Do product pages get cited by AI?

A complete product or products token appeared in the public hostname or path for 272,801 of 8,316,692 cited-page links, or 3.2802%. That is about one in 30 links. The links covered 124,028 pages on 34,212 websites.

Wix Studio and DeltaV Digital classified product pages as a larger part of their source mixes. Those studies use content classifiers and narrower prompt or page samples. This study uses a strict public-address label across the full observed set. The result is a lower-bound prevalence measure, not evidence that product pages earn fewer citations than other formats.

At least one product-labeled page appeared in 161,616 of 860,399 cited answers, or 18.7838%. At least one appeared somewhere in the observed history for 14,277 of 18,449 organic prompts, or 77.3863%.

Link share and answer reach answer different questions. Product-labeled pages touched many prompts, but they remained a small part of the source mix inside those answers.

ChatGPT Search used 109,836 product-labeled links among 2,280,333 cited-page links, or 4.8167%. Google AI Mode used 162,965 of 6,036,359, or 2.6997%. The rate ratio was 1.7841 to 1, and the gap was 2.1169 percentage points.

Keep the engine attached to every product-page audit. The aggregate rate hides a material difference in the observed source mixes, but this comparison does not establish an engine preference or causal effect.

Prompts containing the complete word product or products recorded 33,641 product-labeled links among 264,724 cited-page links, or 12.7080%. Other prompts recorded 239,160 of 8,051,968, or 2.9702%. The rate ratio was 4.2785 to 1.

Page format should follow the buyer question. This cut shows an association with prompt wording. It does not show that adding a product label causes citations.

The ten leading websites supplied 26,673 of 272,801 product-labeled links, or 9.7775%. The leading site, liberia.ubuy.com, supplied 5,209, or 1.9095%. G2 supplied 4,078 and HubSpot supplied 2,767.

There was no single product-page gatekeeper in this corpus. Compare the websites that appear for the same market and prompts instead of copying one publisher.

Ecommerce websites supplied 63,667 product-labeled links among 676,197 cited-page links, or 9.4155%. General websites recorded 4.0671%, SaaS 3.8989%, documentation 2.0368%, editorial 1.9227%, and social 0.4394%.

Publisher type changes the useful baseline. The 21.4281-to-1 observed gap between ecommerce and social websites describes source composition, not publisher quality or causal preference.

Natural Resources recorded 935 product-labeled links among 12,405 cited-page links, or 7.5373%. Agriculture and Farming recorded 7.1334%, Consumer Goods 6.3195%, Financial Services 1.0273%, Gaming 0.6530%, and Blockchain and Cryptocurrency 0.4393%.

Use the industry cut to select the right comparison set. It does not hold prompt intent, publisher type, or engine mix constant and does not show that industry caused the rate.

The main hostname-or-path rule returned 3.2802%. Path labels alone returned 3.2784%, title labels alone returned 1.2001%, and a broader path rule that also included shop, store, catalog, item, and items returned 3.9153%. The final set had zero duplicate answer-source rows, and 176 product-labeled links had no title.

The rule misses product pages without a public label and can include non-product content inside a product directory. It does not read the page, classify its intent, test factual support, measure traffic, or estimate causal lift.

Product-labeled pages supplied 3.2802% of links, the engine rates differed by 1.7841 to 1, and the top ten websites supplied 9.7775% of product-page links.

Start with buyer questions that name a product or require a product decision. Keep the exact page, engine, prompt, publisher type, and industry together. Compare product pages with editorial and comparison pages for the same prompts. Rerun the fixed public-address rule next quarter before calling any difference movement.