oLLM eCommerce analysis – 2025
Overview:
Great analysis from Prof Maximilian Kaiser & Prof Christian Schulze on oLLM’s (outbound Large Language Model referral e.g ChatGPT, Perplexity, Co Pilot, Claude, etc)
Full paper available on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5585812
Direct PDF: https://papers.ssrn.com/sol3/Delivery.cfm/5585812.pdf?abstractid=5585812&mirid=1
Summary:
This research constituted 12 months of 1st party Google Analytics data from 973 online shops & ~$20bn in cumulative ecommerce sales = decent Quant/Qual analysis
The 68 page paper(!) denoted quality analysis of where oLLM traffic actually sits in today’s ecommerce channel mix. The findings dispel some of the hype and align with what many performance marketing teams are seeing “on the ground” from 1st party data.
Core Findings:
The study shows that oLLM traffic remains a marginal contributor in 2025 for most within the ecommerce channel mix. Volumes are low and sit consistently at the bottom when compared with established channels (SEO, PPC, Direct, etc). Sales performance is relatively weak with CR% below all major sources of demand. Organic search converts @ c.13% higher than oLLMs and the only channel oLLMs marginally outperform is paid social. Revenue per session is also structurally low which limits the commercial impact even where relevance signals appear reasonable.
Overall the data directly challenges the view from some that oLLM traffic is already a meaningful sales driver. One year (2025) of observed performance shows that it underperforms almost every traditional channel on core commercial metrics.
Performance trends show a mixed picture. CR% from oLLM traffic has improved steadily over time which suggests growing user familiarity + better intent matching. However AOV actually declined slightly during 2025 and the detailed analysis suggests this could fall further (I’m less so sure on this assertion though as I believe ollm’s could play a bigger role in larger £ticket items + in the likes of B2B marketing during 2026 > larger “baskets”). The channel is generally skewed towards small simple purchases to date and narrow buying tasks. It does not support complex or higher-value journeys (yet) and is not behaving like a substitute for organic or paid search in commercial terms.
The Future
Looking into my crystal ball for 2026 I see the next 12 months ramping up OLLM usage and bucking some of the 2025 trends that Prof Maximilian Kaiser & Prof Christian Schulze analysed. As an example i see further usage growth in 2026 though still mainly from relatively “early adopters” + younger generations; gradual improvement (clicks, revenue per session, AOV increase, CR% etc) will be noted Vs mass (digital) channel disruption.
AI results will be inserted into SERPs more seamlessly and people will use the likes of Gemini AI on Google without even realising it. I see ads being tested > rolled out on the likes of ChatGPT (mainly in the USA though other test markets such as the UK trialled in ’26) yet they will be more subtle, context aware and “woven into the narrative” Vs the more obvious PPC ads of Bing, Baidu & Google.
ChatGPT & Gemini will become the dominant (western) players though “lighter models”, backed by the likes of the Chinese government, such as open-weight players including DeepSeek-R1 will gain further adoption%.
OLLMs are not really (yet; excluding the likes of Gemini tech in Google AI Overviews) built for shopping environments and lack native features that support exploration comparison, trust, recommendation, etc > further broader awareness of their benefits + format adaption are needed for demographic penetration.
People also need to stop talking of oLLMs killing the likes of Google. I do believe ChatGPT et al will take some market share from the likes of Google though instead of widespread Substitution (e.g ChatGPT replaces Google usage) there will be a bigger% of expansion (e.g ChatGPT use adds to total search behaviour) > as noted in my research article HERE.
What I’m not suggesting for one moment though is that companies should not take oLLMs seriously; absolutely the contrary! What I am saying though is that we need to watch the current hype whilst preparing for the future where awareness > search > conversion will become ever more £%impacted by the likes of Gemini, ChatGPT & “dark horse”, government backed, open-weight models such as Deep Seek.