boutique-exemple.fr
Cross-section report · AI agents
Produced on UTC by LogGrain, version 0.1, agent list 2026-11-01.3, from 16,000 log lines from 2026-09-25 to 2026-10-29. Sample report, produced from fictitious logs: it describes no real shop.
Sample · fictitious data
1. What this report does not say
This report counts requests seen in logs. It says nothing about the following.
- Why an agent came, and what it answered its user.
- The revenue from agents that read without sending a visitor.
- Agents that drive a real browser without signing their requests: they look like an ordinary visitor.
- What Gemini reads through Googlebot, and what Copilot reads through Bingbot: inseparable from search, not counted.
- AI-referred visitors who arrive with neither a referrer nor a utm_source.
This report does not claim a cause-and-effect relationship between any two figures. It promises no visibility in assistants’ answers.
2. What LogGrain read
16,000 lines, of which 0 unreadable; formats: cloudflare 16,000. 2,897 requests attributed to AI agents (18.1% of readable lines), 1,241 probable impersonations, 640 requests from AI-referred humans.
3. Agents, by assistant
| Assistant | Requests | Pages | Volume per week | Roles | Verification |
|---|---|---|---|---|---|
| ChatGPT | 1,668 | 1,668 | 17.3 MB | training 747, indexing 496, live user question 425 | verified address 1,668 |
| Claude | 831 | 831 | 6.4 MB | training 554, live user question 277 | verified address 831 |
| Perplexity | 398 | 398 | 4.2 MB | indexing 240, live user question 158 | verified address 398 |
Every agent is checked against the addresses published by its operator. The “live user question” role is the closest to a purchase: a human is asking a question right now.
4. Agents refused by the site
| Agent | Assistant | Refusals | Statuses | Pages affected |
|---|---|---|---|---|
| Claude-User | Claude | 227 of 277 | 403 227 | 50 |
A refusal is a 401, 403 or 429 status, or a CDN security action that blocks or challenges. A 404 is not one.
5. Product pages most read by agents
| Page | Reads | By assistant |
|---|---|---|
/livraison | 78 | ChatGPT 41, Claude 25, Perplexity 12 |
/categorie/sacs | 76 | ChatGPT 48, Claude 17, Perplexity 11 |
/retours | 75 | ChatGPT 50, Claude 20, Perplexity 5 |
/magasins | 71 | ChatGPT 41, Claude 18, Perplexity 12 |
/ | 69 | ChatGPT 46, Claude 14, Perplexity 9 |
/produit/article-008 | 69 | ChatGPT 40, Claude 15, Perplexity 14 |
/produit/article-011 | 64 | ChatGPT 35, Claude 24, Perplexity 5 |
/produit/article-026 | 64 | ChatGPT 31, Claude 25, Perplexity 8 |
6. Probable impersonation
| User-agent borrowed | Requests | Addresses | Volume |
|---|---|---|---|
| fake GPTBot | 1,241 | 1,241 | 64.8 MB |
Requests that carry an agent’s name but come from addresses outside the ranges published by its operator. They are attributed to no assistant.
7. Humans referred by an AI
| Assistant | Page views | Estimated visitor-days | Detected by |
|---|---|---|---|
| ChatGPT | 430 | 430 | utm_source 327, referrer 103 |
| Gemini | 104 | 104 | referrer 104 |
| Perplexity | 106 | 106 | referrer 106 |
8. The two signals
- Catalogue: 44 of 60 product pages read by at least one agent in 30 days (73.3%); threshold of interest 20%.
- Growth: cannot be measured: 4 full week(s), 26 needed (6 months).