Without web search, AI named a large chain in 63–70 per cent of answers and a small shop in 10 per cent, according to Vaer AI
Chatbots and Google's AI features recommend large national chains to shoppers more often than small and independent shops, Vaer AI co-founder and principal consultant Tom Wells said following a study commissioned by Lightspeed Commerce.
Vaer AI published the results on 29 September. The company analysed almost half a million answers to 20,000 shopping queries in the US and Canada: 200,000 with web search switched off and 260,000 with it switched on. ChatGPT, Google AI Mode and AI Overviews were tested with search, and ChatGPT and Gemini without it.
Vaer AI classed stores with revenue of $1 billion or more as large, and those with revenue below $50 million as small. The company said the full technical report and methodology are available on request.
Without web search, ChatGPT and Gemini named a large national chain in 63–70 per cent of cases, and a small or local shop in about 10 per cent. When the models were shown a large chain and a smaller shop with no indication of size, they chose the large one in 90–94 per cent of cases.
With search switched on, the gap narrowed but did not close. A large retailer was the top recommendation in 46–58 per cent of answers, depending on the platform, while a small or local shop was named in about a quarter of cases, according to Vaer AI.
AI Overviews performed worst: in two out of three answers from this Google feature, no small shop was named at all.
Small shops featured more prominently in sources than in recommendations. Among the shops the AI linked to, large and small accounted for about 38 per cent each, but a large chain became the top recommendation about 2.5 times as often.
In AI Mode, small and local shops made up about 58 per cent of sources and 24 per cent of top recommendations. In ChatGPT, large retailers accounted for 41 per cent of sources and 58 per cent of top recommendations. Wells explained the gap by saying the models fall back on what they are more confident about, that is, the large chains in their training data.
The more specific the query, the higher the share of large chains. For a broad query such as “toys for a six-year-old”, a small shop was the top recommendation in about a third of cases, and for a query naming a product, such as LEGO Technic Ferrari, in about 10 per cent. The share of large chains rose from 40 per cent to 60 per cent over the same range.
The result also depends on the category. A local electronics shop appeared in 17 per cent of AI Overviews answers, a toy or pet supplies shop in 38–45 per cent.
One word in the query changed the picture. Adding “independent” more than doubled the share of small and local shops in recommendations, from about a third to almost four-fifths in a random sample of queries. The share of large chains among the sources the AI read before answering fell from about 44 per cent to 9 per cent on Google's platforms.
“Local” and “near me” had a noticeably smaller effect, and on ChatGPT “local” changed almost nothing. Vaer AI links this to the fact that the AI may count the nearest branch of a national chain as a local shop.
