We changed one page. PerplexityBot went from 42 requests to 511.
PerplexityBot made 511 requests to Genymotion’s AI site in the week of March 10–17, up from 42 the week before. That’s a 12x increase. It crawled 172 Q&A pages, 256 GEO pages, and swept the entire topic taxonomy twice. Five days earlier, we had redesigned the site’s index page. We can’t prove the two are connected. But PerplexityBot had been stuck on the same four pages for six weeks, and then it wasn’t.
For months, PerplexityBot was the bot we couldn’t figure out. It showed up in early February, visited a rooting guide, the Android 16 beta page, the SaaS product page, and the system requirements page. Then it came back and visited the same pages again. And again. Total across six weeks: never more than 42 requests in a given week. No Q&A deep dives. No topic pages. No systematic indexing. On March 9, we redesigned the AI site’s index page. On March 10, PerplexityBot ran its first real content crawl.
Key findings
- PerplexityBot: 511 requests (Mar 10–17), up from 42 the prior week (12x)
- PerplexityBot crawled 172 Q&A pages and 256 GEO pages (84% content)
- PerplexityBot swept the topic taxonomy twice (March 15 and March 17)
- OpenAI SearchBot: 328 requests, up from ~75 (4x)
- Combined search index traffic (PerplexityBot + SearchBot): 839, or 71% of ChatGPT-User citation volume
- ChatGPT-User: 1,176 requests, stable
- ClaudeBot: 123 requests, 0 Q&A pages, 0 Claude-SearchBot traffic
The data
All LLM bots (Mar 10–17, 2026)
| Bot | Requests | Category | Change vs. prior week |
|---|---|---|---|
| ChatGPT-User | 1,176 | Citation | Stable (~1,200) |
| PerplexityBot | 511 | Search index | 12x (was 42) |
| OpenAI SearchBot | 328 | Search index | 4x (was ~75) |
| OpenAI GPTBot | 171 | Training | Moderate growth |
| ClaudeBot | 123 | Training | Down from 505 |
| CCBot | 100 | Training | Stable |
| ByteSpider | 69 | Training | Down |
| Meta AI | 52 | Training | Stable |
PerplexityBot content breakdown
| Content type | Requests | % |
|---|---|---|
| GEO Pages | 256 | 50.1% |
| Q&A Pages | 172 | 33.7% |
| Topic Pages + Other | 61 | 11.9% |
| Robots.txt | 12 | 2.3% |
| Homepage | 4 | 0.8% |
84% of PerplexityBot’s requests hit content pages. It skipped the homepage almost entirely. It went straight to the content.
What we changed
On March 9, we redesigned the AI site’s index page at rozz.genymotion.com.
The old index was an infrastructure page. It listed API endpoints, content counts, and navigation links. It told a human reader “here’s 177 Q&A pages and 450 GEO pages, here’s how to browse them.” But AI crawlers don’t use JSON APIs. They GET HTML pages. When a bot landed on the old index, it got a page that described the site’s architecture without giving it any signal about what Genymotion actually is or which content to read first.
We had already identified this as a problem. In Entry #6, we found that 28% of ChatGPT-User sessions hit the index and stopped. Dead ends. The bot arrived, found no useful signal, and left.
The new index opens with a product description: what Genymotion is, what it does, who uses it. Then a topic directory with inline summaries for each topic. Not just “Subscription & Billing” but context about pricing tiers, free personal use, SaaS vs Desktop costs, plan differences. Enough signal for a crawler to either answer a query from the index itself or know exactly which page to fetch next.
What PerplexityBot did after the change
Day 1: March 10, content burst
The day after the index revamp, PerplexityBot ran a content crawl at 21:03 UTC. 50+ pages in under 20 seconds. GEO pages and Q&A pages. ARM transition content, Android 16 pages, pricing, installation, virtualization, case studies, release notes. It came from 8 different IP addresses running in parallel.
This was the first time PerplexityBot had ever crawled more than 4 pages in a single session on the AI site.
Day 5: March 14, Q&A deep dive
PerplexityBot returned at 21:02 UTC and crawled 40+ Q&A pages in a single burst. One page every 1–2 seconds. The questions it grabbed: pricing (how much does it cost, pricing limits, free trial, discounts, monthly cost), installation (Windows steps, Linux steps, macOS, Ubuntu, general setup), compatibility (MacBook Air 8GB RAM, ARM support, VirtualBox dependency, Windows 10 Intel), and product questions (50 concurrent devices, license keys, online vs download).
This is the content format that drives 75% of ChatGPT citations. PerplexityBot had never touched a Q&A page before this week.
Day 6: March 15, topic sweeps + GEO content
Two sessions. At 21:21, PerplexityBot swept 14 topic pages in rapid succession, all within seconds of each other. These are CollectionPage-marked pages that list every piece of content in a topic: android-development-toolkit, ci-cd-toolchain, cloud-platform-ecosystem, hypervisor-platforms, licensing-eula, saas-paas-tools, and more.
Then at 22:59, it came back for more GEO pages: ARM webinar recordings, CircleCI integration, mobile testing, VPN setup, proxy configuration, biometric authentication, Bitrise CI, Applitools visual testing.
Day 8: March 17, second topic sweep
Another sweep of topic pages at 07:13. genymotion-virtual-devices, gpu-arm-hardware, technical-support-troubleshooting, android-testing-stack, network-system-security, marketplace-cloud-providers.
PerplexityBot reads the map and the pages
PerplexityBot is the second bot we’ve seen systematically crawl topic pages. ClaudeBot did it first, on March 2, sweeping 13 topic pages in five minutes.
The difference: PerplexityBot did the topic sweep AND the content deep dive. It read the map and then read the chapters. ClaudeBot read the map and stopped.
Here’s how the bots compare on content type distribution this week:
| Bot | GEO Pages | Q&A Pages | Topic/Other | Homepage |
|---|---|---|---|---|
| PerplexityBot | 50% | 34% | 12% | 1% |
| OpenAI SearchBot | 45% | 24% | 16% | 3% |
| ChatGPT-User | 9% | 34% | 20% | 37% |
| ClaudeBot | 1% | 0% | 64% | 5% |
PerplexityBot and OpenAI SearchBot share a similar profile: content-heavy, minimal discovery overhead. Both are search index bots. Both skip the homepage and go straight to content pages.
ChatGPT-User has a different pattern: it uses the homepage as a discovery hub (37% of requests), then drills into Q&A pages (34%). This makes sense. ChatGPT-User operates during live conversations. It needs a starting point.
ClaudeBot is in a category of its own. 64% of its requests are robots.txt, sitemaps, and topic pages. 0% Q&A. 1% GEO content. It’s still in discovery mode.
The search index layer is catching up
PerplexityBot (511) and OpenAI SearchBot (328) together made 839 requests this week. That’s 71% of ChatGPT-User’s citation volume (1,176).
For context, search index bots were negligible two months ago. SearchBot was at 66 requests in January. PerplexityBot was at 14. Combined: 80. Now: 839. A 10x increase.
| Week | Search index bots | ChatGPT-User citations | Ratio |
|---|---|---|---|
| Jan 3–Feb 2 | 80 | 42 | 1.9x |
| Feb 17–24 | 89 | 1,329 | 0.07x |
| Mar 10–17 | 839 | 1,176 | 0.71x |
In January, search index activity exceeded citation activity. Then ChatGPT citations exploded and the search bots fell behind. Now the search index layer is catching back up.
If the ChatGPT pattern holds (deep indexing followed by citations approximately three weeks later), Perplexity citations should start appearing around early April.
ChatGPT: steady state
ChatGPT-User made 1,176 requests this week. The trajectory:
| Period | ChatGPT-User | Change |
|---|---|---|
| January (30 days) | 42 | Baseline |
| Feb 2–9 | 345 | 8x |
| Feb 10–17 | 1,077 | 3x |
| Feb 17–24 | 1,329 | +23% |
| Mar 3–10 | ~1,200 | Stable |
| Mar 10–17 | 1,176 | Stable |
The exponential phase is over. ChatGPT citations have plateaued around 1,200 requests per week. That’s roughly 44 content sessions per day, each averaging 3.1 pages. 40% of sessions involve multiple turns: users asking follow-up questions in the same conversation.
This week’s deepest session: 22 pages, 8 turns, 38 minutes. A user in the US-Central region started with macOS system requirements, drilled into Android versions, explored device templates, checked plugin setup, and came back multiple times. This isn’t a lookup. It’s a full product evaluation happening inside ChatGPT, with the AI site as the knowledge source.
The growth frontier has moved. ChatGPT is now a steady-state citation channel. The next growth story is Perplexity.
Claude: still waiting
In Entry #5, we wrote about ClaudeBot’s distinctive behavior: sweeping the topic taxonomy, reading the site’s organizational structure rather than individual content pages. We called it “reading the map, not the pages.”
Three weeks later, ClaudeBot is still reading the map. And only the map.
This week: 123 requests. Content breakdown: 64% robots.txt/sitemap/topic pages. 0 Q&A pages. 1 GEO page.
The contrast with PerplexityBot is direct. Both bots did topic sweeps. PerplexityBot followed up with 172 Q&A pages and 256 GEO pages. ClaudeBot followed up with nothing.
No Claude-SearchBot traffic was observed this week. Zero. Claude has a documented search bot (Claude-SearchBot) that builds the retrieval index for Claude’s web search feature. It hasn’t visited the AI site at all.
| Bot event | GPTBot | PerplexityBot | ClaudeBot |
|---|---|---|---|
| First topic sweep | Never | March 15 | March 2 |
| First Q&A deep dive | Jan 7 (mass crawl) | March 14 | Never |
| Search bot active | SearchBot: 328/week | PerplexityBot IS the search bot | Claude-SearchBot: 0 |
| Current citation rate | 83% | 17% (expected to grow) | 21% (not growing) |
PerplexityBot treated the topic sweep as step one of a three-step process: learn the structure, then read the Q&As, then read the GEO pages. ClaudeBot treated the topic sweep as the whole job.
We don’t know why. We don’t know if this is a crawl budget decision, an architectural difference in how Anthropic builds its retrieval index, or something about the AI site that works for Perplexity’s crawler but not for Claude’s. What we do know: until ClaudeBot or Claude-SearchBot starts reading content pages, Claude’s citation rate is unlikely to improve. We’re watching this one.
Get This for Your Site
ROZZ builds this infrastructure automatically. AI site. Q&A pages from your chatbot. Schema.org markup on every page. Topic taxonomy that search index bots actually use.
$997/month | ChatGPT steady at 1,200/week. Perplexity indexing at 511 and growing.
→ Book a call | → See how it works | → rozz@rozz.site
→ Data source:
CloudFront access logs for rozz.genymotion.com, March 10–17, 2026.
Bot classification based on User-Agent strings.
ChatGPT session analysis via IP subnet and edge clustering from ChatGPT-User/1.0 log entries.