Your e-commerce report now includes visits from ChatGPT. The team proposes spending more on content for this channel, but you want to know whether those visitors actually buy. Do they purchase more often than visitors from other channels? In a peer-reviewed Marketing Science article, researchers found that ChatGPT traffic converted better than paid social but worse than other channels after accounting for website, device and month differences. Conversion rate means the share of sessions ending in a transaction. DijitalPi commentary: Assess traffic volume alongside revenue contribution before allocating budget. The study does not measure additional sales caused by visibility work. Some alternative analyses find no significant gap against organic search, and the findings are not separately validated for Türkiye or Turkish.
ACADEMIC WORK REVIEWED
Frontiers: ChatGPT Referrals to E-Commerce Websites: How Do LLMs Compare Against Traditional Channels?
ResearchersMaximilian Kaiser, Christian Schulze
Marketing Science · 21 April 2026 · Journal version, online publication · Peer-reviewed article · Source language: English
English review: 9 October 2026 · The source date appears in the citation above.
Editorial review draft
The source has been checked; the review still carries its original editorial status. DijitalPi did not conduct a customer experiment for this paper.
LET'S READ THE RESEARCH TOGETHER
We first explain the researchers' question, method and findings. We then discuss how to interpret the results, clearly separating DijitalPi's commentary from the source.
01 / WHAT DID THE RESEARCHERS WANT TO UNDERSTAND?
How do visits from ChatGPT compare with other channels in generating e-commerce transactions and revenue?
The researchers compared traffic from ChatGPT’s organic outgoing links with traditional marketing channels. They examined purchasing frequency, order value and revenue per session alongside traffic volume. They also investigated variation over time and across websites with different product complexity and audience characteristics.
Explaining the context · DijitalPi commentary
DijitalPi commentary: The appearance of a new traffic source does not establish how much marketing budget it deserves. The practical decision is which products to monitor and which measures to use. This study offers a comparison for testing expectations about the direct sales value of ChatGPT referrals; it does not calculate the return on visibility work.
02 / HOW WAS THE RESEARCH CONDUCTED?
How did they test the question?
Kaiser and Schulze used Google Analytics data from 973 e-commerce websites, accessed through Grips Intelligence. The 12-month dataset covering August 2024–July 2025 includes $20 billion in revenue, more than 50,000 ChatGPT transactions and 164 million transactions from other channels. The main analysis covers 2 February–2 August 2025, with 44,437 ChatGPT transactions and 4,149,187 ChatGPT sessions. The data span 24 website categories and all continents, with most sessions coming from the Americas and Europe. There are 49 countries contributing at least 10 million sessions each; Türkiye is not separately identified. This was an observational study without random assignment of visitors to channels.
Understanding the method · DijitalPi commentary
Data were collected weekly, with each observation representing a week, website, device and channel combination. The main conversion model contains 340,292 observations. The researchers accounted for website, device and month differences when comparing each traditional channel against ChatGPT traffic. These adjustments do not eliminate all possible explanations, including differences between the users attracted by each channel.
The main measures were the share of sessions ending in a transaction, average order value and revenue per session; bounce rate, session duration and page views were also examined. A session is a period of website activity that typically ends after 30 minutes of inactivity. Channel attribution uses the last click, leaving earlier discovery contributions incompletely captured. Links were organic during the study period; the platforms had not yet introduced advertising or affiliate-like systems.
03 / RESEARCH FINDINGS
ChatGPT traffic showed no general advantage on direct sales measures.
The researchers report that, one year after launch, ChatGPT accounted for less than 0.2% of total traffic and was about 200 times smaller than Google organic search traffic. In the main analysis accounting for website, device and month differences, conversion likelihood was 53% lower for paid social, 13% higher for organic search and 86% higher for affiliate traffic than for ChatGPT. These are not percentage-point differences. ChatGPT revenue per session exceeded paid social but was significantly below every remaining channel. Average order value differed significantly from only four of eight channels; differences against organic search, paid search, paid social and the other-channel group were not significant. Affiliate traffic had the largest significant order value difference, at $24.7 above ChatGPT. The conversion gap against organic search became insignificant in five alternative analyses; its direction reversed under a minimum threshold of 1,000 sessions, but that result was also insignificant.
ChatGPT’s share of total traffic
Less than 0.2%The observed traffic share one year after launch within the dataset, not a share of sales or revenue.
Organic search conversion likelihood
13% higherRelative to ChatGPT in the main model; not a percentage-point difference, and insignificant in some alternative analyses.
DijitalPi commentary: Start a channel assessment by examining volume alongside revenue per session. Do not treat a lower bounce rate as direct sales success: the researchers found ChatGPT performed better than most channels on this measure, while organic and paid search still performed better. Reporting conversion rate, order value and revenue separately for your own product group is a cautious starting point suggested by these findings.
The overall average does not describe every website. The researchers found an average 4.6 times higher ChatGPT traffic share for websites in complex product categories, with conversion rates exceeding some other channels. Conversion gaps also narrowed where audience characteristics suggested greater proficiency, using technological interest and younger audiences as indicators. These are website-level measures, not direct assessments of individual users’ proficiency or product effects. Differences in how categories define a conversion may also explain part of the complexity finding.
05 / CONCLUSION AND OPEN QUESTIONS
What did we learn, and what do we still not know?
The peer-reviewed journal version depicts ChatGPT as a low-volume channel with limited direct revenue performance during the period studied. Conversion rate and revenue per session rose over the first 12 months while average order value declined; the authors regard future projections as highly uncertain. The findings are not causal, last-click attribution may understate discovery contributions, and the results cannot be directly generalised to Türkiye or Turkish-language use. Grips Intelligence supplied data and computing resources, with no direct financial support to the authors. First author Kaiser is employed by the company. According to the authors, its involvement was limited to data access, computing costs and technical guidance; it neither commissioned the study nor approved the paper.
Our open question at DijitalPi: do ChatGPT traffic volume and revenue per session in your own product group support the effort you plan to allocate, and how will you assess discovery touchpoints absent from last-click reporting?
We created these scenarios to make the topic concrete. They are not cases from the paper or measured client results.
EXAMPLE 01
E-commerce: effort spent on product comparisons
An illustrative electronics retailer is considering improving product comparison pages for visitors arriving from ChatGPT.
Which evidence should guide further investment in this content?
Open the recommendation for example 01+
Assess transaction rates and revenue per session alongside traffic volume.
Report ChatGPT, organic search and paid social sessions within the same periods and device groups. For sessions reaching comparison content, show transactions, average order value and revenue per session separately. Flag uncertainty when there are few transactions. This observational comparison cannot establish that a content change generated additional sales.
EXAMPLE 02
Tourism: information needs before booking
An illustrative accommodation business suspects that visitors from ChatGPT check room conditions but may complete their booking on a later visit.
Does an absence of attributed sales mean this channel contributes nothing?
Open the recommendation for example 02+
Examine completed bookings and earlier discovery touchpoints separately.
Define a booking transaction as the conversion and report transaction and revenue data for ChatGPT sessions. Where existing measurement allows, examine channel touchpoints preceding the booking separately. Last-click records can understate the contribution of earlier research. This proposal is not a revenue effect validated by the study for accommodation businesses.
EXAMPLE 03
B2B: visitor value when choosing technical products
An illustrative industrial parts supplier notices visitors arriving through ChatGPT while researching technical product choices.
Can the finding about complex products apply to my own range?
Open the recommendation for example 03+
Examine the technical product group separately and distinguish quotation requests from sales.
Report ChatGPT and search sessions reaching pages that compare technical specifications. Use separate measures for quotation requests, completed orders and revenue. The study measures complexity at website level rather than for individual products. Do not translate its finding into an expected gain for your product range.
TRY IT WITH YOUR TEAM
Build your channel comparison.
We suggest preparing a table of sessions, transactions, order value and revenue per session for ChatGPT, organic search and paid social within the same periods and device groups. Record that last-click attribution may understate discovery contributions when interpreting the results. This proposal has not yet been implemented.
Maximilian Kaiser, Christian Schulze Maximilian Kaiser — University of Hamburg, Germany; Grips Intelligence, Berlin, Germany. Christian Schulze — Frankfurt School of Finance & Management, Germany.
The academic work belongs to the researchers named above. This page contains DijitalPi's explanatory review and original business examples; it is not a full translation of the paper.
Cookie preferences