The Diurnal Shape of AI Crawl Traffic
Every site operator with a CDN dashboard already knows that human traffic has a daily rhythm. AI crawler traffic has one too, but it looks different. This page reports what that rhythm actually looks like for presenc.ai in April 2026, hour by hour, broken down by major AI bot. The data comes from the same Cloudflare Worker plus D1 logging stack used for the rest of our first-party crawl analytics research.
Hourly Volume by Bot
The table below summarises which hours of the day (UTC) each major AI bot is most active during April 2026. "Peak hours" reflect the top quartile of hourly volume for that bot.
| Bot | Peak hours (UTC) | Quietest hours (UTC) | Diurnal pattern |
|---|---|---|---|
| GPTBot | 02:00 to 08:00 | 14:00 to 18:00 | Strong overnight bias, scheduled pulls |
| OAI-SearchBot | 12:00 to 22:00 | 04:00 to 08:00 | Mirrors human business hours, search-index style |
| PerplexityBot | 14:00 to 22:00 | 02:00 to 06:00 | Tracks user-query patterns, business-hours weighted |
| ClaudeBot | Episodic, no clear daily rhythm | Most days are quiet | Bursty, often days-long gaps then heavy fetches |
| Google-Extended | Distributed across day | None pronounced | Smooth, low variance |
| ChatGPT-User | 13:00 to 21:00 | 03:00 to 07:00 | Tracks human use of ChatGPT |
Why GPTBot and OAI-SearchBot Look Opposite
GPTBot does most of its work overnight UTC. OAI-SearchBot does most of its work during the day. This is consistent with their stated purposes: GPTBot is a training-data crawler that runs scheduled pulls without a real-time obligation, so OpenAI can run it when origin servers are quietest and infrastructure is cheapest. OAI-SearchBot serves a live search index that needs fresh data when humans are searching, so its work is concentrated in the hours those humans are awake.
The practical consequence for serving strategy is that GPTBot can be safely rate-limited during peak human hours without affecting any live AI product. OAI-SearchBot cannot, because its fetches feed real-time answers that users see seconds later.
PerplexityBot and ChatGPT-User Track Human Activity
PerplexityBot peaks in the late afternoon and early evening UTC, which corresponds to working hours in Europe and the Americas. The shape closely matches the diurnal shape of human Perplexity usage. Similarly, ChatGPT-User (the on-demand fetcher) peaks during the same hours and is essentially absent overnight. Both bots are fetching in response to current human queries, so their pattern is a low-amplitude version of the human pattern itself.
ClaudeBot Is the Outlier
ClaudeBot does not have a clean daily rhythm. Most days see very little ClaudeBot activity, then a single day or 1-2 day window will produce 5x to 10x the normal volume. This pattern is consistent with batch crawls run by Anthropic for training, evaluation, or refresh purposes rather than continuous indexing. The implication is that you cannot predict ClaudeBot load by time of day, but you can predict that overall ClaudeBot volume per month will be roughly stable.
Implications for Origin and Edge Strategy
Sites that have ever struggled with AI crawler load almost always struggle with two specific patterns: the GPTBot overnight pull and the ClaudeBot burst. Both are concentrated in narrow windows. The right defence is not blocking AI crawlers wholesale, since that destroys downstream visibility. The right defence is rate-limiting at the edge during those specific windows for those specific bots, while leaving OAI-SearchBot, PerplexityBot, and ChatGPT-User unconstrained because they directly feed live human-facing answers.
Methodology
Data source: Cloudflare Worker logging every inbound request to presenc.ai during April 2026 to a Cloudflare D1 table. Hourly buckets are constructed in UTC. Bot classification is by declared user agent string, cross-referenced against published documentation from each operator. Peak and quiet hours are defined as the top and bottom quartile of hourly volume for that bot during the period.
How Presenc AI Helps
The hourly shape of your own AI crawler traffic is something most teams have never measured. Presenc AI customers can run the same Cloudflare Worker against their own zone, route logs into managed storage, and read the resulting hourly view in the same dashboard as their AI visibility scores. The combination of "what AI says about you" and "when AI fetches you" is the operational layer most brands are missing.