Key Takeaways

  • A study of 973 ecommerce sites with roughly $20B in combined revenue found ChatGPT drives more than 90% of LLM-referred sessions and generated over 50,000 tracked transactions between August 2024 and July 2025.
  • ChatGPT referrals still make up less than 0.2% of total ecommerce traffic in the academic sample, and they convert worse than organic or affiliate on most sites, even as the absolute volume climbs fast.
  • Category concentration matters more than the average: Digiday reports ChatGPT now drives about 20% of Walmart's referral traffic, with meaningful share to Etsy, Target, and eBay as well.
  • Search Engine Land reports ChatGPT traffic converts 31% higher than non-branded organic search on the sites they measured, which flips the story from volume to intent quality.
  • The right first move is not a new AEO budget line, it is instrumenting GA4 to see ChatGPT sessions cleanly so the number you argue about is your number, not somebody else's benchmark.

Six months ago, if a CMO asked how much traffic ChatGPT was actually sending to ecommerce sites, the answer was a vendor deck, a single-site anecdote, and a shrug. In April, Kaiser and colleagues published first-party GA4 data from 973 ecommerce websites, roughly $20B in combined revenue, tracking more than 50,000 ChatGPT-referred transactions between August 2024 and July 2025. That is the first time we have had a real denominator under any of this, and it is worth reading carefully because the headline number and the footnote number point in almost opposite directions.

It's interesting to me that the same dataset is being used to argue two different things. One camp is waving the 50,000 transactions and the 90%-of-LLM-share figure and calling it the beginning of a new channel. The other camp, including one of the paper's own authors on LinkedIn, is pointing out that ChatGPT referrals are still under 0.2% of total traffic and convert worse than organic or affiliate on most sites. Both are true. The interesting question for a CMO is not which camp is right, it is what you should actually do this quarter given that both are right at the same time.

My read is that this is the moment to instrument, not to invest. Get the measurement in place, watch your own numbers against the benchmarks, and let the data tell you when the curve bends.

What the 973-site study actually measured

The paper, published in Marketing Science, is doing something most vendor reports are not: looking at first-party GA4 data across a large, heterogeneous set of ecommerce sites over overlapping 3, 6, and 12-month windows. That last detail matters, because ChatGPT referral volume is growing so quickly that a snapshot from Q4 2024 and a snapshot from Q2 2025 look like different planets. The overlapping windows let the authors separate the growth curve from the level.

Two findings are worth pulling out.

First, ChatGPT dominates the LLM referral category so completely that treating "LLM traffic" as a category is misleading, it is a ChatGPT number with a rounding error of Perplexity, Claude, and Gemini around it.

Second, the transactions are real. Fifty thousand of them, attributable, tracked in Google Analytics, tied to sessions that came from a chat.openai.com or chatgpt.com referrer. This is not modeled, it is counted.

What the paper does not do, and what the LinkedIn commentary from Prof. Schulze makes clear, is claim that this is a large channel yet. Under 0.2% of total sessions is under 0.2%. If you have a $50M ecommerce business, that is a rounding error on your monthly revenue.

If you have a $5B business like the ones Digiday was writing about, 0.2% is a real number, and the category concentration means some of you are already seeing double-digit percentages of your referral traffic coming from ChatGPT.

Why the same data supports two opposite stories

Averages across 973 sites hide the categories where ChatGPT is already meaningful. Digiday reported in September that ChatGPT now accounts for roughly 20% of Walmart's referral traffic, with Etsy, Target, and eBay also seeing meaningful share from the same source. Those are not fringe sites. But referral traffic is a slice of total traffic, and for most retailers referral is a small slice of the pie. So "20% of referral" and "under 0.2% of total" can both be true for the same brand in the same week.

The conversion picture has the same shape. Prof. Schulze notes that ChatGPT referrals convert worse than organic or affiliate on average across the 973 sites. Search Engine Land, working from a different dataset, reports ChatGPT traffic converts 31% higher than non-branded organic search. Those are not contradictions once you look at what is being compared. Non-branded organic is a low-intent baseline. Affiliate and branded organic are high-intent. ChatGPT sits somewhere in between, and where it sits depends heavily on the product category, the prompt patterns people use, and whether the site is being cited inside answers or just linked from a follow-up.

The distribution is the story, and you can only see your own place in it if you are measuring your own site.

Where to point your analytics this quarter

The good news is that this is not hard to set up. Practical Ecommerce has a clean walkthrough for GA4: Acquisition, Traffic acquisition, then filter session source to include chatgpt.com and chat.openai.com. Do that today.

From there, save it as a segment rather than leaving it as a one-off filter, because a saved GA4 segment for ChatGPT-referred sessions lets you overlay it onto every other report, landing pages, product pages, checkout funnel, without rebuilding the query. This is where you will learn which pages are actually pulling the traffic, which is almost never the pages your SEO team assumes.

If you already have ecommerce tracking in GA4, transactions will appear in the segment automatically, and it is worth adding a custom dimension for referrer at the transaction level so you can join this to your data warehouse and compare AOV, return rate, and lifetime value against your organic and paid cohorts. The 973-site paper had to work with GA-level data. You do not.

Then watch the slope. If your ChatGPT sessions are doubling quarter over quarter, that is a signal even if the absolute number is 400. If they are flat, that is a signal too, and it probably means you are not being cited in the answers for your category.

When investing ahead of the data is the wrong move

We should all be aware of the other side of this, because the AEO consulting market is running well ahead of the evidence. If you sell a considered B2B product with a six-month sales cycle, ChatGPT referral traffic is not going to move your pipeline this year, and building out a dedicated AEO program with headcount and agency spend is almost certainly premature. The Azoma analysis of promising categories is heavy on shopping, travel, and consumer research use cases, which makes sense because those are the prompts people actually type into ChatGPT. Enterprise software procurement is not one of those prompts, yet.

Similarly, if you are a mid-market ecommerce brand with a healthy paid and organic mix, the right posture is to instrument, watch, and be ready, rather than reallocate 10% of your budget to AEO because a vendor deck told you to. The 50,000 transactions across 973 sites averages out to about 51 transactions per site over a year. Some sites got thousands, most got a handful. Assume you are in the handful until your own data says otherwise.

Where I would spend real effort right now is on the content itself, making sure your product pages, comparison pages, and category explainers are structured so a model can actually cite them. That work pays off in traditional search too, so the downside is bounded. If you want a framework we use with clients for that, we wrote it up in our marketing blueprint.

What to do today

The move worth making this month is to set up a ChatGPT segment in every GA4 property that does not already have one, and add a monthly line to the reporting dashboard that shows sessions, conversion rate, and revenue for the segment against a rolling 90-day baseline.

The point is not that the numbers will be big yet. The point is that they will change, and you want the change visible the month it happens instead of the quarter after.

The Kaiser paper is the first hard floor under this conversation. Use it as a floor, not a ceiling. Your numbers are the ones that matter.

Frequently Asked Questions

How much ecommerce traffic does ChatGPT actually drive today?

Across the 973-site academic sample published in Marketing Science, ChatGPT referrals were under 0.2% of total sessions on average, but drove over 90% of all LLM-referred traffic and generated more than 50,000 tracked transactions between August 2024 and July 2025. Category concentration is heavy: Digiday reports ChatGPT now accounts for roughly 20% of Walmart's referral traffic, with Etsy, Target, and eBay also seeing meaningful share.

Both, depending on the comparison. The 973-site study found ChatGPT converts worse than affiliate and branded organic on average. Search Engine Land reported ChatGPT converts 31% higher than non-branded organic. The honest read is that ChatGPT sits between low-intent and high-intent traffic and the answer depends on your category and how you are being cited.

How do I track ChatGPT referrals in GA4?

Go to Acquisition, then Traffic acquisition, and filter session source to include chatgpt.com and chat.openai.com. Save it as a segment so you can overlay it onto every other GA4 report. Practical Ecommerce has a step-by-step walkthrough linked in the sources below.

Should I be investing in Answer Engine Optimization now?

For most B2B and mid-market brands, the right move this quarter is to instrument, not invest. Get the measurement in place, watch your slope, and be ready to move when your own data shows the curve bending. For consumer categories already seeing double-digit referral share, the investment case is already there.

Which ecommerce categories are seeing the most ChatGPT traffic?

Shopping and consumer research categories dominate. Azoma's analysis identifies 13 promising categories, and Digiday's reporting confirms marketplaces like Walmart, Etsy, Target, and eBay are seeing the strongest referral share. Considered B2B purchases show up much less, which matches the prompt patterns people actually use.

Sources