I wanted to understand whether AI systems have distinct bot personalities, not in what they say, but how they show up on your site.
Here’s what I did:
- Asked the same prompt
- About the same site
- Observed behavior via server logs + citations
Here’s what I found 👀
What AI bots actually look like in the wild
| LLM | Citations | AI Bots Observed | Self-Identifying? | Example User Agent |
|---|---|---|---|---|
| ChatGPT | 10 | 7 | Yes | ChatGPT-User/1.0; +https://openai.com/bot |
| Gemini | 7 | 0 | No | Googlebot/2.1 (mobile Chrome UA) |
| Claude | 10 | 2 | Mixed | ClaudeBot/1.0; +claudebot@anthropic.com |
| Grok | 81 | 39 | No | Standard Chrome on macOS |
A few things stood out
Self-identification is optional
Only ChatGPT consistently raises its hand. Gemini appears to rely heavily on search ingestion. Claude will both self-identify and send anonymous fetchers.
“No bots” doesn’t mean no traffic
Gemini produced citations without any clearly identifiable Gemini bots. Other LLMs likely used cached or previously ingested data. This aligns with recent cost-optimization behavior in newer models (e.g. ChatGPT 5-2 auto).
Anonymous ≠ low impact
Grok generated the most citations while being fully anonymous at the UA level.
Each LLM has a bot personality
Fetch patterns, timing, headers, and disclosure habits differ significantly. We don’t fully understand what drives this behavior yet, but cost is clearly a major factor. We observed a sharp drop in bot activity after GPT-5 launched, followed by a partial reversal to improve response quality.
AI bot traffic is the best signal we have for understanding how your site is represented inside LLMs—but it’s still an imperfect science.