What ChatGPT receives from the web can depend on the source. OpenAI partner pages are supplied as pre-generated summaries, while other web pages appear as raw text snippets. The partner entries also come before the web results in the prompt sent to ChatGPT.
That difference matters for how information is presented. In ranked lists, the summaries described in the analysis named more entries in the first positions than entries near the bottom. The findings raise questions about how source format and placement may affect what ChatGPT can draw on when it answers.
Partner pages arrive in a prepared form
According to the source article, ChatGPT does not fetch partner pages live. Instead, partner content reaches it as summaries that have already been generated. Regular web pages are provided differently, as raw text snippets.
The ordering is also distinct: partner entries are placed ahead of all web results in the prompt. This means partner content is not only presented in a condensed form, but is also positioned before the other material described in the article.
These details describe how information is supplied to the system, rather than proving that every partner source will be cited in an answer. The article notes that partners do not always receive correct citations.
Ranked lists show a gap by position
Konitzny analyzed 1,000 cases involving ranked lists. In those cases, the summaries named 70 percent of entries in positions 1 through 3. For entries in positions 7 through 10, the figure was 23 percent.
The figures point to a difference in which entries were named across list positions. They do not, on their own, explain why the difference occurred or establish how every ChatGPT answer handles ranked information. The article reports the results for the cases analyzed.
Still, the pattern illustrates why the format of supplied material can matter. When a summary names entries near the top more often than those near the bottom, a model receiving that summary has a different set of details in front of it for those positions.
Placement promises and citation limits
OpenAI had promised partners more prominent placement in ChatGPT back in 2024. A study from Hamburg later found Springer media appearing far more often in the web interface than through the API.
The findings about summaries add another part to the picture: partner material is placed ahead of other web results and arrives in a prepared summary format. The article also cautions that prominence does not guarantee correct attribution. Even partners may not always be cited accurately.
For readers, the distinction is useful context when considering how ChatGPT presents information found online. A response may be based on material delivered in different forms and in a particular order, while citations can still be imperfect. The reported analysis offers a view of those inputs, but the source does not claim that it accounts for every answer or every partner page.