Cloudflare already knows which bots are knocking. As of Thursday, it is also starting to grade how well they behave once admitted. In a blog post, the company described a system that observes whether Verified bots — operators who have cryptographically proven their identity — keep honoring site preferences over time, and collapses those observations into an explainable “trust band.” The company’s own framing is apt: identity is a driver’s license; it says who you are, not how you drive.
How the identity half already works
To be a Cloudflare Verified Bot, an operator must meet three criteria: identify itself accurately and securely, do only what it says it will do, and not engage in abusive behavior. The first is handled by Web Bot Auth, in which bots sign their HTTP requests to prove who operates them; BotBase then catalogues those identities in an open format. Cloudflare poses the verification problem as two questions — “Are you who you say you are?” and “Can we trust you?” — and the new system exists because the second question has no point-in-time answer. The framing matches the company’s earlier work tracking account behavior histories to catch fraud that beats identity checks.
What the system measures
The initial signals are deliberately mundane. First: does the bot’s observed traffic stay consistent with the site’s robots.txt file over time, per RFC 9309? Isolated violations and one-offs don’t count; what the system looks for are repeated patterns that disregard published preferences. Cloudflare is explicit that robots.txt is a floor, not a ceiling — the file states what bots must not do, but nothing in it implies that everything unstated is allowed, and a site can communicate that continued access is unwelcome through other signals.
Second: has the bot changed its behavior in response to direct feedback, specifically HTTP 429 “Too Many Requests” responses? A 429 is a directive to reduce request rate, sometimes with a hint on how long to wait. A bot can stay entirely within allowed paths and still bury an origin in requests — a denial-of-service incident wearing a polite signature. The trust band rises for bots that consistently slow down or pause after a 429, and falls for those that keep sending at the same pace. Cloudflare acknowledges that misbehavior is often misconfiguration or a bug rather than malice, which is why the band grades sustained patterns rather than permanent culpability.
Trust band, not verdict
The company draws a clean line between risk and trust. Risk is the likelihood that a single request or point-in-time action is harmful, and it can change moment to moment. Trust accumulates through repeated, consistent behavior. The output is a range showing how consistently a bot respects preferences, annotated with the observed behaviors that produced it — “rather than a permanent judgment about a bot or its operator,” in the post’s words. The thesis: verify, then continuously re-verify.
Cloudflare says the scoring runs across internal evaluation of the Verified bot population now, and this initial phase is for building behavioral history and refining what the observations mean. It plans to publish the scores on Radar, expand the measured signals, and eventually open the system to all customers, who will be able to choose their own trust threshold for handling Verified bot traffic. Together with Bot Preference Sync, which lets sites publish preferences, and Bot Management, which enforces them, the pieces form what the company calls an accountability loop: site declares, bot is observed, customer decides.
Two qualifications before granting the model its due. First, no data yet: the post announces a scoring system with no published scores, no population statistics, and no false-positive methodology, so today there is nothing to audit. Second, the one concrete enforcement action on record is instructive. Cloudflare de-listed Perplexity’s bot from the Verified list in 2025 after it attempted to access resources on websites that had opted out of scraping and then tried to circumvent the controls blocking it. That precedent — deliberate evasion, judged at the network layer — is exactly the behavior a trust band is meant to surface earlier. It is also Cloudflare acting as identity issuer, behavior scorer, and traffic gatekeeper simultaneously, a vertical stack worth remembering when the scores go public.