AI agents and crawlers such as GPTBot, ClaudeBot or PerplexityBot visit websites like any other client. The page “AI Agent Control” of a key shows which of them visit your site and lets you decide what they are allowed to do. This post shows that page. The film below shows it on the real dashboard with sample data, silent and with short captions. The text under it says the same in words, so you can follow along without playing the film.
The film in short
The film opens “AI Agent Control”, which still says “Not connected yet”. It scrolls to “Preset”, chooses “Balanced” and confirms the question about the preset, shows the “Categories”, the card “What do the actions mean?” and the card “Stage” with “Observing (nothing is blocked)”. It does not connect a real site and shows no visits. It has no sound and does not start by itself.
Step 1: set up the observer
Under “Set up observation” the page offers code for “WordPress”, “Next.js”, “Node.js”, “PHP” and “Go”. It runs on your server, sets no cookies, blocks nothing and only records detected bots, never your human visitors. Under “Test the connection” a curl command sends an empty test report with your key; “Check again” asks whether the page now hears from your site. Until then the page says “Not connected yet”.
Step 2: preset and categories
The “Preset” card offers “Maximum visibility”, “Balanced”, “Maximum protection” and “Custom rules”. A preset asks before it changes anything: the question lists what it sets and notes that you are in the observing stage. Below, “Categories” shows one row per kind of agent (“AI search”, “AI agents”, “AI training”, “Faked identity”) with “Allow”, “Observe” or “Block”, and a recommendation based on the purpose of the bot and its verification record. You decide what fits your site.
Step 3: actions and stage
“What do the actions mean?” explains the three choices: “Allow” welcomes the agents of a category, “Observe” records every visit and blocks nothing (the safe default), and “Block” creates a deny rule for the category. The “Stage” card is the important one: your rules are saved but not enforced yet, so “Block” only takes effect once you activate your rules. In the film the stage is “Observing (nothing is blocked)” and enforcement is not enabled for the account, which the page says in plain words. Further down, “robots.txt for this policy” generates a file from your rules; robots.txt is advisory and well-behaved crawlers follow it.
What the film does not show
The film does not show real visits, a blocked agent or the activation of the rules; whether enforcement is available depends on your account. The documentation page for AI agent rules describes the rules and categories in detail. What happens to suspicious visitors of your forms is a different setting, shown in set up the security policy.