Connect the evidence you need
#Open Agent Analytics and choose Setup. Check the project's public domain, then connect the sources you need.
| Source | Evidence it provides |
|---|---|
| Google Analytics 4 | Website sessions and identified AI referral traffic. |
| Google Search Console | Search clicks, impressions, click-through rate, and position. |
| Crawler source | AI crawler requests to the project's domain, sent from your site's request logs or code. |
| Tracked answer citations | Pages linked in the AI answers collected for your project. |
A source can be connected while still waiting for observations. Check its status and latest available data before comparing periods.
Read Overview, Pages, and Visitors
#Overview brings the available source summaries together and shows coverage states. Pages combines page-level evidence where matching data is available. Open a page record to inspect its evidence and any follow-up shown there.
Visitors uses connected traffic and search data. Use the data source control, date range, and available filters to inspect the part of the traffic you need.
An AI referral is an observed visit attributed through the traffic source. A citation is an answer link. A crawler request is a log event. Treat them as separate observations when you assess a page.
Set up crawler requests
#Crawler request setup now has its own guides under Crawler logs. Start with Getting started, or go straight to your platform:
Read Crawlers and Logs
#Crawlers shows exact totals for the selected dates: crawler requests, user-triggered fetches (an AI assistant fetched the page because someone asked it something), URLs crawled, and the success rate of requests with a recorded status. The chart stacks requests per UTC day by operator. The Crawlers table lists each crawler with its purpose, requests, share, verified share, error rate, and the last day it was seen. Crawler operators and Crawler purposes rank requests by operator and by the purpose each operator declares. Pages crawled lists every URL crawlers requested, 25 per page, with search, an operator filter, and sorting by most crawled, errors, or most recent. Select a crawler or a page to open its requests in Logs.
Logs lists individual requests, newest first, 25 per page, grouped by UTC day. Each row shows the time (UTC), the crawler with its operator's mark, its purpose, the request, its status, and, when more than one method delivered in the dates, its source. Search by host or path, or enter exact: followed by a host and path for one page, and filter by operator, purpose, verification, status, and, when more than one method delivered in the dates, source. Export CSV downloads the requests that match your filters, up to the newest 10,000. A search reads a bounded number of requests at a time; choose Keep searching to look further back.
A crawler request is labeled verified when its IP address is inside the ranges the crawler's operator publishes, and declared otherwise. OpenAI, Anthropic, Perplexity, Google, Apple, Common Crawl, Mistral (for its user and index crawlers), and DuckDuckGo publish ranges. Meta, Amazon, ByteDance, You.com, and Ai2 don't, so their requests are always declared and their verified share shows no value. Requests from methods that send no IP address, Next.js middleware and the WordPress plugin, are declared too. A label names the crawler, not the person or task behind a request. AI agents that browse with an ordinary browser user agent can't be told apart from people and aren't counted. Googlebot and Bingbot aren't counted either; Search Console and Bing Webmaster Tools report their crawling.
A day with no crawler requests counts as zero only when your source delivered to Searchly that day, setup checks included. On a day without a delivery the count is unknown, not zero, because a source that sends only AI crawler requests sends nothing on a quiet day. Searchly stores up to 50,000 crawler requests per project each UTC day; requests over that limit are counted but not stored, and totals are then shown as lower bounds. Crawler requests and their daily totals are kept for 90 days. Submitted events show observed requests; they do not establish that a model trained on a page or that a particular crawl caused a referral.
If a source shows Waiting, Setup incomplete, Partial coverage, or Unavailable, check the connection and scope first. For connection steps, read Connect integrations.