Every AI crawler separately
Search index, live query or training: what each bot does, how often it visits and whether that is going up or down.
Found in AI · part of Edge SEO
See how often each AI crawler visits, which pages it reads and how many visitors AI assistants send you. The edge blocks fake bots that pretend to be ChatGPT or Perplexity. And what AI needs to read your site properly, the edge takes care of.
Search index, live questions or training: for each bot, what it does and how often. Anything that pretends to be ChatGPT or Google but doesn't come from their addresses doesn't count and is blocked.
Example of 28 days · hover over a bot to see the colors
Search index, live query or training: what each bot does, how often it visits and whether that is going up or down.
Scrapers that pose as ChatGPT, Perplexity or Google are not counted and are blocked.
The pages AI fetches most often. Make sure those are correct.
How many people reach your site via ChatGPT, Perplexity, Gemini, Copilot and Claude, and on which page.
A summary of your site for AI assistants, built from your own pages and served live via the edge.
28 AI bots, allow or block per bot, with three quick presets. In your robots.txt and with a hard block, also for bots that ignore robots.txt.
mention you, or your competitor?Every week we put your questions to ChatGPT, Claude, Gemini and Perplexity. You see whether you are mentioned, in which position and who is mentioned instead. Nobody can "train" an AI on your business. You can see where you stand and improve in a targeted way.
ChatGPT, Perplexity and Google pull answers from YouTube, Reddit, LinkedIn, forums, comparison sites and reviews. A short video, an honest answer on a forum or a LinkedIn post about a question from your field, under your real name and without spam, becomes a source that way. Under "Where do the AIs get their answers?" you see which sources count for your questions.
Google, Trustpilot, Kiyoh or industry sites. AIs mention companies more often when several independent sources speak well of them.
What you do, for whom, where, since when and what it costs. Short, factual sentences that an AI can copy word for word work best.
Organization, Product and FAQ schema make your data readable for machines. With llms.txt you tell AIs where your most important pages are. The edge serves llms.txt and completes basic schema.
"X vs Y" and "best ... for ..." are exactly the questions people ask an AI. An honest comparison on your own site is often used as a source.
If you block GPTBot, ClaudeBot or PerplexityBot, they can't recommend you either. In Search engines and AI you can see which bots really read you.
Being read is not the same as being mentioned. An AI assistant chooses which source to mention for each question, and nobody can guarantee that. What you can do: make sure AI can read you well, make sure your most important pages are correct, and measure what it delivers. That is what Netslim Edge does.
Not automatically, but it is the first step: AI can't mention what it doesn't read. The search index bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot) and live questions (ChatGPT-User and similar) count most for being found. Training bots use your text for a next version of a model, and that takes months.
That is up to you. Blocking training (such as GPTBot, ClaudeBot and CCBot) does not remove you from ChatGPT Search or Perplexity, because they use other bots. The quick preset "In AI answers, no training" sets that up with one click. If you want to be found in AI, always allow the search and live bots.
28, in two groups. Training: GPTBot, ClaudeBot, CCBot, Google-Extended, Applebot-Extended, the Meta bots, Bytespider, Amazonbot, Cohere, Diffbot and more. Search and answers: OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, DuckAssistBot, Mistral, You.com and Alexa. For each bot you see how often it visited and how many requests the edge blocked. We never block Googlebot or Bingbot.
Yes. A blocked bot gets a 403 at the edge, even if it skips robots.txt. Live fetchers such as ChatGPT-User and Perplexity-User sometimes do this, because a user asks them to. Only Google-Extended and Applebot-Extended work solely through robots.txt: they are not separate bots but agreements, and Google and Apple stick to them.
OpenAI and Perplexity publish the IP addresses of their bots. We compare every visit with those lists, which we refresh every day. We check Google, Bing and Apple via DNS. Bots whose maker doesn't publish a list are still shown, but without a "real" checkmark.
Most AI crawlers do not run JavaScript. If your site only loads its text through JavaScript, they see an almost empty page. The edge shows you which pages are affected.
Through the edge, most pages come from the cache, so your server barely notices. We stop fake bots before they reach your server.
Put your site behind the edge for free and watch what it does. Your DNS, your hosting and your visitors won't notice a thing.
The chat goes along to WhatsApp, so you don't have to repeat anything.