A Lightspeed-commissioned audit by Vaer AI found that large retailers were more likely than small businesses to be named or recommended in the shopping prompts it tested—even when AI systems cited local stores as sources. The result is a bounded study of selected AI systems and queries, not proof that AI always favors chains or that the recommendations changed sales.
What the study measured
Lightspeed founder and CEO Dax Dasilva’s October 2, 2026 Fortune commentary draws on an audit conducted by Vaer AI and commissioned by Lightspeed. Vaer says it analyzed shopping prompts from June through August 2026. The test covered 20,000 prompts in 10 product categories across Los Angeles, San Francisco, New York City and Montreal.
As an Amazon Associate I earn from qualifying purchases.
The audit tested prompts under two conditions: using a model’s knowledge without live search, and with live web search. Across those runs, the study describes roughly 460,000 AI responses—200,000 without search and 260,000 with search. Lightspeed’s September 29 announcement says the analysis covered more than 2.4 million links; Vaer’s study page describes 2.5 million. Those are source-specific figures, not a single reconciled total.
The platforms varied by condition. Vaer says the no-search runs used ChatGPT and Gemini; live-search runs used ChatGPT, Google AI Mode and Google AI Overviews. Lightspeed’s release lists ChatGPT, Google AI Mode, Google AI Overview and Gemini across the study. These descriptions are not identical, so platform comparisons should be read in light of the specific run being discussed.
Why a citation is not the same as a recommendation
An AI answer can cite a business without putting it forward as the best place to shop. That distinction is central to Dasilva’s argument: showing up among an answer’s sources may not mean a retailer receives the shopper’s attention or the first recommendation.
In the New York office-supply example in the Fortune commentary, three of six ChatGPT sources were small local retailers, but Staples appeared as the first recommendation. Across the results Dasilva describes, large and small retailers each accounted for roughly 38% of cited sources, while large chains received 52% of top recommendations and smaller businesses 20%. These figures describe different outcomes—source share and top-recommendation share—and should not be treated as interchangeable measures of visibility.
What the audit reported about retailer size
Vaer classified retailers in large, medium and small bands. It says the classification used one classifier, checked against a hand-built answer key and a second AI model. Its headline findings point to an advantage for large chains in the tested prompts, with the outcome differing between search conditions.
| Test condition or result | Reported finding | How to read it |
|---|---|---|
| Model knowledge, no live search | National chains appeared in 63–70% of answers; small or local shops appeared about one time in ten. In neutral head-to-head choices, the larger retailer was selected 90–94% of the time. | Vaer AI’s results for its no-search prompt set, not a rate established for all AI answers. |
| Live web search | National chains remained the named recommendation in 46–58% of responses, depending on the AI system; local shops were named about a quarter of the time. | Vaer AI’s platform-dependent results for its live-search runs. |
| Broad versus specific shopping searches | Lightspeed says local retailers were a top recommendation in about one-third of broad shopping searches, but in as little as one in ten specific product searches. | Lightspeed’s summary of the study; the result varies with query intent. |
| Google AI Overviews | Lightspeed says no local store was named in 68% of its shopping responses. | A result for this platform in the tested prompts, not a finding about every Google AI Overview. |
| ChatGPT and Google AI Mode | Lightspeed says each surfaced a local retailer in roughly 70% of responses. | “Surfaced” a retailer does not necessarily mean it was the top recommendation. |
Vaer’s results suggest a distinction between general discovery and specific product intent: local businesses appeared less often when the shopper’s request narrowed to a particular product. The figures do not establish why a model chose a retailer, nor do they show what a shopper ultimately bought.
Rank #3
What changing the prompt did
Vaer also tested changes to prompt wording, including adding “local,” “near me” or “independent.” In that test, adding “independent” more than doubled small or local shops among recommendations and reduced national-chain retailer sources from about 44% to as low as 9%. Vaer reported a substantially stronger effect from “independent” than from “local” or “near me.”
This was an intervention within the study, not a guarantee for a real shopper’s query. “Local” and “near me” can describe geography without ruling out a chain nearby; “independent” more directly expresses a preference about the kind of business. A shopper who wants to prioritize independent retailers can try wording the request that way, while checking the answer’s sources and confirming details such as location, stock and hours directly with the store.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What consumers said—and what that does not prove
A separate Lightspeed-commissioned Censuswide survey of 2,000 North American consumers in 2026 found that 56% said they had used AI for shopping decisions and 41% said they trusted AI shopping recommendations. Half said they would be more likely to shop locally if AI made nearby independent retailers easier to discover. Asked what AI should prioritize, 33% favored small or local businesses, compared with 13% who favored large brands.
Recommended Free Tools
These are survey responses about reported use, trust and stated preferences. They are not observed purchase records, a population-wide census or evidence that a particular AI recommendation causes a local sale.
How far the findings can be generalized
The audit is evidence of what happened under its chosen prompts, cities, platforms, retailer classifications and study period. It is not an independent audit: Lightspeed commissioned it, and the public pages reviewed do not provide response-level data or the complete prompt set for independent reproduction. Lightspeed says the headline figures carry 95% confidence intervals, but that does not resolve those transparency limits.
- It does establish: In this study’s tested conditions, large chains were often named or recommended, while outcomes varied by platform, search access and prompt wording.
- It does not establish: That all AI shopping answers favor chains, that the same rates apply to every city or query, why a model produced a particular answer, or whether the findings translated into sales.
The practical implication is narrower than a universal claim about AI: being included in a set of sources does not ensure that an independent retailer will lead the answer. The study raises a visibility question for local businesses, but its results should be treated as a measured snapshot of one commissioned audit rather than a complete account of AI-driven retail discovery.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
