Product Pages Get 24.1% of AI Citations, Reddit 4.2%

US B2B content agency Ten Speed tallied the types of pages cited in AI answers and found that product pages accounted for the largest share, at 24.1%. Reddit, YouTube, forums and discussion threads combined made up 4.2%, most of which was Reddit. YouTube stayed at around 1%.

The data was collected with Peec AI, a tool that tracks which URLs ChatGPT, Perplexity, Claude and Gemini cite in their answers. Ten Speed ran 170 prompts against competitors similar to its own clients, which span fintech, physical security, hospitality and IT automation, and obtained 7,387 citation appearances.

The prompts mimicked the stage where a buyer who already knows the category compares specific vendor names, questions such as "Pipedrive vs. HubSpot" or "How does X handle SOC 2 reporting."

By type, after product pages at 24.1% came articles, combining blogs, news and PR, at 17.4%. Comparison pages and listicles each came in at around 13%, how-to guides fell just short of 9%, homepages were at 7.8%, and directory profiles such as G2 and Capterra were at 7.2%. Content written and owned directly by marketing teams accounted for 88.3% of all citations.

Comparison-type prompts made up 20% of all prompts but generated about 27% of citations, a 1.33x ratio. The original article suggested writing product pages and homepages in plain sentences that make clear what the product does, who it is for, and what it integrates with. It also suggested building comparison content covering two or more competitors first, and treating G2 and Capterra profiles as content to be managed as well.

Limits Uncovered by the Reporter's Follow-Up Questions

Before covering the study, Search Engine Journal, which wrote the article, sent Ten Speed six questions. One prompt breakdown table listed a total of 220, but every other chart used 170 as the denominator, and Ten Speed replied that 170 was correct and that it would fix the mislabeled chart. The comparison category, which had been classified differently across charts, was clarified as 34 prompts.

Ten Speed did not disclose even the range of how many clients or industries were involved. Its reason was that pairing a small, undisclosed client with an industry name could reveal who it is. The article noted that this means an outside researcher has no way to judge for themselves how broad the sample is.

This data collection did not break results down by platform, ChatGPT, Perplexity, Claude and Gemini, at all. The gap between 4% and 24% was reported only as descriptive statistics, without a statistical test. Ten Speed used a nonparametric test when comparing citation rates across page types, but not for this headline figure.

The original article listed three remaining limitations. The scope was limited to Ten Speed's B2B SaaS and professional-services clients, the data was pulled once at a single point in time rather than as a trend, and citation counts are not the same as clicks, demo requests or closed deals. Ten Speed said it does not yet have data to connect these citations to a sales pipeline.

In Korea

These numbers were not gathered using Korean-language questions and focus on B2B SaaS, so they cannot be applied directly to the Korean market. What can be borrowed here is the method. Put comparison questions from your own industry into several AI engines and count for yourself which types of pages get cited.

Source: Search Engine Journal, Greg Jarboe, September 3, 2026. Checked September 5, 2026. Read the original