Skip to main content
Plans: All plans, limits vary · In the app: Sidebar → Analytics → Agent Analytics
Agent Analytics shows how AI systems reach your website. You connect the request logs your CDN, host or server already produces, and Quotable picks out the requests made by AI agents: training crawlers like GPTBot, search indexers like OAI-SearchBot and PerplexityBot, and assistants fetching a page for a live user, like ChatGPT-User or Claude-User. Collection is server-side. There is no JavaScript tag to install, and AI agents rarely run JavaScript anyway, so a client-side tracker would miss them.

What you can answer with it

  • Who is visiting. Requests per AI provider and per agent, with the share that is verified.
  • Why they visit. Each agent is labelled with its purpose: training, search indexing, or fetching for a user.
  • What they read. The pages AI agents request most, with request counts, distinct agents, errors and p95 response time per page.
  • Where they fail. URLs that return 4xx or 5xx to AI agents, and slow responses.
  • Whether you allow them. An on-demand robots.txt audit that shows which AI agents your current rules allow at the site root.
  • What is happening right now. A live feed of the latest accepted agent requests.
Agent Analytics measures crawler and agent requests. Human visitors who click through from an AI answer are a different dataset: see AI Referrals. A crawler request does not imply a citation, and a citation does not imply a crawl.

How it works

1

You connect a log source

Pick the platform that serves your site (Cloudflare, Vercel, CloudFront, Fastly, Akamai, Netlify, Google Cloud CDN, or a custom HTTP sender) and point its log delivery at the endpoint URL Quotable gives you. Each source is bound to one domain and has its own token.
2

Quotable keeps only AI agent requests

Every delivered record is checked against the agent registry. Human traffic is dropped on arrival and never stored or counted against your quota.
3

Each agent request is verified

When the record carries the client IP, Quotable checks it against the provider’s published IP ranges or confirms it with reverse DNS. The IP is used for that check only and is then discarded.
4

You read the results

The Agent Analytics page of each brand shows KPIs, trends, agents, pages, crawlability checks and live traffic.

Recognized AI agents

Detection matches the user agent the request declares. The current registry recognizes these agents: Every agent in the table rests on its operator’s own documentation. A crawler that only appears in third-party bot lists is not added. Some entries cover more than one user agent: GoogleOther includes GoogleOther-Image and GoogleOther-Video, Google-GeminiNotebook includes its earlier name Google-NotebookLM, YandexAdditional includes YandexAdditionalBot, Baiduspider includes its image, video and news variants, and ExaSearchBot and ExaUser include their image fetchers. Anthropic and DuckDuckGo don’t publish a machine-readable range list for the agents marked “IP ranges maintained by Quotable”, so those ranges are maintained by Quotable. When no range is loaded for an agent, its requests stay unverified. Google-Extended is a robots.txt control token, not a user agent Google sends. A request that claims to be Google-Extended is still recorded, but it can never be verified. Search crawlers such as Googlebot, Bingbot, YandexBot, Baiduspider and Slurp are listed because the indexes they build also feed AI answers. Requests from agents that are not in the registry are not stored. Googlebot-Image, Googlebot-Video and AdIdxBot have their own rows since 11 November 2026. Before that their requests were counted as Googlebot and Bingbot, and events stored before it keep that label.
“None available” means the provider publishes no IP ranges or hostnames for that agent. “Not supported yet” means the provider publishes a way to check the agent that Quotable doesn’t read yet, such as signed requests or a list in another format. In both cases the requests show as unverified. That is a statement about the evidence, not an accusation. See How verification works.

Intents

Intent comes from the purpose each provider documents for its agent, plus one behavioural signal: the HTTP method. Blocking a training crawler does not stop search indexers or user fetchers from the same provider. Each agent obeys its own robots.txt rules.

Privacy by design

Agent Analytics stores a reduced, privacy-safe record of each agent request. These never cross into storage:
  • Client IP addresses. Used in memory for verification, then discarded. The record keeps only the verification result.
  • Full user agent strings. Reduced to the recognized agent and provider.
  • Query strings and URL fragments. Stripped from every path.
  • Cookies, request headers and bodies. Not read from the log record.
  • Full referrer URLs. Reduced to the referring host.
Paths are also scrubbed before storage: segments that look like IDs, tokens or email addresses are replaced with :id, :token or :redacted, and paths are cut to 2,048 characters. Requests to these paths are dropped and not stored: /admin, /auth, /private, /_next, /assets, /static, and static files ending in .css, .js, .map, .png, .jpg, .jpeg, .gif, .webp, .svg, .ico, .woff, .woff2, .ttf or .eot.

Plans and quotas

Agent Analytics is available on every plan. Your plan sets a monthly allowance of accepted agent events for the whole workspace, a daily limit for each source, and the largest log file one import can upload. See brand plans and agency plans for the numbers. Human traffic, excluded paths, invalid records and duplicates don’t count toward either limit. Enterprise limits can be raised by contract. See Quotas for what happens when a source reaches its limit.

Next steps

Get started

Pick a connector and confirm your first delivery.

Read the dashboard

What each KPI, tab and alert means.

How verification works

Verified, unverified and spoofed, explained.
Last modified on November 18, 2026