OpenAI's automated web crawling activity has roughly tripled since the launch of GPT-5, according to new analysis of billions of crawler logs. The company's search bot is now generating more activity than its training bot, a reversal from the pre-GPT-5 era that reveals how the latest model retrieves information differently than its predecessors.

What Changed After GPT-5 Launched in August 2025?

Researchers at Botify, a web analytics firm, analyzed approximately 7 billion OpenAI bot log events spanning November 2024 through March 2026. They tracked three distinct OpenAI user agents, or automated crawlers, to understand how the company's data collection patterns shifted after GPT-5's release in August 2025.

The findings paint a clear picture of changing priorities. OAI-SearchBot, which retrieves content when ChatGPT performs web searches, recorded about 3.5 times more events after the GPT-5 launch. That translates to roughly 2.2 billion additional crawler events in the dataset. GPTBot, which collects training data for model improvement, recorded about 2.9 times more events over the same period, adding another 1.8 billion events.

Before GPT-5, the two bots ran at roughly equal volumes in the dataset, with a ratio of about 0.95 search events per training event. After GPT-5, that ratio rose to about 1.14, meaning search activity now outpaces training activity.

Why Does This Matter for How ChatGPT Works?

The shift in crawling patterns suggests OpenAI is fundamentally changing how GPT-5 answers user questions. Rather than relying primarily on knowledge learned during training, the model appears to be pulling more answers from live web searches in real time. This approach has practical implications: it means ChatGPT can provide more current information, but it also requires constant access to the internet.

Interestingly, the third user agent tracked, ChatGPT-User, moved in the opposite direction. This bot fires when a ChatGPT session fetches a page on behalf of a user, and it recorded a 28% drop in activity between December 2025 and March 2026. Researchers offered two possible explanations: either fewer ChatGPT sessions are triggering real-time page fetches, or OpenAI is relying more on cached or indexed resources, reducing the need to fetch pages fresh each time.

How to Understand OpenAI's Crawling Strategy Across Different Industries

• Healthcare Sites: Experienced the largest increase in OAI-SearchBot activity at approximately 740% after GPT-5 launched, suggesting health queries increasingly rely on live search results.
• Media and Publishing: Saw about 702% more search bot activity, with a 256% difference favoring search over training crawls, indicating news-related queries trigger live web searches.
• Retail and Software: Recorded 190-216% increases in search activity, with these sectors leaning toward live search for product and software queries.
• Travel Sites: Had the smallest rise at 30%, suggesting travel queries may rely more on trained knowledge than current web data.

The pattern suggests OpenAI routes different types of questions through different pathways. News inquiries trigger live search, while health and product queries may rely more on trained knowledge.

Even after tripling, OpenAI's crawl activity remains much smaller than Google's dominance. In Botify's most recent 30-day window, Googlebot registered 18.2 billion events compared with 887 million events from OpenAI's crawlers combined. That puts OpenAI at approximately 4% of Google's crawl volume, up from about 1.38% a year earlier. The gap is closing, though Google's crawl is still roughly 20 times larger in absolute terms.

What This Means for Website Owners and Content Creators

The findings have important implications for how websites should manage AI bot access. Sites that block only GPTBot are not blocking OAI-SearchBot, the crawler OpenAI uses to surface websites in ChatGPT search answers. Conversely, sites that block OAI-SearchBot may be excluding themselves from appearing in ChatGPT search results.

The data comes from Botify's enterprise client dataset, which skews toward large websites in retail, ecommerce, technology, publishing, travel, and marketplaces rather than representing the broader web. However, the findings align with patterns other vendors have reported. An Alli AI analysis found OpenAI's ChatGPT-User made 3.6 times more requests than Googlebot in a smaller sample, while a Hostinger analysis found OAI-SearchBot's website coverage reaching 55%.

The broader takeaway is that AI training crawls and AI search crawls need to be measured and managed separately, especially as OAI-SearchBot activity continues to grow. Website operators who want to appear in ChatGPT search results should ensure they're not blocking the search bot, while those concerned about training data collection can target GPTBot specifically.