 1 for every 31 human visits by the end of the year. Over the same stretch, click-through from AI apps to publishers fell from 0.8% to 0.27%. Agents are reading the pages; the people behind them mostly aren't arriving.
AI bot visits per 1,000 visits on TollBit publisher sites (1 in 200, 1 in 50, 1 in 31). TollBit's earlier quarters count against all visitors and Q4 2025 against human visits. TollBit Q2 2025, TollBit Apr 2026.
That gap is the problem for anyone who sells ads to readers. The agent reads the article text and leaves the banners and sidebars behind, so the visits grow while the ad revenue doesn't.
The money that's coming
Forecasts of shopping done by or through AI agents. Morgan Stanley (US e-commerce by agentic shoppers, 2030); McKinsey (US retail revenue orchestrated by agents, 2030; $3T–$5T globally); eMarketer (e-commerce driven by AI platforms, 2029). The three measure different things.
The platforms are building the pipes
OpenAI's scale-back and the ads launch are sourced to news reports; the rest to the companies' own announcements.
What agents cite when you ask them to shop
Engines disagree. On e-commerce prompts, ChatGPT sent 36% of its citations to retailers while Google's AI Overviews sent 4%, leaning on YouTube, Reddit and Quora instead (BrightEdge). In our own 100-question study, Claude's search leaned almost entirely on independent review sites, with Tom's Guide, RTINGS and TechRadar cited in more answers than any others. And Adobe scores the average retail product page at 66% readable by AI (Adobe), so a third of what's on the page doesn't reach the agent at all.
What it means
- For publishers: agent visits are growing and the shoppers who do click through convert well, but the ad formats on the page were built for human eyes. The revenue follows whichever format the agent actually reads.
- For brands: the agent's shortlist is built from a handful of sources per category. Being in the text of those pages matters more than being on the page.
- For everyone: the engines now sell ads beside their answers. The pages they read to write those answers are a separate, open surface.
Figures are quoted as published, with the source linked beside each. Surveys and forecasts use their own methods and definitions, so they aren't directly comparable.