What 96 Million Sessions Show About AI Referral Traffic and Conversions

Over the past year, AI assistants sent the online stores we manage about the same number of visits, which turned into almost three times as many orders.

To find out what AI actually sends, which AI assistants bring traffic, and why orders went up while the traffic stayed the same, we analyzed 96 million sessions and 977,000 orders across 73 stores on Shopify, Magento, BigCommerce and WooCommerce.

Key takeaways:

• Across 96 million sessions and 977,000 orders, visits from AI assistants were almost unchanged year over year (896 a day in September 2025, 917 in September 2026), while orders from those visits rose from 5.9 to 16.1 a day.
• Stores where at least half of AI visits start on a product page convert that traffic at 1.62%, against 1.06% for the rest.
• Of the 54 stores that recorded an AI order in either September, 45 recorded more this year and 9 recorded fewer.
• The biggest change was in August 2026, when AI orders went from 8.0 a day in July to 14.7 a day.
• On 41 of the 61 stores with at least 200 AI visits this year, AI visitors converted better than Google organic visitors. The median store saw a 25% higher conversion rate and 23% more revenue per visit, with orders 15% smaller.
• The conversion rate went up for every AI assistant and on 28 of the 35 stores with enough AI visits in both Septembers. Engagement stayed flat at about 60%.
• ChatGPT sent 88% of AI visits, and half of all AI visits started on a product page.

How the AI referral traffic was measured

We looked at the GA4 data of 91+ online stores Shero manages or supports. 73 of them had at least 100 orders and 20,000 sessions in 2026, which is enough to measure AI orders, so those 73 are the stores in this study.

Before calculating anything, test, staging and duplicate properties were removed. When one GA4 property tracks several brand storefronts, each storefront was counted as its own store. When a brand also has its own property, its numbers come from that property so no order is counted twice.

We also left out properties that are not online stores. When a brand tracks its content site and its online store in one property, only the store was counted.

An AI referral is a session that came from ChatGPT, Gemini, Perplexity, Copilot, Claude or another AI assistant, and an order is a GA4 purchase. The data covers September 1, 2025 to September 26, 2026. September 2026 only has 26 days, so every month is compared on visits and orders per day.

These are all stores we work with, so there is some selection bias, and this is not a random sample of eCommerce stores. Results are shared as totals and medians, and no store is named.

AI visits stayed flat and AI orders almost tripled

AI traffic to these stores went up and down during the year but never really grew. It dropped to between 640 and 680 visits a day through the winter and spring, then came back to about 900 a day by September, right where it started.

Orders went up in two steps. The first was in November 2025, when the AI conversion rate went from 0.63% to 0.95%, and it stayed above 0.9% after the holidays were over. The second came in August 2026, when AI orders went from 8.0 a day in July to 14.7 a day and the conversion rate went from 1.24% to 1.71%.

Orders from AI assistants rose to 2.74 times their September 2025 level, while visits ended the year close to where they started.

The August jump was not only our stores. BrightEdge also reported that ChatGPT referral traffic jumped 36% in August across its own data, its biggest month of the year.

BrightEdge's release on ChatGPT referral traffic in August 2026.

Big retailers tell a different story. Adobe's analysis of more than 1 trillion US retail visits put AI traffic up 393% year over year in the first quarter of 2026. The mid market and enterprise stores we work with saw nothing close to that in visits, and their gains came from conversion.

Adobe's first quarter numbers, as reported by TechCrunch. Large retailers saw AI visits go up, but the stores in our study did not.

Why AI visitors buy more than they did a year ago

The consideration stage of the buying journey now happens inside the assistant, so the people who click through are much closer to buying. GA4 can't prove that on its own, but our data indicates that in three places.

1. Every assistant converts better. ChatGPT's conversion rate went from 0.69% in September 2025 to 1.77% in September 2026, and the other assistants combined went from 0.34% to 1.38%. When ChatGPT, Gemini, Perplexity, Copilot and Claude all show the same shift in conversion, the change is in how people shop with AI.
2. It happened on most stores, not just a few big ones. In a matched cohort of 35 stores with at least 100 AI visits in both Septembers, 28 converted AI traffic at a higher rate this year. 45 stores added AI orders in total, although the top 10 account for about 70% of the increase.
3. Visitors are not browsing more. The engagement rate of AI visits stayed at about 60% all year. They spend about the same time on the site once they land and buy more often, which is what you would expect from someone who picked the product before clicking.

The timing lines up with new features OpenAI launched. Instant Checkout went live on September 29, 2025, and shopping research, which builds a buyer's guide from a few questions about budget and needs, followed on November 24, 2025. Our first step up came in November. The holiday season is a confounding factor, so I can't pin it on a feature launch, but the rate stayed up in January and February. I couldn't tie the August jump to a single launch.

The monthly AI conversion rate from the dataset. The first increase came in November, and the second in August.

I think a few things are going on, from most to least likely:

1. The AI assistant does the comparing. ChatGPT now answers a buying question with product cards, prices, reviews and a link to the product page, like the example below. By the time someone clicks, the decision is mostly made.
2. More people ask it what to buy. A year ago a lot of AI visits came from general questions. More people now go to ChatGPT with a purchase in mind, the way they used to go to Google. Yet, we can't measure this hypothesis directly.
3. AI assistants link to better pages. Half of AI visits now start on a product page, the page type that converts them best. We only have the landing page split for 2026, so we can't say how much that share grew.

I asked ChatGPT which waterproof hiking boot to buy for wide feet. It named one boot, showed the price, and linked straight to a retailer's product page, and not their homepage.

Tracking does not explain this shift either. If GA4 had simply started showing more AI visits, visits would have gone up, but they stayed approximately the same YoY. Orders placed with Instant Checkout happen inside ChatGPT and usually never reach GA4, so if anything these numbers undercount.

Two stores that show what changed

Two Shopify stores with enough AI traffic show the change clearly at store level. We describe them without names or exact figures to keep them anonymous.

The first is a Shopify store where AI visits stayed roughly flat from last September to this one. Orders from those visits increased more than seven times over the same year, and most of its AI visits start on a product page.

The second, also on Shopify, is the growth case. Its AI visits roughly doubled and its AI orders more than tripled. Their AI visitors convert at close to twice the rate of their Google organic visitors. Most AI visits started on a product page as well.

The same is true beyond these two stores. On stores where at least half of AI visits start on a product page, AI traffic converts at a median 1.62%, against 1.06% on stores where it does not. That comparison covers the 49 stores with at least 500 AI visits this year.

AI visitors convert better than Google organic visitors

On 41 of the 61 stores with at least 200 AI visits in 2026, AI visitors converted at a higher rate than visitors from Google organic search. On the median store, AI visitors converted at 1.27% against 1.09% for organic.

Measured store by store, the median AI conversion rate was 25% higher than organic, and revenue per visit was 23% higher. However, orders from AI visitors were 15% smaller on the median store, and their engagement rate was 5% lower.

The median store, AI visitors against Google organic visitors.

Other studies confirm the same. For example, 180 Marketing's analysis of 94 eCommerce stores in 2025 found ChatGPT traffic converting at 1.81% against 1.39% for non-branded organic search. Adobe found AI traffic converting 42% better than non-AI traffic in March 2026, a year after it converted 38% worse.

180 Marketing compared ChatGPT traffic with non-branded organic search only, so their gap is a cleaner comparison than ours.

Additionally, our organic numbers include branded searches, which usually convert higher than non-branded ones, so this comparison is harder on AI traffic than 180 Marketing's. Still, AI visits brought in 23% more revenue per visit than organic on the median store.

ChatGPT sends 88% of AI referral traffic

ChatGPT sent 88.1% of AI assistant visits to these stores in 2026 and sent visits to all 73 stores. It also converted better than every other assistant.

BrightEdge measured ChatGPT at 95.1% of AI referrals in August, so its lead is consistent across very different data sets. Gemini, Perplexity, Copilot and Claude together sent 11.9% of AI visits to our stores, and each of them converted below ChatGPT.

If you are deciding where to start with AEO, I would start with ChatGPT. Prompt it about your brand or products to see what it says, and which of your pages it links to.

Half of AI visits start on a product page

51.0% of AI visits to these stores started on a product page in 2026, and those visits converted at 1.77%. Homepage visits made up 22.8% and converted at 0.57%.

Product pages receive half of all AI visits and convert three times better than the homepage.

Blog posts and guides started 5.1% of AI visits and converted at 0.29%. Collection pages were small at 4.2% of visits but converted almost as well as product pages, at 1.78%.

The description, specifications, price and availability on a product page are what an AI assistant reads before it recommends the product, so the product page is where AI visits turn into orders. That is why I focused our citation study and these product descriptions on product content.

Blog content still earns citations and brand mentions, which you can measure with AI visibility tracking. Across these stores, blog posts and guides got 5.1% of AI visits and about 1% of AI orders.

How to convert more of your AI visits into orders

AI visitors arrive with a short list and a question already answered, so the page only needs to confirm the choice and make buying easy. These are the five changes I would tackle first, and each one comes from the numbers above.

1. Make the product page answer the question the shopper asked the assistant. Sizing, compatibility, materials, what is in the box, price, stock, delivery time and the returns policy belong in plain text on the page, where both the assistant and the shopper can read them. Product pages are also the page type you control most directly.
2. Give assistants a better page than your homepage. A homepage visit converts at a third of the product page rate. Collection pages convert almost as well as product pages, so collections built around a real need, use case, or intent, give the assistant a better page to link to.
3. Link a product or category on every tool, guide and info page. These pages draw AI visits because they answer a question well, but blog and guide visits rarely turn into orders. A clear link to the products each page is about turns traffic that already arrives into orders.
4. Raise the cart size on the pages AI sends people to. AI orders were 15% smaller than organic orders on the median store. Bundles, frequently bought together blocks and a free shipping threshold shown on the product page give a decided buyer a reason to add a second item.
5. Remove the friction between the landing page and the checkout. A shopper who arrives from ChatGPT ready to buy should see the add to cart and express checkout or buy now buttons first, before any email popup. I would test turning off entry popups for AI traffic and watch the conversion rate on those sessions.

After that, I would check that AI crawlers can fetch your store at all, and plan for agentic commerce, where the assistant completes the order itself and the product page may never be seen.

How to track AI traffic in GA4

GA4 files most AI visits under Referral, mixed in with every other website that links to you, so the first thing I'd set up is a channel of their own for AI assistants. GA4 applies custom channel groups to your historical data as well, so you do not have to wait a month to see numbers.

Set up a GA4 channel for AI assistants

1. In GA4, open Admin, then Data display, then Channel groups, and click Create new channel group.
2. Name the group, click Add new channel and call it AI assistants. Set the condition to Source matches regex and paste the expression below.

.*(chatgpt|openai|perplexity|gemini\.google|bard\.google|copilot|edgeservices|claude\.ai|anthropic|meta\.ai|deepseek|grok|you\.com|poe\.com|mistral).*

Our channel still uses an older list of exact domains. The expression above covers more assistants.

Keep the .* at both ends. GA4's matches regex has to match the whole source name, so without them a source like chatgpt.com does not match and your ChatGPT visits stay in Referral. I ran into this while building the exploration below: without the wildcards, only claude.ai came through.

3. Use Reorder to move AI assistants above Referral, then save the group. GA4 gives each session to the first channel it matches, so an AI channel sitting below Referral stays empty.

The AI channel sits first in the list, so GA4 checks it before Referral.

Build the AI exploration

GA4 does not let you share an exploration outside your own property, so you need to create this one yourself. It takes six steps.

1. Open Explore and start a Blank exploration. Name it AI referral traffic.
2. Set the date range to Last 12 months.
3. Next to Dimensions, click the plus sign and import Session source and Landing page.
4. Next to Metrics, import Sessions, Transactions and Purchase revenue.
5. Double click Session source and Landing page to add them as rows, then double click the three metrics to add them as values.
6. Under Filters, pick Session source, set it to matches regex, paste the same expression and click Apply.

One setting to check first: explorations only go back as far as GA4 keeps event data, and the default is two months. I hit this on one of the stores in the study. Under Admin, then Data collection and modification, then Data retention, set it to 14 months so next year's comparison is possible.

The finished exploration on one of the stores in the study. I blurred the store details and left the orders column, which shows ChatGPT visits turning into orders.

This is the same expression I used for the study, so your numbers will be comparable to mine. Our guide to track AI traffic covers the setup in more detail.

Or use our Looker Studio report

If you would rather skip the setup, open the AI Assistant Looker Studio template I built for this study. It compares AI assistants with organic search and everything else, breaks AI traffic down by assistant and by landing page, and uses the same expression as this study.

1. Open the report and click the data box at the top left to pick your own GA4 property. You only see the properties your Google account can access, and nobody else sees your numbers.
2. The date range starts at the last 12 months. Change it from the box at the top right.

Pick your property on the left and the date range on the right. The values here are from our own site, blurred. Further down, the report breaks AI traffic out by assistant and by landing page.

3. To keep your own version, open the three dot menu at the top right and choose Make a copy.

I would report AI orders and revenue per visit every month next to organic search, because traffic share alone would have told these stores that nothing changed in twelve months. If you pay for an AI visibility tool, check its scores against these numbers, because if the score goes up and AI orders don't, the score is not measuring anything that makes you money.

Why the real AI numbers are higher than GA4 shows

GA4 only counts an AI visit when the assistant passes a referrer, and part of AI-driven shopping never does. So every AI number in this study is the minimum, and the real figure for these stores is higher. The table shows where the rest goes.

So the 0.21% share of sessions understates what AI assistants do for these stores. The conversion comparisons are less affected, because they compare visits GA4 did attribute to AI with visits it attributed to Google.

Earlier this year, Rand Fishkin of SparkToro tested this with Gumshoe. 600 volunteers ran the same 12 prompts through ChatGPT, Claude and Google's AI 2,961 times. ChatGPT and Google's AI had less than a 1 in 100 chance of giving the same list of brands twice, and seeing the same list in the same order took closer to 1,000 runs. In his words, any tool that gives a "ranking position in AI" is "full of baloney."

I agree with him, and it is why this study looks at visits and orders. A visibility score changes every time the question is asked. Visits and orders happened, and the only open question is how many of them GA4 missed.

SparkToro and Gumshoe ran the same prompts thousands of times and rarely got the same list of brands twice.

There are two more limits to keep in mind. Orders are GA4 purchases, which can differ from the commerce platform's order count, and one year of data is too short to predict where this will go next.

Explore the dataset

I made the full dataset public: monthly totals for all 73 stores, an anonymized row for each of the 73 stores, the AI assistant and landing page breakdowns, and the exact rules used to classify AI traffic, so you can reproduce the numbers in your own GA4. Download the dataset, copy it, filter it, and chart your own cut. If you publish anything from it, credit Shero Commerce and link back to this study.

The Store Data tab has one row per store, with random IDs and volumes shown as ranges so no client can be identified.

Conclusion

AI traffic to these stores did not grow this year, and it still sent almost three times the orders. Anyone looking only at AI visits would think nothing happened.

If I managed an eCommerce store, I would spend the next quarter on the product pages AI assistants link to and on the tracking above, so AI orders show up next to organic search in the monthly report.

Frequently asked questions