What the AI Site reveals about AI-mediated discovery

Source: https://rozz.site/insights/chatgpt-sessions-as-analytics.html

Published:

Keywords: ChatGPT-User, CloudFront logs, AI-mediated discovery, Genymotion

Document ID: rozz_site_insights_chatgpt-sessions-as-analytics.html

Entry #6 · Mar 10, 2026

What the AI Site reveals about AI-mediated discovery

ChatGPT-User made 681 visits to Genymotion’s AI site in one week. By grouping those visits into sessions using IP hashes and timing, we can reconstruct how user sessions in ChatGPT unfold.

We’ve spent several months tracking AI citations: running queries against the ChatGPT API, counting how often Genymotion appeared in responses, and watching the trend. That method helps measure whether your content exists in the model’s internal knowledge and its retrieval index. This week we found a way to look from the user’s perspective.

Session data: March 3–10

| Metric | Value | |---|---| | ChatGPT-User visits | 681 | | Reconstructed sessions | 168 | | Content sessions (non-index) | 127 | | Pages fetched | 587 | | Avg pages per content session | 4.6 | | Multi-turn sessions | 38 (30%) | | Unique questions identified | 109 |

The session reconstruction is imperfect. We use heuristics to infer behaviors. The patterns that emerge are coherent enough to be meaningful. Also, from previous weeks you might have noticed the number of ChatGPT-User visits has gone down. That’s because we refined our way of counting visits.

The signal: ChatGPT-User in CloudFront logs

ChatGPT-User is a documented OpenAI bot. ChatGPT-User fetches web pages during response generation. ChatGPT-User is distinct from GPTBot. GPTBot indexes content for training. ChatGPT-User is distinct from OAI-SearchBot. OAI-SearchBot builds the retrieval index. ChatGPT-User shows up at response time. A human uses ChatGPT. The human uses ChatGPT (not the API). ChatGPT-User queries the AI Site in real time.

Because the AI site is hosted on CloudFront, every ChatGPT-User request is stored in our logs. Each stored request includes a timestamp. Each stored request includes a URI. Each stored request includes an IP hash. We used those 3 pieces of information to reconstruct multi-turn ChatGPT sessions.

The result is a session model. Each session has one or more turns. Each turn corresponds to a fetch event. ChatGPT-User pulls a set of pages to respond to something. We analyze page contents to infer what the user was asking about at that turn.

Finding 1: ChatGPT-User fetches multiple pages per turn

A naive model of AI retrieval assumes one query maps to one fetch. In the naive model, the bot finds the best page. In the naive model, the bot reads the page. In the naive model, the bot answers. The log data does not support that model.

In the turns we can reconstruct, ChatGPT-User on average pulls 4.6 pages. The 4.6 pages are pulled in a tight burst. After the burst, ChatGPT-User moves on.

Pricing questions generated the most fragmented fetch patterns. Across the week, four near-duplicate Q&A pages covering pricing were fetched a combined 47 times:

| Page | Visits | |---|---| | what-pricing-plans-are-available-for-genymotion | 18 | | what-are-genymotion-s-pricing-options | 15 | | what-are-the-pricing-options-for-genymotion | 7 | | what-are-the-costs-for-using-genymotion-saas | 7 |

The same pattern appeared for macOS compatibility. Three pages were fetched for macOS compatibility. The three pages had 38 combined visits. The same pattern appeared for Google Play installation. Four+ pages were fetched for Google Play installation. The four+ pages had 30+ combined visits.

One interpretation: ChatGPT-User is verifying and consolidating across sources.

ChatGPT-User is not just reading the first relevant result.

If this interpretation is right, it justifies an important aspect of our AI site design.

Our AI site design includes multiple facets of a topic covered by distinct Q&A pages.

The distinct Q&A pages exist due to variations in the way users ask questions in the chatbot.

This isn’t redundant information.

This information may be exactly what the bot is looking for.

The information can provide either complementary information or information validation.

Finding 2: 28% of sessions hit the index

40 of 168 sessions fetched only /index.html and stopped. No Q&A pages were fetched. No content pages were fetched. No further navigation was visible.

The previous index was an infrastructure page. The previous index listed API endpoints. The previous index listed content counts. The previous index included navigation links.

ChatGPT-User does not use JSON APIs. ChatGPT-User GETs HTML pages. When ChatGPT-User arrived at that index, not enough signal existed to decide which page to fetch next. The session ended on our end. No further turns were visible. No additional page queries were visible. The session ended as if the bot had moved on.

We adapted our index page. Our adapted index page opens with a product description. Our adapted index page includes a topic directory with inline descriptive summaries. The inline descriptive summaries give the bot enough context to proceed. We will check over the next weeks if this improves bot performance on the AI Site.

Finding 3: 30% of sessions are multi-turn

Nearly 1/3 of sessions involved a second or third fetch cluster. The second or third fetch cluster can be attributed to the same session. We found that we could follow a meaningful conversation. The meaningful conversation happened in ChatGPT. The meaningful conversation did not happen in our own chatbot.

| Date | Turns | Pages | Duration | Fetch pattern | |---|---:|---:|---:|---| | Mar 3 | 4 | 11 | 248s | Web emulator → installation → pricing → user guide | | Mar 4 | 3 | 21 | 95s | macOS compatibility → pricing → Play Store setup | | Mar 6 | 4 | 6 | 637s | SaaS templates → template count → index → user guide | | Mar 7 | 2 | 9 | 1,377s | Linux install + KVM errors → ARM transition |

The March 4 session fetched 21 pages across 3 turns in 95 seconds. The March 7 session had a 23-minute gap between turns. We can almost see the user shifting away. We can almost see the user trying something. We can almost see the user returning with a follow-up question.

Multi-turn sessions reveal something that citation tracking cannot reveal. Citation tracking uses metrics such as Share of Citations. Multi-turn sessions reveal the sequential nature of AI-mediated discovery. The fetch patterns show natural progressions. The natural progressions can go from compatibility to pricing to setup. The natural progressions can go from troubleshooting to related technical questions. If we get these progressions right, it affects the chances that the AI will recommend us.

Get This for Your Site

ROZZ builds this infrastructure automatically. AI site. Q&A pages come from your chatbot. Schema.org markup is on every page. Session analytics are derived from weekly log analysis.

$997/month | 168 sessions reconstructed in one week

Data source: CloudFront access logs for rozz.genymotion.com, March 3–10, 2026. Session reconstruction is based on IP hash grouping and timing heuristics. Bot classification is based on User-Agent strings.

Serial tech entrepreneur with 10+ years experience building AI systems including Aristotle (conversational AI analytics) and products for eBay and Cartier. Previously founded Squid Solutions. Built AI products like Aristotle, the conversational big data analytics chatbot. Built an AR jewelry try-on device for Cartier.

March 10, 2026 | Data period: March 3–10, 2026

rozz @ rozz.site | © 2026 ROZZ. All rights reserved.