AI NewsGo-to-marketReported

AI engines cite pages that lead with the answer, 193,507-answer study finds

TJ Robertson studied 193,507 AI answers from ChatGPT, Google AI Mode and AI Overviews. Cited pages opened with a direct answer at 20 percent, against 5 percent for pages the engines passed over.

AI News

Editorial2 min read

LinkedInX

Why it mattersA page that buries the answer is a page an AI engine quotes less. The research also shows that most cited pages are product, home and category pages, not blog posts.

A page can rank in Google's top 10 and still get passed over by AI engines. TJ Robertson is a researcher who measures what AI quotes. His new study examines what cited pages do differently. Search Engine Journal reported the findings this week.

Robertson tracked 4,167 buyer questions through ChatGPT, Google AI Mode and Google AI Overviews. He ran eight weekly snapshots over about two months. The run collected 193,507 answers. That is 63,033 from AI Mode, 62,931 from AI Overviews and 67,543 from ChatGPT. There were 943,673 citations to 209,859 different URLs. 69,401 pages were downloaded and read by blind AI raters in two rounds.

What the citations opened with

Cited pages were more likely to open with a direct answer. Robertson says cited pages opened with a direct answer 20 percent of the time. Passed-over pages, which the engines ranked but did not cite, opened with a direct answer 5 percent of the time. Of the pages the engines pulled up, 23 percent of the cited ones led with a direct answer. For the uncited ones the share was 14 percent.

Completeness pulled the same way. On a 1 to 5 scale, cited pages scored 2.3 and passed-over pages scored 1.7. About half of the passed-over pages never answered the question. The share for cited pages was 23 percent.

Which pages the engines quote

The split by page type is the result Robertson leads with on his own site. 28 percent of cited pages were product pages. 12 percent were homepages. 10 percent were category pages. The rest were guides, blog posts and other pages. Robertson writes that most of what AI cites is the pages that describe what a business sells.

Robertson states two limits on his own data. The sample runs on buyer questions in nine business sectors. The result may not hold for every topic. And association is not cause. A page that leads with a direct answer may share other traits. One of those other traits may be what the engines pick up. Only a controlled test can separate the two.

A product page that opens with a one-sentence description is cheap to write. It also matches what the engines appear to reward. The signal in Robertson's data is a clear one.

Source

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project