Short answer: consumers rely on AI for shopping because it does the research step for them. A shopper describes a situation, a budget or a constraint, gets a short list with reasons, and refines it in the same conversation instead of opening a dozen tabs. Adobe's survey of 5,000 U.S. consumers found that 39 percent have used generative AI for online shopping, mostly for research and product recommendations. The growth is fast but starts from a small base, and the assistants themselves are now being wired directly into product catalogues and checkout.
This article sticks to what can be traced to a primary source: Adobe's own analytics reports, Shopify's and Amazon's own announcements, Perplexity's own blog and OpenAI's merchant documentation. Where the evidence runs out, it says so.
What the measured data actually shows
The best public measurement of AI shopping traffic comes from Adobe Analytics, which counts visits to U.S. retail sites where a shopper clicked a link from a generative AI service. Shopify publishes a comparable figure for its own merchants. Read together, they show steep growth over two holiday seasons and into 2026.
| Source | Period | What was measured | Reported change |
|---|---|---|---|
| Adobe Analytics | Nov. 1 to Dec. 31, 2024 | Generative AI traffic to U.S. retail sites, year over year | Up 1,300 percent (1,950 percent on Cyber Monday) |
| Adobe Analytics | February 2025 vs July 2024 | Generative AI traffic to U.S. retail sites | Up 1,200 percent |
| Adobe Analytics | Nov. 1 to Dec. 31, 2025 | Generative AI traffic to retail sites, year over year | Up 693.4% (670% on Cyber Monday) |
| Shopify | Q1 2026 | AI-driven traffic and orders at Shopify stores, year over year | Traffic grew 8 times; orders from AI-powered searches nearly 13 times |
Two caveats belong next to these numbers. First, they are growth rates on a small base. Adobe says plainly that generative AI traffic remains modest compared with channels such as paid search or email, even while it was doubling every two months from September 2024. Second, Adobe counts clicks. A shopper who reads an assistant's answer and then types the brand name into a search bar, or buys inside the assistant, does not show up as AI traffic in that count. The real influence of AI answers on purchases is therefore not measured by either figure, in either direction.
Set side by side, the two Adobe holiday figures show growth that is still very fast but slowing in percentage terms as the base grows. That is the normal shape of a young channel, and a reason to plan for it as a growing share of research rather than as a sudden replacement for search.
Why consumers rely on AI for shopping: the research step, compressed
Adobe's survey asked what people actually use generative AI for when they shop. The answers point at research, not at buying:
- Conducting research: 55 percent of respondents
- Receiving product recommendations: 47 percent
- Seeking deals: 43 percent
- Getting present ideas and finding unique products: 35 percent each
- Creating shopping lists: 33 percent
Respondents also said an AI chat interface shortens the time it takes to get information personalised to their needs. Of those who had used AI for shopping, 92 percent said it enhanced their experience, and 87 percent said they are more likely to use it for larger or more complex purchases. That last figure is the important one. The more options and specifications a purchase involves, the more a shopper gains from handing the comparison to an assistant.
Adobe's conversion data tells the same story from the retailer's side. In its early 2025 report, traffic from generative AI sources was 9 percent less likely to convert than other traffic, down from a gap of 43 percent in July 2024. Adobe reads the gap as AI being used in the research and consideration stage, before the shopper is ready to buy. Conversion was highest in electronics and jewellery and lowest in apparel, home goods and grocery, which fits the same pattern: AI helps most when a shopper is narrowing options by screen size, resolution and price, and least when the choice is about taste or routine.
There is also a structural reason the conversation format works. Shoppers do not naturally think in keywords; they think in situations. A question that carries a budget, a constraint or a use case produces a different set of named brands than the bare category term, which is why an assistant that accepts the whole situation, and the follow-up question after it, feels closer to asking a knowledgeable person than to running a search.
The assistants are turning into shops
The second driver is supply. Over the same period, the major assistants and commerce platforms built shopping directly into the conversation, which removes the step where the shopper had to leave and search again.
- Amazon. Amazon describes Rufus as a shopping assistant trained on its product catalogue, customer reviews, community Q&As and information from across the web. On May 13, 2026 it was renamed Alexa for Shopping.
- Perplexity. On Nov 18, 2024 Perplexity launched Buy with Pro, which allows checkout inside Perplexity for select products from select merchants, together with Snap to Shop for searching by photo and a free merchant program in which retailers share product data. Perplexity states that its product cards are not sponsored.
- ChatGPT. OpenAI publishes a product feed specification so merchants can send structured product data that ChatGPT indexes and displays. The required fields are item ID, title, description, URL, brand, seller name, image URL, availability and price, and OpenAI notes that being eligible does not guarantee being displayed.
- Shopify. Shopify's Agentic Storefronts is automatically activated for eligible merchants and syndicates titles, descriptions, images, pricing, inventory and shipping through Shopify Catalog to ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app. Purchasing across these channels is currently available to U.S. buyers only.
The result is that the path from an answer to a basket keeps getting shorter. Each of these is a reason for consumers to stay inside the assistant, and each also changes what a brand has to supply: not just pages that can be crawled, but product data that arrives in a structured, current form.
What the evidence does not show
Several claims that circulate about AI shopping could not be traced to a primary source, and you should be careful with them wherever you see them:
- That shoppers trust AI more than reviews. We found no primary survey that supports a specific figure. What is documented is that reviews feed the answers: Amazon names customer reviews as one of Rufus's training sources.
- That a fixed share of product discovery already happens in AI. Adobe and Shopify measure traffic and orders in their own datasets. Neither publishes a share of all product discovery.
- That AI visitors always convert better. Adobe's early 2025 data says the opposite, although the gap was closing quickly.
The honest summary is narrower but still significant: a growing minority of shoppers use AI for the research step, especially for complex purchases, and the platforms are making it easier every quarter.
What consumers relying on AI for shopping means for a brand
If the research step moves into an assistant, the question for a brand is whether it is named in that step. Product categories make this harder than other segments, because answers consolidate. In our own Q3 2026 benchmark of 1,188 answers across 12 categories and five models, the three categories with a clear default answer were all product categories: project management software at 38% of answers, skincare at 35% and electric bikes at 28%, while every local-service category had a leader below 13%. The full data is in the AI recommendation benchmark.
The same benchmark showed that in product categories a large share of citations went to marketplaces, review sites and publisher round-ups rather than to brand sites. So the work has four parts:
- Make sure the assistants can read your product pages. OpenAI's crawler documentation separates OAI-SearchBot, which builds the index behind ChatGPT search, from GPTBot, which collects training data. Blocking the wrong one removes you from search answers. Our AI crawler access guide lists every documented agent.
- Write product facts as sentences. Assistants lift sentences that answer a constraint: a dimension, a material, a return window, a compatibility. Category copy written for brand tone gives them nothing to lift.
- Keep structured product data and third-party listings current. Check that the marketplaces and review sites cited in your category carry your current facts, prices and availability, and that any feed you send matches the live page.
- Measure with questions that do not name you. A question containing your brand name will be answered with your brand. In one of our own runs, 13 of 15 generated questions contained the brand name and the score read 13% when the honest value was zero. Building a prompt set explains how to avoid that.
The practical, step-by-step version of this, including Shopify and WordPress specifics, is in our companion post on how e-commerce brands can boost AI shopping recommendation exposure. If you want to see which questions in your category name a competitor and not you, that is what the InsightWonder e-commerce solution measures.
Sources
- Adobe, traffic to U.S. retail websites from generative AI sources, March 17, 2025 (survey of 5,000 consumers, traffic and conversion data).
- Adobe, 2025 holiday shopping season release, January 7, 2026.
- Shopify, how agentic commerce works, June 18, 2026.
- Amazon, Rufus, now Alexa for Shopping.
- Perplexity, Shop like a Pro, November 18, 2024.
- OpenAI, product feed specification and crawler documentation.
- InsightWonder Q3 2026 AI recommendation benchmark and product incident records (first-party).
An earlier version of this article cited survey and forecast figures attributed to Nielsen, McKinsey, Gartner and Statista, charts with no traceable source and claims about how much individual assistants had changed purchasing. We could not verify any of them at a primary source, so they were removed. It also misdescribed Adobe's 1,300 percent figure as a Cyber Weekend number; Adobe reports it for the whole November to December 2024 season. Every number that remains links to the publisher's own page.