Additional commentary by Emily Lyons
Welcome back to AI Pulse, our blog series in which we share insights on the state of the AI bots and the AI agent landscape.
If you missed any of the previous blog posts in the AI Pulse series, be sure to check them out:
This post’s topic: OpenAI. We’re unpacking the volume of OpenAI traffic hitting Akamai and how it outpaces every other AI platform.
OpenAI leads the pack
Throughout our AI Pulse series, we’ve highlighted just how much of today’s AI bot traffic is coming from OpenAI. Four vendors make up more than 90% of all AI bot requests hitting Akamai customers — and OpenAI alone accounts for 42.4% which is more than twice the volume of the next-largest vendor (Figure 1).
OpenAI operates one of the most widely adopted AI ecosystems, spanning training crawlers, fetchers, and search-focused bots. Its models power millions of downstream applications, integrations, and agents that continuously pull fresh content, perform user-driven lookups, and maintain model performance.
As customer use grows, so does the volume of automated requests, pushing OpenAI far ahead of other AI platforms in overall traffic.
A closer look
When we dig into OpenAI’s traffic patterns, a clearer picture emerges: The volume isn’t coming from one bot, but from a diverse set of automated behaviors that each play a different role in OpenAI’s ecosystem.
Today, four main OpenAI bots show up consistently across Akamai’s network (we’ll save our discussion of the new Atlas AI browser for another post):
ChatGPT-Agent: A newer, lighter bot tied to agentic task execution. Its footprint is small and stable, reflecting early-stage adoption of autonomous agent workflows.
GPTBot: OpenAI’s long-standing training crawler, designed to gather data used to refine and update foundation models. Its traffic is steady and predictable, matching established training cycles.
ChatGPT-User: The fetcher responsible for real-time retrieval when users or integrated apps request fresh information. This is one of the fastest-growing categories, fueled by expanding API use and increasing reliance on live data.
OAI-SearchBot: OpenAI’s search crawler, indexing content to support searchlike capabilities inside ChatGPT and other OpenAI products. Its growth aligns with OpenAI’s investment in more structured, search-driven experiences.
Each bot serves a distinct operational purpose, creating a layered pattern of activity rather than a single monolithic source of traffic (Figure 2). This variety is a defining characteristic of OpenAI’s presence on the web and a key reason that their traffic looks fundamentally different from other AI vendors.
How OpenAI’s traffic differs from everyone else’s
When we continue to explore the breakdown of OpenAI’s four bots, the next thing that becomes clear is how unusual OpenAI’s traffic profile is compared with the traffic of all the other AI platforms.
OpenAI isn’t simply operating across training, user-driven fetching, and search crawling; it has significant traffic in all three categories, creating a multilayered pattern that no other vendor comes close to matching (Figure 3).
Specialists vs. a generalist
Most other platforms behave like specialists. For example:
Anthropic is overwhelmingly training, with ClaudeBot making up nearly all activity.
Perplexity is almost entirely defined by PerplexityBot, a search crawler with occasional spikes tied to index refreshes. In earlier days, Perplexity-User accounted for a small but substantial amount of traffic.
OpenAI, meanwhile, looks like a generalist; each bot type contributes meaningful volume, which produces a stacked, blended traffic pattern that reflects how many different surfaces OpenAI now touches.
This odd mix matters because it reveals a platform that’s powering multiple types of AI behavior at once: retrieval for user queries, structured crawling for search, and steady collection for model improvement. That diversity not only drives higher total volume but also gives OpenAI a broader and more persistent presence across Akamai customers than any other AI vendor today.
Mitigation patterns across OpenAI bots
When we look at how OpenAI bots are being managed across Akamai customers, a clear pattern emerges: Fetchers trigger far more mitigations than training or search crawlers. While GPTBot (which is a training crawler; Figure 4), and OAI-SearchBot (which is a search crawler; Figure 5) show some level of blocking or slowing, the volume and percentage of mitigations are dramatically higher for ChatGPT-User, OpenAI’s user-driven fetcher (Figure 6).
This makes sense as fetchers generate high-intent, high-frequency requests tied to real-user activity or application-level calls. They often touch sensitive endpoints, inventory data, pricing, or content APIs, and customers are more likely to intervene when this traffic becomes noisy or excessive.
How customers respond to ChatGPT-User traffic
By zooming in on ChatGPT-User mitigation specifically, we see clear growth across all mitigation types (Figure 7).
Tarpit grew steadily and consistently over the past few months.
Serve slow saw a noticeable uptick recently, suggesting more customers are opting to throttle rather than flat-out block.
Delay increased in August as users experimented with soft friction methods.
Deny or outright block experienced the steepest rise, ballooning in October as overall traffic volume spiked.
The trend line tells a simple story: As ChatGPT-User activity accelerates, customers are responding with a wider mix of defenses.
Advanced mitigations and controlled friction
What stands out even more is the distribution of advanced mitigations. While ChatGPT-User has only slightly more total triggers than GPTBot and roughly twice the triggers of OAI-SearchBot, it accounts for the overwhelming majority of advanced mitigation actions, including tarpit, serve slow, serve alternate content, and challenge flows.
In fact, nearly two of every three advanced mitigations across all AI bots are attributed to ChatGPT-User (Figure 8).
What stands out even more is the distribution of advanced mitigations. While ChatGPT-User has only slightly more total triggers than GPTBot and roughly twice the triggers of OAI-SearchBot, it accounts for the overwhelming majority of advanced mitigation actions, including tarpit, serve slow, serve alternate content, and challenge flows.
In fact, nearly two of every three advanced mitigations across all AI bots are attributed to ChatGPT-User (Figure 8).
The so what
OpenAI is the largest source of AI bot traffic hitting customer sites, and its bots (especially ChatGPT-User) are growing fast and triggering the highest levels of mitigation.
For most organizations, the immediate priority is getting visibility and control over this traffic. Businesses should be actively deciding how they want to manage that goal: which bots to allow, tarpit, deny, or block; how to protect sensitive data; and, for publishers and content owners, whether to pursue licensing or usage agreements.
In short, OpenAI’s footprint is too large to ignore, and companies need a clear strategy for how they want these bots to interact with their applications and content.
Learn more
To learn how to gain visibility and control over your organization’s AI bot traffic, contact an Akamai expert.
Tags
