Methodology and limits
Sample. 625 audits of 511 websites between 31 March 2026 and 28 September 2026, run by Bodhi & Co’s GET SEEN audit engine. Where a site was audited more than once, only the latest audit is counted. The sample is not random. It is the set of sites that went through our audit pipeline, and 344 of the 511 are real-estate sites. Businesses were audited from public information, and none is named here.
Two versions of the audit. 400 sites were last audited between 31 March 2026 and 5 August 2026 with an earlier version that produced an AI-visibility score only. 111 were audited from 5 August 2026 with the current version, which adds an SEO score and the live citation test. We do not pool the two. Pooled, the 511 sites would average 13.3 on AI visibility (median 8), a figure that mostly reflects the earlier audit.
Scores. Both scores are graded, not measured. Our audit engine crawls the site, and an AI model grades what the crawler found against a written rubric: content quality and originality, named people and credentials, question-led structure, and access for AI crawlers. In the current version, code recomputes the SEO score from the category grades and caps the AI-visibility score at 25 when a site blocks indexing. Graded scores can vary from run to run.
Live citation test. 108 businesses, 5 August 2026 to 28 September 2026. Each was asked 3, 5 or 10 buyer questions, 422 in total. The questions were written by the audit engine from the business’s own website, as the kind of question a buyer would put to an AI assistant. They were not collected from real customers. 73 businesses were tested on Claude alone, 27 on two or more engines (21 of them on all four), and 8 earlier tests did not record the engine.
What counts as a citation. The business’s domain appeared among the sources an engine cited, or in the text of its answer. Detection is by matching web addresses in code. “Never cited” means no citation across every question and engine that business was tested on. Of the 41 never-cited businesses, 27 were tested on Claude alone and 13 on two or more engines.
Engines. Claude, ChatGPT, Gemini and Perplexity were queried through their developer APIs with live web search switched on. The consumer apps may use different models and return different answers.
Sources. Source counts come from 34 businesses tested between 1 September 2026 and 28 September 2026. For each we recorded the 10 most-cited sources, not every source in every answer.
Region. A business is counted as Upstate when the service area recorded by the audit names an Upstate city. 195 sites have no service area recorded.
Other limits. AI engines give different answers to the same question at different times. Industry groups with fewer than 7 businesses are left out of the charts. The checklist figures come from 28 audits.