Study: AI assistants cite business directories 2.3x more than brand websites

When someone asks an AI assistant to recommend a plumber, a software vendor, or an engineering firm, the answer they get is far more likely to come from a business directory than from the company's own website. A study released in September 2026 by Citations.press examined 3,850 commercial prompts across four major AI platforms—ChatGPT, Google AI Mode, Perplexity, and Google AI Overviews—and found that sector directories supplied 41% of the citations behind AI-generated supplier recommendations, compared to just 18% for brand-owned pages. The gap was even starker when the AI was asked about a category rather than a specific company: brand homepages appeared in only 6% of those responses.

The research, conducted between mid-June and late August 2026, tested prompts across six industry verticals: engineering, IT supply, channel partnerships, agency selection, training provision, and consumer services. The disparity between directories and brand sites was not uniform across sectors. In engineering, the ratio was most extreme—directories were cited 4.2 times more often than brand websites. The IT supplier and channel partner categories showed similar patterns, suggesting that procurement-style searches, where a buyer is looking for options within a category rather than information about a known company, trigger higher directory citation rates.

The structure of the directory itself mattered significantly. Directories with one entity per entry and consistent, standardized attributes were cited 3.1 times more often than directories organized as free-text listings. This finding suggests that AI models, when answering a narrow category question, favor sources that resolve that question completely and systematically over sources that are merely well-known. The citation rates also varied substantially by platform. On Perplexity, sector directories appeared in 47% of category-prompt responses. On Google AI Overviews, the figure was 39%. Google AI Mode returned directories 36% of the time. ChatGPT cited them in only 22% of responses. Across all four platforms, there was only 31% overlap in which sources were actually cited, suggesting that each AI system draws from a different pool of trusted sources.

The researchers found a weak correlation between a directory's domain authority—a measure of a website's overall trustworthiness and link profile—and how often it was cited. The Spearman rank correlation was 0.19, indicating that raw domain authority was not a strong predictor of citation frequency. Coverage depth within a stated category proved to be the stronger signal. As the study notes, this suggests that AI models are optimizing for completeness and relevance to the specific query rather than deferring to general measures of website authority. The researchers emphasized that this finding comes from a single dataset collected over one summer in a handful of sectors, and that replication in other categories would be necessary to confirm whether the pattern holds more broadly.

The study's methodology involved classifying every cited URL in each response by source type. The first 400 responses were classified manually, and subsequent responses were classified using rule-based automation. A 200-response sample was re-checked by hand, showing 94% agreement with the automated classification. The researchers published the complete prompt set, raw citation records, classification rules, and the manual audit alongside the study, allowing other researchers to attempt replication. They acknowledged several limitations: AI responses can vary across repeated runs, different accounts, and different regions; the prompt set was weighted toward the six sectors tested; some judgment was involved in source-type classification; and the study measured citation frequency rather than any actual effect on purchasing behavior.

The findings align with patterns reported in three other large-scale studies published over the previous year. DeltaV Digital examined more than 25,000 citations across 21,000 AI responses and found that listicles accounted for 61% of citations in B2B technology services, with brand-owned domains cited 0% of the time in that sector. Semrush analyzed more than 100 million citations across 230,000 prompts and reported that citation weight concentrated among a small number of aggregating platforms. Peec AI reviewed 30 million cited sources and found that listing and review platforms like G2 and Yelp appeared frequently in recommendation queries. These studies measured different samples over different periods and were not designed to test the same question, but they point toward a consistent pattern: when AI assistants answer questions about where to find a service provider, they are drawing heavily from aggregated, categorized sources rather than from individual company websites.

The implications are significant for how businesses approach visibility in an AI-driven search landscape. If a company's own website is cited less than half as often as a sector directory when a potential customer asks an AI for a recommendation, then being listed in the right directories—and being listed with consistent, structured information—becomes a critical part of visibility strategy. The finding that directory structure matters more than domain authority suggests that the traditional SEO playbook, which has long emphasized building domain authority and earning links, may need to be supplemented or even reoriented. For directories themselves, the research indicates that maintaining clean, consistent data and comprehensive coverage within a stated category is more valuable to AI citation patterns than general measures of website authority.