Claude crawled our sites 258 times last week and didn't read a single post.
We watched 7 AI bots crawl 3 live domains for a week. Each one behaved completely differently. None of it shows up in Google Analytics. Here's what the server logs actually say.
Domains
3
Window
Apr 5 to Apr 12, 2026
Requests
29,556
AI-identified
1,139
The cast
Seven bots. Seven different habits.
The mistake is treating AI traffic as one bucket. These crawlers do not behave like a channel. They behave like separate systems with separate incentives.
Total hits are not attention. Claude was active, but almost none of that activity reached real content.
ChatGPT
The homepage tourist
367
Hammers the root of your site and rarely ventures deeper. Ignores your sitemap. Barely touches robots.txt.
59%Hits to /
Meta AI
The silent reader
293
Reads blog posts and almost nothing else. Didn't touch robots.txt once. Nobody is optimizing for it.
73%Hits to /blog
Claude
The map-maker
258
Spent 76% of its budget re-reading robots.txt and sitemap.xml. Didn't fetch a single blog post.
76%robots.txt + sitemap.xml
Applebot
The browser
110
Downloads JavaScript bundles, stylesheets, manifests. Treats your site like a real browser render.
84%JS / CSS / assets
Amazonbot
The blog scraper
89
No homepage, no sitemap, no robots. Just content, over and over again.
92%/blog/* posts
Perplexity
The ghost
14
Fourteen total requests in 7 days. That looks like a scheduled heartbeat, not an index build.
exact2 hits per day
Gemini
The phantom
2
Two requests in a week, both on the same day. Google's flagship AI is essentially absent from our logs.
2 hitsone path, one day
Findings
Five things you won't read in a client-side analytics dashboard.
Finding 01
2 hits/day
Perplexity's daily cadence was exact.
Perplexity visited our sites twice a day, every day, for seven days. Not 1. Not 3. Two. That is not crawling behavior. It looks like a scheduled heartbeat, not an index build.
Observed paths
robots.txt and sporadic content fetches
Server-side read
14 requests across 7 consecutive days
Finding 02
0 posts
Claude hit us 258 times and never read content.
Claude spent 89 requests on /sitemap.xml, 85 on /robots.txt, and zero on any actual blog post or landing page. If you only look at total hits, it's one of the most active bots. If you look at what it actually read, it's the least.
Observed paths
/sitemap.xml, /robots.txt
Server-side read
174 map requests, 0 blog requests
Finding 03
5x more
Meta AI read far more blog posts than ChatGPT.
Across the same site, Meta AI fetched 186 blog posts while ChatGPT fetched 36. Meta AI is almost invisible in AI-SEO conversations, and it's arguably reading your content more carefully than OpenAI is.
Observed paths
/blog/* posts
Server-side read
186 Meta AI blog fetches vs 36 ChatGPT fetches
Finding 04
84% assets
Applebot was rendering, not scraping.
Applebot spent 84% of its traffic on JavaScript bundles, CSS files, and web manifests. Most AI bots fetch HTML and stop. Applebot is behaving like a headless browser, which is how Apple Intelligence will need to parse modern JS-heavy sites.
Observed paths
.js, .css, manifest, image assets
Server-side read
92 of 110 requests were render-supporting assets
Finding 05
39%
A huge share of traffic was neither human nor AI.
39% of requests came from generic scrapers: SEO tools, archive bots, unidentified crawlers. Before you even start worrying about AI, roughly four out of every ten visitors to your site are bots nobody is tracking.
Observed paths
mixed routes across three domains
Server-side read
generic automation exceeded AI-identified traffic
Full table
7 days. 3 sites. Every bot.
| Bot | Hits | Dominant path | Concentration | Behavior |
|---|---|---|---|---|
| ChatGPT | 367 | / (homepage) | 59% | Homepage hammer |
| Meta AI | 293 | /blog/* posts | 73% | Content reader |
| Claude | 258 | sitemap + robots.txt | 76% | Maps, never reads |
| Applebot | 110 | JS / CSS / assets | 84% | Browser renderer |
| Amazonbot | 89 | /blog/* posts | 92% | Content reader |
| Perplexity | 14 | robots + sporadic | exact cadence | Heartbeat only |
| Gemini | 2 | one path, one day | effectively zero | Absent |
Concentration means the share of a bot's total hits that landed on its single most-visited path or path type. High concentration means the bot is predictable; low concentration means it roams.
Methodology
Observed via SeeLLM's Cloudflare Worker proxy running on applyr.co, aisearchmonitor.com, and seellm.link between April 5 and April 12, 2026. Every request was captured at the edge, before the browser or JavaScript was involved.
- Google Analytics can miss this because many AI crawlers and agent requests never execute client-side JavaScript.
- robots.txt tells you who is allowed. Server logs tell you who actually came and what they read.
- AI traffic is not one channel. ChatGPT, Claude, Meta AI, Applebot, Amazonbot, Perplexity, and Gemini all behave differently.
From report to action
Want to see what your own server logs say?
Run a free scan first. When the signal matters, move from one-off checks to monitoring the pages tied to traffic, pipeline, and revenue.