AI Assistant vs Search Shoppers: Six Months of US and UK Store Data

Industry reports this year say shoppers who arrive from AI assistants convert better than shoppers from search: Shopify’s Q1 2026 data puts the difference on product pages at nearly 50%. We run the same jewelry brand in two markets, so we checked six months of our own Google Analytics 4 data, covering 17,077 AI assistant and organic search sessions on ifshe.com in the US and ifshe.co.uk in the UK.

The short version: AI assistant visitors did add to cart more often, in both markets and in every period. But for the same brand the gap ranged from 1.9x to 5.8x, and it moved mostly with our search traffic rather than with AI.

1. The findings

1. AI assistant visitors added to cart more often than search visitors in both markets and in every period we measured. Over six months that was 8.6 versus 2.6 add-to-cart events per 100 sessions in the US (3.3x) and 8.7 versus 2.5 in the UK (3.6x).
2. The same brand’s gap ranged from 1.9x to 5.8x. In the UK it was 5.8x from March to May and 1.9x from June to August. A single six-month number would have hidden that.
3. AI assistants send people to shopping pages; search sends them to articles. In the US, 85% of AI assistant sessions started on a product, collection or home page, and in the UK 75%. For search it was 30% and 33%; most search visits started on a blog article.
4. Page mix explained 55–70% of the gap while our search traffic was mostly informational. The rest was on-page behavior: counting only sessions that began on a product page, AI visitors added to cart 1.7x as often as search visitors in the US and 1.6x as often in the UK.
5. When our UK search traffic turned commercial, the gap fell to 1.9x. From June to August, 79% of UK search sessions started on a shopping page, close to AI’s 83%, and page mix explained only about 15% of the gap.
6. GA4’s AI Assistant channel only covers our data from June 2026. Before that, 74% of US and 76% of UK AI assistant sessions were recorded as Unassigned, not Referral, so a channel-group trend line shows a jump in June that is really a relabeling.

2. Methodology, and what this data is not

The data comes from Google Analytics 4 for our two largest storefronts, ifshe.com (US) and ifshe.co.uk (UK), covering March 1 to August 31, 2026, and retrieved on September 13, 2026. Both stores run on Shopify and sell the same brand, and 479 products are listed on both (each store lists about 800).

• AI assistant sessions: sessions whose source contains chatgpt, openai, perplexity, gemini, copilot or claude, whatever channel GA4 assigned them (section 7 explains why). ChatGPT accounted for 97.8% of them in the US and 98.4% in the UK.
• Search sessions: GA4’s Organic Search channel, excluding the AI sources above.
• Add-to-cart rate: add_to_cart events per 100 sessions. It counts events, so a session that adds two items counts twice; it is not the share of sessions that added something.
• Landing page type: the first page of the session, grouped by URL into product page, collection page, homepage, blog article and other. “Shopping pages” means product, collection and home pages together.
• Intervals: 95% intervals for the ratio of two add-to-cart rates, treating events as Poisson counts. Where an interval includes 1, we treat the difference as directional rather than as a finding.
• Splitting the gap: explained in section 5. Search page types with fewer than 30 sessions in a period use search’s overall rate for that period.

Sample sizes: 1,762 AI assistant and 6,155 search sessions in the US; 1,332 and 7,828 in the UK. GA4 reported no sampling, no thresholding and no data loss to an “(other)” row in any of the reports behind the tables and charts.

What this data is not. It comes from one brand in one category, personalized jewelry, so it is not a market benchmark. We do not publish conversion rates: orders from AI assistants in this window are too few to support a rate, so add-to-cart is the deepest step we report. These are channels as they reached our stores, not a controlled experiment. People who click through from an AI answer have already chosen to click, and “the same type of page” does not mean the same page. We also left out engagement rate, because it differed between our two GA4 properties in ways we could not attribute to shopper behavior.

3. AI visitors add to cart more often, in every period

Over the full six months, AI assistant visitors added to cart at almost the same rate in both markets: 8.6 events per 100 sessions in the US and 8.7 in the UK. Search visitors were close to each other too, at 2.6 and 2.5. On those totals the gap looks like a fixed property of AI traffic: 3.3x in the US and 3.6x in the UK.

Split the window in half and the gap stops looking fixed.

Every row favors AI assistant traffic, and every interval sits above 1. But within one brand the gap runs from 1.9x to 5.8x, and the UK figure moved the most. The ratio fell there because search rose much faster: UK search visitors went from 0.9 to 6.4 add-to-carts per 100 sessions, while AI visitors went from 5.4 to 12.1.

For scale: from June to August, AI assistants sent about 27 sessions to the US store and 30 to the UK store for every 100 that came from organic search.

4. AI sends people to product pages; search sends them to articles

The first reason for the gap is where each channel drops people off. A visitor who starts on a product page is one click from the cart. A visitor who starts on an article is not.

Six months, March to August 2026. Percentages are rounded, so columns may not add to exactly 100.

In the US, 67% of AI assistant sessions began on a product page, against 19% of search sessions, while 65% of search sessions began on a blog article. The UK showed the same pattern. That lines up with Shopify’s network-wide figure for Q1 2026: more than half of AI-referred sessions started on a product detail page, compared with about 20% for organic search.

AI assistant traffic was also more mobile: 83% of US AI sessions and 86% of UK ones came from phones, against 70% and 77% for search. The add-to-cart gap held on phones and on desktops in both markets.

5. Where they land vs. what they do there

If landing pages were the whole story, AI and search visitors who start on the same type of page would add to cart at the same rate. To separate the two effects, we asked a simple question: how often would search visitors add to cart if they started on the same mix of pages as AI visitors, at search visitors’ own rate for each page type? The step from search’s actual rate to that figure is the landing effect. The step from that figure to AI’s actual rate is the on-page effect. Multiplied together, the two give the total gap.

Effects are rounded to one decimal; before rounding they multiply exactly to the total gap. The share of the gap is measured on that multiplicative scale.

The on-page effect also shows up without any modeling. Counting only sessions that began on a product page, over six months AI visitors added to cart 9.0 times per 100 sessions in the US against 5.4 for search (1.7x, interval 1.2–2.3), and 10.2 against 6.3 in the UK (1.6x, interval 1.1–2.3). Shopify reported the same direction for orders on product detail pages. Our measure is add-to-cart rather than orders, so the two figures are not directly comparable.

6. What happened when our UK search traffic changed

From March to May, 73% of UK search sessions (4,125 of them) landed on a cluster of “angel number” meaning articles: spiritual content with no connection to jewelry shopping. Those sessions produced no add-to-carts at all. With 82% of UK search sessions starting on a blog article, search visitors added to cart 0.9 times per 100 sessions, and the AI gap was 5.8x.

By June those articles were drawing almost no search visits, and we have since removed them. From June to August, 79% of UK search sessions started on a shopping page, close to AI’s 83%. Search add-to-carts rose to 6.4 per 100 sessions, the gap fell to 1.9x, and page mix explained only about 15% of that gap.

The US is the contrast case. Its search traffic stayed informational for all six months, with 68% and then 64% of search sessions starting on a blog article, and its gap stayed between 3.0x and 3.9x.

For anyone quoting an “AI traffic converts N times better” figure: in our data, the size of N depended more on the search baseline than on the AI traffic. Our two markets suggest that a store whose search visitors land on articles will see a large multiplier, and a store whose search visitors land on products will see a small one.

7. A measurement trap: GA4’s AI Assistant channel

Google Analytics 4 now has an AI Assistant default channel. Google’s channel definitions say GA4 sets the medium to “ai-assistant” when the referrer matches its list of AI assistants. Coverage of the launch, such as Search Engine Journal’s report, described AI chatbot traffic as previously landing in Referral. In our two properties, most of it was not in Referral.

From March to May, 74% of US and 76% of UK AI assistant sessions were recorded as Unassigned, which is GA4’s label when no channel rule matches, and the rest as Referral. Of those Unassigned sessions, 93% in the US and 98% in the UK came from chatgpt.com with no medium recorded. The AI Assistant channel first appears in June (59% of US and 57% of UK sessions) and holds 95–100% of them in July and August. It was not applied to earlier months.

That has two consequences if you track AI traffic by channel group. A chart of the AI Assistant channel will show AI traffic appearing from nothing in June 2026. And a count of AI sessions in Referral before June will miss most of them, at least in properties like ours. Classifying by session source, as this page does, gives a continuous series.

8. What to take from this

• Compare like with like. Before reporting that AI traffic outperforms search, split both channels by landing page type. In three of our four market-periods, most of the headline gap was page mix.
• There is still a real on-page difference. On product pages, AI visitors added to cart about 1.6–1.7x as often as search visitors in both markets. That is a more useful planning number than the headline.
• Product pages are the front door for AI traffic. 67% of US and 51% of UK AI assistant sessions started on one, so a product page is often the first thing an AI-referred shopper reads.
• Informational content moves your baseline. Articles that attract search visits with no buying intent inflate the apparent AI multiplier without changing anything about AI traffic.
• Track AI traffic by source, not by channel group, if your trend needs to start before June 2026.

9. Using and citing this data

All figures on this page, including the tables and charts, are free to reproduce with attribution to IfShe and a link to this page. No permission needed. Every finding and section has its own anchor link if you want to point at a specific number.

Suggested citation
IfShe (2026). AI Assistant vs Search Shoppers: US and UK Store Data. First-party Google Analytics 4 data, March 1 – August 31, 2026. https://www.ifshe.com/pages/ai-assistant-vs-search-shopper-data

Media inquiries and data requests. If you are working on a story and need a cut we have not published, such as a different page type, monthly detail or how these numbers change after August, write to [email protected] and we will pull it. We publish what we find, including results that do not flatter us. For search demand rather than on-site behavior, see our companion page, US vs UK personalized jewelry search data.

Data window March 1 – August 31, 2026, retrieved September 13, 2026. Published by IfShe, personalized jewelry since 2013.