Cloudflare today published research accusing Perplexity, the AI-powered answer engine, of using stealth, undeclared crawlers to bypass website no-crawl directives. According to a blog post by Cloudflare researchers Gabriel Corral, Vaibhav Singhal, Brian Mitchell, and Reid Tatoris, Perplexity initially crawls using its declared user agents—PerplexityBot and Perplexity-User—but when blocked, rotates to a generic Chrome-on-macOS user agent and undisclosed IP ranges while ignoring robots.txt. The company says it observed this behavior across tens of thousands of domains and millions of requests per day.

Cloudflare says it confirmed the stealth crawling by creating new, non-indexed test domains with strict robots.txt disallows. When queried, Perplexity still returned detailed information about those domains. Cloudflare has de-listed Perplexity’s declared bots as verified crawlers and rolled out managed rules to block the stealth activity, available to all customers including free users.

Perplexity spokesperson Jesse Dwyer dismissed the report as a “sales pitch,” adding that screenshots in the post show no content was accessed. Dwyer also claimed the bot identified in Cloudflare’s analysis “isn’t even ours.” The dispute escalates the ongoing battle between AI companies and content publishers over scraping consent, as Cloudflare last month launched a marketplace to let sites charge AI crawlers.