What results does Rozz deliver? Genymotion case study.

Detailed Explanation

The client

Genymotion makes an Android emulator used by developers, QA teams, and enterprises for app testing. They had a comprehensive website with stable Google rankings. But when prospects asked ChatGPT "what's the best Android emulator?" or "how do I set up Genymotion?", the content rarely appeared in the answer.

The problem was not the content itself. The problem was how the content was packaged. The main site loads 69 scripts and renders 3,249 DOM nodes. AI crawlers have finite budgets. Most of that budget was spent parsing framework overhead instead of reading answers.

Baseline citation rate: 14%. Out of every 100 relevant AI queries, Genymotion appeared in about 14 answers.

What Rozz deployed

Rozz built an AI site at rozz.genymotion.com: a structured content layer designed for AI agents. Same content as the main site, different format.

The AI site renders in under 100ms with 2 scripts and 61 DOM nodes. Every page has Schema.org JSON-LD markup. An llms.txt discovery file tells AI crawlers where to find structured content. Canonical tags on every page point back to genymotion.com, so the mirror site does not compete for Google rankings.

Rozz also deployed a chatbot on genymotion.com. Every visitor question becomes a candidate for a new Q&A page. The system processes 500+ questions per week, deduplicates them, and publishes fresh Q&A content automatically.

Setup from Genymotion's side: two DNS records. Rozz handles everything else.

Month 1: Discovery (Weeks 1-4)

Week 1: Quiet start

The AI site went live. Minimal crawler attention. ClaudeBot made the first discovery, visiting 13 pages over two days. GPTBot checked in with a handful of requests per day.

Total LLM bot requests, Week 1: under 50.

Week 2: GPTBot finds the site

On a single day in Week 2, GPTBot made 547 requests to the AI site. That one day accounted for 47% of all training bot activity in the entire first month. GPTBot followed the sitemap systematically, prioritizing content pages (57%) over Q&A pages (37%).

This was the trigger event. GPTBot had decided the site was worth indexing at scale.

Weeks 3-4: Follow-up waves

GPTBot returned in waves. A secondary crawl of 124 requests. A tertiary wave of 409. Then a targeted Q&A session: 40+ Q&A pages crawled in rapid succession, roughly one per second.

OAI-SearchBot appeared separately, making 46 requests to build retrieval indexes for ChatGPT's search feature.

Month 1 totals:

42 citations in 30 days. The pipeline was working, but slowly.

Month 2: Exponential growth (Weeks 5-8)

Week 5: Citations accelerate

ChatGPT-User requests jumped from a trickle to a stream. 345 citation events in 7 days, up from 42 the entire previous month.

Q&A pages accounted for 75% of all citations. The question-answer format matched how users phrase queries to AI, and Schema.org QAPage markup made extraction trivial.

The questions getting cited were purchase-decision queries: system requirements, pricing, macOS compatibility, Play Store setup. These are the questions someone asks when they are deciding whether to use the product.

Week 6: 1,077 citations

Three times the previous week. A single day hit 252 citations, exceeding the entire first month.

PerplexityBot began visiting, up 5x from the prior week. The story was expanding beyond ChatGPT.

The automated content pipeline was feeding the growth: 500+ chatbot questions processed weekly, fresh Q&A pages published, AI crawlers indexing the new content, successful citations reinforcing the source.

Week 7: BingBot arrives

BingBot made 1,556 requests in 7 days, more than any other single bot that week. BingBot feeds Microsoft Copilot, Bing AI, and Azure OpenAI. This was the same training-to-citation pipeline we had watched with OpenAI, now starting at Microsoft scale.

ChatGPT-User citations: 1,329. Still growing.

Six platforms were now indexing the AI site: OpenAI, Microsoft, Anthropic, Meta, ByteDance, and Perplexity. None required separate optimization. One architecture, six platforms.

Week 8: Sustained momentum

Citations stabilized around 1,000+ per week. The initial exponential spike had settled into a sustained high baseline.

Month 2 weekly citation growth:

60-day cumulative results

Bot activity across platforms

Citation rate: 14% to 95%

Before the AI site, Genymotion appeared in roughly 14% of relevant AI queries on ChatGPT. Continuous citation tracking now shows a 95% citation rate.

Current citation tracking across platforms:

ChatGPT is the primary success story. Perplexity is emerging. Gemini and Claude represent future opportunity.

Use-case validation: 24 queries across 4 platforms

To stress-test beyond the tracking queries, we ran 24 real-world use-case queries across ChatGPT, Claude, Perplexity, and Gemini. These cover 12 distinct use cases: CI/CD testing, app development, manual QA, mobile security, field data collection, customer support, social media management, and more.

ChatGPT cited Genymotion with a link in 83% of use-case queries and mentioned the brand in 96%. Genymotion ranked at position #1 in 10 of those 24 queries.

Example queries where Genymotion ranked #1:

• "Secure Android virtual device environment for regulatory compliance" (61 brand mentions in response)
• "How to run multiple Android instances for social media management" (52 brand mentions)
• "Best Android emulator for mobile app penetration testing" (45 brand mentions)
• "Virtual Android device for manual app testing without physical hardware" (42 brand mentions)
• "Stream Android application to a web browser for product demos" (38 brand mentions)

Use case breakdown across all platforms:

App Development and Manual Testing hit 100% brand mention across all four platforms. Every AI system knows Genymotion exists for these use cases. The citation gap between "brand mentioned" and "cited with a link" represents the next optimization opportunity: the AI knows about you, but does not yet trust the content enough to link to it on every platform.

Where citations go: the canonical finding

This is the data point that addresses the "split authority" concern.

94.8% of all ChatGPT citations link to www.genymotion.com. 5.2% link to support.genymotion.com. 0% link to the mirror site.

ChatGPT-User crawls the AI site's Q&A and content pages to build understanding. Then it cites the main domain in its responses. The AI site is a knowledge funnel, not a competing authority.

Every URL ChatGPT has ever cited for Genymotion was also crawled by ChatGPT-User on the AI site. 100% overlap. The bot reads the structured content, then attributes to the canonical source.

What gets cited

Q&A pages dominate. The question-answer format matches how users query AI systems, and Schema.org QAPage markup makes the content trivially extractable.

The specific topics getting cited are purchase-decision queries: system requirements, pricing and plans, macOS compatibility, how to install and download, Play Store setup, free vs paid versions.

These are not support questions. They are the questions someone asks before they decide to become a customer.

Competitor displacement

When Genymotion is not cited, these companies appear instead:

AI citation is a zero-sum game within a category. When you are in the answer, your competitors are displaced. When you are not, they fill the slot.

The three-phase pipeline

The data reveals a consistent pattern across platforms.

Phase 1: Training. AI crawlers discover and mass-index the content. GPTBot's 547-request day was the trigger event for OpenAI. BingBot's 1,556-request week was the same for Microsoft.

Phase 2: Search indexing. Separate bots build retrieval indexes. OAI-SearchBot (441 requests) feeds ChatGPT's search features. PerplexityBot (98 requests) feeds Perplexity's answers.

Phase 3: Citations begin. Real users asking AI questions receive the content in responses. ChatGPT-User requests confirm live citations. The timeline from mass training crawl to first citations: approximately 3 weeks.

This pipeline ran independently for each platform. OpenAI's started in Week 2. Microsoft's started in Week 7. The same content, the same architecture, different timelines.

What made this work

Dedicated AI site, not on-page tweaks. The main genymotion.com serves humans: tracking scripts, analytics, A/B frameworks, dynamic rendering. The AI site at rozz.genymotion.com serves AI agents: clean HTML, Schema.org markup, answer-first structure. Two users, two layers.

Q&A pages from real questions. 178 Q&A pages generated from actual chatbot conversations. These match how users phrase questions to AI systems. Q&A pages receive 4x more citations than standard content pages.

Schema.org markup on every page. QAPage for Q&As, WebPage for content pages, CollectionPage for topic pages. Full JSON-LD in the head of every page. AI crawlers can extract structured answers without parsing HTML.

llms.txt discovery file. Tells AI crawlers exactly where to find structured content. 25 requests to llms.txt in the first month alone.

Weekly content refresh. Rozz crawls the main site weekly and regenerates content. New chatbot questions become new Q&A pages automatically. Fresh content signals to AI crawlers that the site is active and current.

Canonical tags preserve SEO. Every page on the AI site references the original URL on genymotion.com. The AI site does not compete for Google rankings. The data confirms this: 0% of ChatGPT citations link to the mirror site.

Summary of results

Timeline summary