Which AI actually reads us.
Everyone is being told that AI search matters. Almost nobody can show which engines are actually reading them, because the evidence sits in server logs nobody opens. These are ours, published unedited.
On this store, AI crawlers fetch pages 1.9 times as often as Google and Bing combined. That is not a prediction about the future of search. It already happened, on a shop nobody has heard of.
Answer engines
| Crawler | Fetches | Belongs to |
|---|---|---|
| Meta-ExternalAgent | 1,545 | Meta |
| ClaudeBot | 1,534 | Anthropic |
| GPTBot | 1,114 | OpenAI |
| Bytespider | 186 | ByteDance |
| OAI-SearchBot | 154 | OpenAI |
| Claude-User | 45 | Anthropic |
| ChatGPT-User | 8 | OpenAI |
| PerplexityBot | 1 | Perplexity |
Search engines, for comparison
| Crawler | Fetches |
|---|---|
| Googlebot | 1,063 |
| Bingbot | 855 |
| YandexBot | 247 |
| Applebot | 217 |
| DuckDuckBot | 6 |
What this is not
One store, one category, one server, 16 days. It is evidence, not a study, and a shop selling something else in another country may see nothing like it. We are publishing it because the alternative - everybody guessing - is worse than one honest sample.
A fetch means a crawler requested a page and got it. It does not mean a model quoted us afterwards; that part is not measurable from a server log, and anyone telling you they can measure it from one is selling something. No IP addresses, no visitor data and no query strings are recorded - only the crawler's own published name and the path it asked for.
Why we bother
Same reason we publish every run's real numbers, the searches we could not answer and the copy we rejected. Anyone can claim they are transparent. Publishing the data is the only version of that claim which costs anything.
Figures update automatically from live server logs. Last recount covers the most recent 16 days on record.