Skip to main content
No. Agent Analytics reads the request logs your CDN, host or server already produces. There’s no JavaScript tag, and AI crawlers rarely run JavaScript, so a tag would miss them anyway.
Log-based connectors (Logpush, Vercel, CloudFront, Google Cloud CDN, Fastly, Akamai, Netlify) run entirely outside your request path. The Cloudflare Worker runs on your route but is fail-open: it reports after the response is sent, never reads the response body, and stops reporting during an outage instead of waiting.
No. The client IP is used in memory to verify the agent and then discarded. User agents, query strings, cookies, headers and full referrer URLs aren’t stored either. Only AI agent requests are kept; human traffic is dropped on arrival. See Privacy by design.
You don’t have to. Filtering to AI agents at the source saves bandwidth, and some connectors (Google Cloud CDN, the Cloudflare Worker) do it for you. Sending everything is also fine: human traffic doesn’t count toward your quota.
49 agents from 22 providers: OpenAI (GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot), Anthropic (ClaudeBot, Claude-SearchBot, Claude-User, Claude-Agent), Perplexity (PerplexityBot, Perplexity-User), Google (Googlebot, Googlebot-Image, Googlebot-Video, Storebot-Google, GoogleOther, Google-CloudVertexBot, Google-GeminiNotebook, Google-Agent), Microsoft (Bingbot, AdIdxBot, MicrosoftPreview), Apple (Applebot, Applebot-Extended), Amazon (Amazonbot, Amzn-SearchBot, Amzn-User), Meta (meta-externalagent, Meta-ExternalFetcher, Meta-WebIndexer), Mistral (MistralAI-User, MistralAI-Index, MistralAI-Training), DuckDuckGo (DuckAssistBot, DuckDuckBot), You.com (YouBot), Kagi (Kagibot), Exa (ExaSearchBot, ExaUser), Diffbot (Diffbot, Diffbot-User), Common Crawl (CCBot), Allen Institute for AI (AI2Bot), Yahoo (Slurp), Yandex (YandexBot, YandexAdditional), Baidu (Baiduspider), Huawei (PetalBot), ByteDance (Bytespider) and Cohere (cohere-ai). Each one rests on its operator’s own documentation. See the full table.
Unverified means there wasn’t decisive evidence: no client IP in the log, no published ranges for that provider, or DNS didn’t answer. Spoofed means the evidence contradicts the claim: the IP is outside every range the provider publishes. See How verification works.
No. A crawler request shows that an AI system read your page. Citations are measured separately by your prompt tracking, and human click-throughs by AI Referrals. The Pages tab links the three for each page, but they don’t share a denominator.
Only the last seven days. Older records are dropped as timestamp_out_of_range. Use Log file import for a short backfill, then keep a live connector running.
Yes. Connect one source per domain and per platform, for example a CDN for example.com and another for docs.example.com. Each source has its own token, endpoint and daily quota, and the dashboard can filter by source. A brand can have one source per connector type and domain.
Yes. A source created for example.com accepts example.com and every subdomain of it, such as www.example.com or docs.example.com. Other domains are dropped as host_mismatch.
Records that fit are stored, and the rest are dropped as quota_exceeded until midnight UTC. A delivery with no room gets 429 with a Retry-After header. At 80% of the quota the source shows a warning, and an alert is raised. See Quotas.
No. For security, the token is shown only once. Rotate it from Connected sources (or from the setup wizard) to get a new one, then update your platform. The old token stops working immediately.
No published plugin or app is available. If your WordPress or Shopify site sits behind a CDN such as Cloudflare, connect the CDN. Otherwise, use Custom HTTP from your own log pipeline.
Owners, integration admins, and editors or billing admins with access to the brand. Viewers can see the dashboard but can’t manage sources. See Team and roles.
No. Disconnecting refuses further deliveries. Events already accepted stay in your reports until they age out.A disconnected source can’t be reopened, but you can connect the same connector for the same domain again: it becomes a new source with a new endpoint and token, and the old one stays listed as disconnected. If you only need new credentials, rotate the token instead.
Last modified on November 18, 2026