ChatGPT Web Lookups Fall Sharply, SISTRIX Analysis Reveals
SISTRIX data shows a sharp decline in web lookup responses in anonymous sessions, dropping from over 15% to under 2.5% within two weeks.
ChatGPT Web Lookups Fall Sharply, SISTRIX Analysis Reveals
Its recent results show that over two weeks, the proportion of responses that prompted web lookups significantly decreased. In anonymous sessions, the rate reportedly decreased from over 15% of responses to less than 2.5%. Although the decrease seems significant, SISTRIX clarifies that its data only shows usage without logged-in accounts and does not offer a particular explanation.
What Changed
According to SISTRIX, it examines a lot of ChatGPT responses every day, employing a variety of prompts to assess the system's performance. The business saw that ChatGPT started web searches far less frequently than before during routine monitoring.
The drop was particularly noticeable when the chatbot was accessed without signing in. In these anonymous sessions, the proportion of answers completed with live web data fell sharply within the observed timeframe.
Important methodological information, such as the quantity of prompts evaluated, the total sample size, or the precise procedure utilized to identify when a web search has been performed, is not disclosed by SISTRIX. Additionally, the business mentions that Chat GPT has traditionally depended on Bing for real-time search results and alludes to rumors that Google data may have occasionally been utilized. However, it does not immediately link the decline it observed to any particular backend tweak.
Related Context
Microsoft Bing Search APIs Retirement
There is no proof that ChatGPT's internal systems were directly impacted by this shift, despite the fact that several third-party services switched to different data sources. However, this development is part of a larger ecosystem change that may be important.
Google’s SERP Access Changes
Additionally, according to SISTRIX, Google no longer supports the "num=100" search parameter, which restricts results to 10 listings per query. This change makes gathering data from extensive search results more labor-intensive. Although this development provides helpful context, the observed decrease in ChatGPT's online lookups cannot be directly linked to it.
Recent Chat GPT Product Notes
OpenAI published release notes on September 16 referencing “improvements to search in ChatGPT.” However, those notes did not explain any changes to underlying data sources or frequency of web retrieval. There is no proof that the date and SISTRIX's findings are related, despite the fact that they coincide.
Why This Matters
Users may start to notice fewer responses that mention current events or quote current internet sources if ChatGPT is using the web less frequently during anonymous sessions. Instead, answers could rely more heavily on the model’s internal training data. This change may have an impact on how frequently comments to users who are not logged in highlight current events or new developments. Although no available evidence supports this distinction, it's also likely that behavior varies for premium tiers like Plus or Enterprise.
Looking Ahead
SISTRIX's findings are exclusive to anonymous usage and are restricted to a brief period of time. As of right now, neither OpenAI nor SISTRIX has provided a clear explanation for the decline in the frequency of ChatGPT's live online searches. One independent investigation found a significant short-term decline in web search activity, which is the most measured result. To ascertain whether this is a transient variation or a more significant strategy change, further observation and confirmation will be required
