A tracked brand’s own website appears in only 16.05 percent of local AI responses, according to the State of AI Search report from AthenaHQ, which drew on seven AI models tracked from December 2025 to March 2026. Search Engine Journal’s Matt G. Southern reported the finding August 3, alongside a separate Uberall benchmark of quick-service restaurant queries that reaches a similar conclusion by a different route. Local operators now compete for space inside an AI answer, not just a search results page.
This publication covered a related finding yesterday: Aleyda Solis’s cross-vertical citation analysis of SaaS, ecommerce and finance queries, which found that most AI citations sit on domains a brand does not own. The AthenaHQ and Uberall data is a separate dataset, from different vendors, focused specifically on local search. Two independent measurements landing on a compatible conclusion inside twenty four hours is itself notable, even though neither confirms the other’s specific figures.
AthenaHQ found AI models cite between 6 and 27 domains per response, averaging around 12, across all industries in its dataset. Individual engines vary widely: ChatGPT averaged 18.86 cited domains per response, Grok averaged 26.99, and Copilot averaged 5.77. Uberall’s restaurant-specific benchmark, which the company describes as directional rather than definitive, found a narrower range: ChatGPT typically cited 16 sources, Copilot and Google’s AI Overviews both cited 8, Perplexity cited 7, and Gemini cited 4.
Both datasets are vendor-supplied and single-sourced per platform. AthenaHQ sells AI visibility tools and Uberall sells listings and reputation software; their own commercial interest is in businesses believing citation counts matter. Neither figure has been independently verified, and Uberall itself calls its restaurant numbers directional. The figures are internally consistent with each other and with the broader citation research this publication covered yesterday, which makes them worth taking seriously without treating them as settled science.
The most useful number in the dataset is not the 16 percent figure. It is the per-engine spread in how many domains get cited. A local business queried on Copilot, which cites under six domains on average, competes against a handful of rivals for space in the answer. The same business queried on Grok, which cites nearly 27, is one name among a crowd nearly five times larger.
On a narrow-citation engine, being one of five or six sources named is close to a direct recommendation. On a wide-citation engine, being one of twenty seven is closer to a footnote a customer may never read past. That distinction should reorder which engines a local team prioritizes. Neither report frames the data that way.
Reddit accounted for 21.85 percent of citations in AthenaHQ’s dataset and YouTube for 10.32 percent, well ahead of any single review site. Businesses ChatGPT recommends average 4.3 stars, per Uberall’s restaurant benchmark, compared with 4.1 for Perplexity and 3.9 for Gemini. Plenty of businesses sitting at 4.0 still rank perfectly well in Google, so treat the ChatGPT number as a description of what recommended businesses tend to look like, not a threshold you have to clear.
For a local business, the practical order of work follows the citation data rather than the website. Third-party listing consistency across platforms such as Google Business Profile, Apple Business Connect and Bing Places matters more than any on-page copy change, because that is the layer AI systems reconcile first. Review volume, rating and recency come next, followed by visibility on community platforms like Reddit and YouTube. On-site content optimization still matters for the 16 percent of responses that do cite a brand’s own domain, but teams with limited time should spend it on listings and reviews before they spend it on the website.
Search Engine Journal published this report by Matt G. Southern on August 3, 2026.
