At Roivenue, we analysed hundreds of e-shops from different countries and segments to find out what real impact the development of AI assistants has on customer shopping behaviour. And the result will be surprising to anyone who believes the AI prophets on LinkedIn: the direct commercial effect of LLM queries was almost negligible in 2025.
In about five per cent of e-shops, we found no demonstrable sales that could be clearly linked to LLM. For the rest, their share ranges from 0.001% to 0.43%, with an average of only 0.07%.
Beware of hype
Roivenue gives companies the ability to analyse the impact of individual channels and marketing activities on sales. Thanks to real data, we know that much of what is said in the market is not true. For example, Facebook is not dead and still performs much better than any other social network in e-commerce.
When we look at aggregated data across e-shops, the same pattern repeats itself. The largest share of traffic and conversions comes from direct visits. However, this traffic can easily hide users who first chatted with AI about the brand or product and then went to the e-shop with a clear idea of what they wanted.
The impact of AI assistants on the e-commerce business is therefore more difficult to analyse. They most often appear at the very beginning of the customer journey, i.e. in the see or think phase. They help customers realise their needs and assist them in their selection, but the purchase takes place through a different channel. AI-assisted traffic and ultimately conversions are therefore difficult to find in the data and are more reminiscent of personal recommendations, i.e. word of mouth.
SEO is not ending, it is just changing context
If AI advises on selection, isn't it most important to get into the AI selection? Forget SEO, do AIO! Our data certainly does not suggest that SEO has stopped working. Quality content, good structure, technical cleanliness, speed and relevance are just as important as they were before the AI boom. The basics haven't changed, yet most companies have weak foundations.
For the more advanced, market leaders, the content on the website will of course also change. Content no longer ends in search results, but is also beginning to appear in the responses of AI search engines and assistants. And language models are naturally closer to factual texts, instructions, comparisons or Q&A formats. On the contrary, long narrative or creative texts that worked on a blog for users are less effective in AI responses because it is not easy to extract clear information from them.
Dalibor Cicman offered an interesting practical example, showing how GymBeam's traffic has changed over the last year. The blog fell by almost a million visits, while traffic from AI rose from zero to 35,000. At first glance, this may look like a decline, but the reality is the opposite. Visits from LLM converted several times better and brought in approximately 1,700 orders.
AI has and will continue to have a significant impact on e-commerce, but much more slowly and in a completely different way than is often presented. You don't have to worry about missing the boat; don't get carried away by the hype. The best preparation for the AI revolution is still the old boring order in data and processes.
The author of the commentary is Milan Knapp, Executive Director of Roivenue. A Czech company that develops an analytical tool for optimising advertising effectiveness.
Source: mam.cz