Directories supply 41% of AI citations on supplier-selection prompts, new GEO and AEO study finds

A study of 3,850 commercial prompts across ChatGPT, Google AI Mode, Perplexity and Google AI Overviews finds sector directories cited more than twice as often as brand-owned pages when buyers ask an assistant to recommend a provider.

London, England, United Kingdom, September 15, 2026 /MarketersMEDIA/ -- London, 15 September 2026. Sector directories accounted for 41% of the citations behind AI-generated provider recommendations, against 18% for brand-owned pages, according to a study published today by Citations.press. The research, Sector Directories Are Driving AI Overview Citations, analysed 3,850 commercial prompts across ChatGPT, Google AI Mode, Perplexity and Google AI Overviews between 10 June and 22 August 2026.

The study reports that brand homepages appeared in 14% of all responses and in 6% of responses to prompts that named a category rather than a company. In the engineering category, directory citations exceeded brand-site citations by 4.2 to 1.

What the study reports

• Directories supplied 41% of citations on supplier-selection prompts, against 18% for brand-owned pages.
• 68% of directory citations came from list pages rather than individual profile pages.
• Directories with one entity per entry and consistent attributes were cited 3.1 times more often than directories with free-text listings.
• Citation rates for sector directories on category prompts were 47% on Perplexity, 39% on Google AI Overviews, 36% on Google AI Mode and 22% on ChatGPT.
• Overlap between the cited source sets across the four platforms was 31%.

Variation between sectors

The gap between directory and brand-site citations was not uniform. It was widest in engineering, where the sample was drawn from listings on Engineering Panel, at 4.2 directory citations for every brand-site citation. The IT supplier category, sampled from IT Suppliers, and the channel partner category, sampled from Partner Base, both showed the procurement-style prompt pattern the study associates with high directory citation rates.

Agency selection, sampled from Agency Roster, and training provider selection, sampled from Trainers List, produced the highest volume of category prompts in the set. The consumer category, sampled from Secret Salons, was the only one in which assistants returned named individual businesses more often than provider lists.

All figures above are first-party measurements produced by Citations.press. They have not been replicated by an independent party. The underlying records are published so that others can attempt replication.

Correlation with domain authority

Within this dataset, the study found a weak relationship between a directory’s domain authority and how often it was cited, reported as a Spearman rank correlation of 0.19. Coverage depth within a stated category was the stronger predictor. The study notes that a correlation of this size across a single dataset indicates an association rather than a causal relationship, and that the result may not hold in categories outside those tested.

“A model answering a narrow category question appears to favour a source that resolves that question completely over a source that is generally trusted. That is what our data points to. It is one dataset, over one summer, in a handful of sectors, and it needs replicating.”

How the findings sit against published research

Three larger studies published over the past year report broadly compatible patterns. Each is listed at the end of this release.

DeltaV Digital examined 25,337 citations across 21,075 AI engine responses covering eight brands over 90 days, and reported that listicles accounted for 19.6% of citations across its portfolio and 61% within B2B technology services, where own-domain citation share was recorded as 0%. Its authors concluded that "AI visibility is won on other people’s websites."

Semrush analysed more than 100 million citations across 230,000 prompts on ChatGPT Search, Google AI Mode and Perplexity over thirteen weeks, and reported citation weight concentrated among a small number of aggregating platforms.

Peec AI reviewed 30 million cited sources across five AI search surfaces and reported that listing and review platforms including G2 and Yelp appeared frequently in recommendation queries.

These studies measure different samples over different periods and were not designed to test the question examined here. They are offered as context, not as corroboration of the figures reported above.

Methodology

3,850 prompts were run across ChatGPT, Google AI Mode, Perplexity and Google AI Overviews between 10 June and 22 August 2026. Prompts were divided into category prompts, which named a sector but no company (2,140 prompts, 56% of the set), comparison prompts (890, 23%) and verification prompts, which named a specific company (820, 21%).

For each response, every cited URL was recorded and the source domain classified by type. Classification was carried out manually for the first 400 responses and by rule thereafter. A 200-response sample was re-checked by hand, showing 94% agreement with the rule-based classification.

The directory test set comprised six vertical directories, one for each sector examined: engineering, IT supply, channel partnerships, agency selection, training provision and independent consumer services. Directory citation rates for every other domain in the dataset were measured using the same method.

Stated limitations include response instability across repeated runs, accounts and regions, a prompt set weighted towards the sectors tested, judgement involved in source-type classification, and the fact that citation frequency was measured rather than any effect on purchasing behaviour.

Materials available to editors

The following are published in full alongside the study and are available to any editor or researcher on request: the complete 3,850-prompt set, raw citation records for every response, the source-type classification rules, the 200-response manual audit used to derive the 94% agreement figure, and the platform-level breakdown behind the citation rates quoted above.

Referenced research

• DeltaV Digital, AI Citation Study, 2026.
• Semrush, The Most-Cited Domains in AI: A Three-Month Study.
• Peec AI, Top Domains Cited by AI Search, as reported by Search Engine Land.

About Citations.press

Citations.press publishes research on how AI assistants select, cite and attribute sources, and operates the sector directories used as the study’s test set. Full methodology and raw citation records are published with the study.

Contact Info:
Name: Terry Owar
Email: Send Email
Organization: Citations Press
Website: http://citations.press/

Release ID: 89203396

In case of identifying any problems, concerns, or inaccuracies in the content shared in this press release, or if a press release needs to be taken down, we urge you to notify us immediately by contacting [email protected] (it is important to note that this email is the authorized channel for such matters, sending multiple emails to multiple addresses does not necessarily help expedite your request). Our dedicated team will be readily accessible to address your concerns and take swift action within 8 hours to rectify any issues identified or assist with the removal process. We are committed to delivering high-quality content and ensuring accuracy for our valued readers.