What Does AI Cite for Price Questions? Company Sites at 83%, Pricing Pages at 16%
Published
This article was first published in Korean by Intrix Lab on September 17, 2026.
When AI answered questions about prices or costs, most of its supporting sources were company sites, but not the pricing pages themselves. As of 2026-09-14, classifying 1,397 citations from 460 responses to 33 price and cost queries across 16 Intrix CT studies showed that company overview, service, and article pages accounted for 73%; pages with pricing-related URL paths, 16%; and third parties such as blogs, communities, and platforms, 10%. Profound's research on US SaaS buyers reported that external review sites shape AI's perception of pricing. For Korean-language local service queries, that role was filled by company sites' "non-pricing pages."
What did we measure, and how?
We used only responses from citation studies already collected and submitted no new questions. We divided citation URLs from 460 responses to 33 price and cost queries across 6 models into three categories: pricing pages, third-party sites, and other company pages. Pricing pages were identified by their URLs, so overview pages containing prices in their body text counted as "other pages." We return to this limitation below.
What were the numbers?
| Category | By response | By citation count |
|---|---|---|
| Answered without citations | 130 (28%) | - |
| Cited pages with pricing-related URL paths | 131 (40% of 330 responses with citations) | 228 (16%) |
| Cited other company pages | 315 (95%) | 1,023 (73%) |
| Cited third-party sites | 79 (24%) | 146 (10%) |
By model, the share that cited pricing pages among responses with citations was 27% for GPT Terra, 35% for GPT Luna, and 53% for Gemini 3.7. Gemini selected pricing pages more often, but it also had the highest share of responses without citations.
Why did the results differ from Profound's?
The populations differ. Profound examined pricing research by US software buyers, a market where review sites such as G2 and Capterra compile pricing information. Our queries mix Korean-language local services and B2B, where third parties that aggregate prices are rare. So AI turns to company sites. Even within company sites, however, overview and service pages served as sources more often than pricing pages. This appears to be because pricing pages are often absent or, when present, contain only table images or scripts that AI cannot read, but these data cannot distinguish the causes.
What should change in practice?
The source used for price questions is a "page that contains prices," rather than a "pricing page." If a price list appears only on a separate page and not in the body text, an overview page is cited instead; if that page has no prices, AI gives numbers from elsewhere. If only distributor or partner sites publish prices, those sites get cited. The recommendation is simple: put prices and conditions together in text on pages that are frequently cited.
Limitations
Pricing pages were identified by URL, so counting again based on page content could increase the pricing-page share. The 16 studies were conducted by Intrix and are not a sample of the entire industry. The language and market differ from Profound's study: these are observations of the same metric in different populations, not a refutation. These findings are not a basis for guaranteeing specific citations or rankings.
Frequently asked questions (FAQ)
Q. Will creating a pricing page get it cited?
The condition is that a page with prices written in text becomes a citation candidate, rather than simply having a page. Adding prices to an overview page that is already being cited is the faster approach.
Q. If third-party citations account for 10%, can we ignore review sites?
The share was low in this population, but it varies by industry. The answer is to check directly using pricing queries for your own industry.
Q. What are the 28% of responses that answered without citations?
These responses either answered from memory without searching or searched but did not attach sources. In these cases, there is no way to verify the source of the prices AI gives.