ChatGPT's Web References Drop Dramatically After Model Update
ChatGPT's reliance on web-based references has significantly decreased, as observed by SEO tool SISTRIX. This change follows a recent update to the model, which has limited its access to real-time internet data.
SISTRIX's analysis shows that, over the past two weeks, the proportion of ChatGPT's responses that involved web searches dropped dramatically. Initially, over 15% of answers were completed via web searches. However, this figure has now fallen below 2.5%.
The reduction in web references is not limited to logged-in users. SISTRIX reports that anonymous sessions also show a decrease in live web searches triggered by ChatGPT. The cause behind this change remains unknown.
Meanwhile, Google has altered its search parameters, now returning only 10 results per request instead of the previous 'num=100'. This change increases the effort required to collect search engine results page (SERP) data at scale.
OpenAI's release notes mention 'improvements to search in ChatGPT' on September 16, but do not specify the backend sourcing changes. Microsoft, on the other hand, has announced the retirement of Bing Search APIs, effective August 11.
The reduction in web-based references by ChatGPT, as observed by SISTRIX, is likely due to the model's update limiting real-time internet access. This change, along with Google's alteration to search parameters and Microsoft's retirement of Bing Search APIs, may impact the way data is collected and used in AI models like ChatGPT.
