Cloudflare Bot Management is the Enterprise control that scores each proxied request from 1 (automated) to 99 (likely human), so a WAF rule can log, challenge, or block it; Bot Fight Mode and Super Bot Fight Mode are lighter products and don't expose that score. If Security Events shows challenges or 403s but you can't find a bot score field, first check which bot control the zone has.
That distinction changes what you can tune. A browser challenge may work on a public page and fail an API client, monitor, mobile app, or webhook. Start with the product and the affected path, then use the event details to make a narrow change rather than turning off protection for the zone.
Table of Contents
- Which Cloudflare Bot Control You Actually Have
- How Cloudflare Scores a Request
- Score Bands Actions and Managed Rules
- How to Tune False Positives Safely
- Origin Integration for Apps and APIs
- Production Failures and How to Read Them
- When to Keep Tuning and When to Escalate
Which Cloudflare Bot Control You Actually Have
Cloudflare Bot Management for Enterprise provides a per-request score. You act on it through a WAF custom rule or a Worker. Bot Fight Mode and Super Bot Fight Mode use different signals, so they don't expose that same score for rules.
| Product | Typical plan | Signal you can use | First action |
|---|---|---|---|
| Bot Fight Mode | Free | Known bad bots only | Leave it on for simple sites; there's no score tuning |
| Super Bot Fight Mode | Pro and Business | Definitely automated, likely automated, verified bots | Block definitely automated traffic; allow verified bots you need |
| Bot Management | Enterprise | Score 1 to 99, JA3 or JA4, verified bot | Use a custom rule for low scores; log the uncertain band |
| Bot Management for AI and crawlers | Enterprise add-on path | AI crawler and training-bot categories where enabled | Allow or block named crawlers; don't treat them as the 1 to 99 score |
For a zone without Enterprise Bot Management, tune the categories it exposes. Allow verified bots that need to crawl public pages, and don't paste a score expression into a rule builder that doesn't offer the score field. With Enterprise Bot Management, start by logging the middle band and acting only on the lowest scores. A healthy first rollout has a small, path-scoped set of blocks or challenges, not a zone-wide block for every score below 50.
Practical rule: A score is a signal to use in a rule, not an action by itself.
The origin still sees Cloudflare edge addresses unless you restore the visitor IP there. Configure that separately from bot decisions, in the origin integration.
How Cloudflare Scores a Request
A customer opens a proxied hostname, then the request reaches Cloudflare's edge. Cloudflare evaluates signals such as TLS fingerprints, HTTP header order, known automation libraries, request behavior, and browser JavaScript detections where applicable. The resulting score is exposed as cf.bot_management.score in WAF custom rules and as request.cf.botManagement.score in Workers. Cloudflare documents those fields and verified-bot behavior in its Bot Management documentation.

A score of 1 indicates traffic that is almost certainly automated; 99 indicates traffic that is very likely human. The score isn't a probability you can quote as a percentage. It also has limits with API clients that never run JavaScript, because browser-specific evidence can't tell you much about a native app or server-to-server client.
Cloudflare separately identifies verified bots. cf.bot_management.verified_bot is true when traffic matches Cloudflare's verified crawler list, such as a major search engine. A verified crawler can be automated and still be a bot you want to allow. Think of the score as a door steward weighing gait and paperwork, while verified_bot is a guest list.
If the dashboard shows Adaptive Intelligence, treat it as an Enterprise layer behind bot scoring that can change detection using attack patterns. Don't assume a fixed threshold or control name applies across configurations. Confirm current names and behavior in Cloudflare's Adaptive Intelligence release notes.
Score Bands Actions and Managed Rules
Use these bands as an operator starting policy, not as a Cloudflare mandate. HTML pages can run browser challenges; API clients usually can't. Keep the actions separate.
| Score or flag | Meaning | HTML action | API action |
|---|---|---|---|
| 1 to 29 | Almost certainly automated | Block or Managed Challenge | Block unless allowlisted |
| 30 to 49 | Likely automated | Managed Challenge | Block or allow only signed clients |
| 50 to 79 | Uncertain | Log only | Log only |
| 80 to 99 | Likely human | Allow | Allow |
verified_bot true | Known good crawler | Allow if indexing matters | Allow only on public read paths |
static_resource true | CSS, images, and similar | Skip challenges | Usually not applicable |
A low-score rule can exclude verified bots and static resources:
(cf.bot_management.score lt 30 and not cf.bot_management.verified_bot and not cf.bot_management.static_resource)
Run rules in a deliberate order. First skip rules for known-good paths, then the low-score action, then a log rule for the uncertain band. Add path conditions to the expression before enforcing it on a mixed site and API zone. The action names available in Cloudflare rules include Allow, Log, Managed Challenge, JS Challenge, Interactive Challenge, and Block. Interactive Challenge is the harshest user-facing option and is a poor fit for APIs, payment webhooks, and mobile apps.
Cloudflare's Bot Management managed ruleset is the prebuilt control alongside custom rules. Enable it for known bad bots, then let the custom low-score expression define your score-band policy. Avoid challenging the same HTML request twice through overlapping rules. On Pro and Business, Super Bot Fight Mode categories replace the 1 to 99 field, so there isn't a score expression to paste.
In multi-tenant hosting, a global score block can look fine on a marketing site and break the API minutes later. The same zone may serve browsers, monitors, mobile clients, and token-authenticated services. Keep path scoping in the rule. For the network-side protections around an origin, see how Cloudflare DDoS protection works.
How to Tune False Positives Safely
Before changing enforcement, confirm that Enterprise Bot Management or Super Bot Fight Mode is enabled, the affected DNS record is proxied, and you can read Security Events. Make the first production change reversible and log-first.

Log before enforcing. Set the low-score rule to Log on a production zone. On a staging hostname, a Managed Challenge can be a useful test if the affected flow is browser-based. Don't begin with a zone-wide block.
Find the affected traffic. In Security Events, filter for the rule ID and group events by path, user agent, and source. Separate browser page requests from API calls before adding an exception. A monitor's user agent and a mobile app's requests aren't the same class of traffic.
Make the exception narrow. For a confirmed browser false positive, such as an old WebView or a corporate TLS inspection proxy, scope a skip to the hostname and path. Don't allow an entire zone, and don't skip based only on a user agent. Attackers can copy user-agent strings.
Use application authentication for non-browser clients. For a mobile app or integration, stop applying browser scoring to that path and require an existing control such as an API token, HMAC signature, or mutual TLS. If search traffic falls after enforcement, explicitly allow verified bots on the public paths that should be indexed.
Promote only after reviewing events. Change the low band from Log to Block or Managed Challenge only when the sampled events are mostly unwanted automation. Keep API and webhook paths out of browser-challenge actions.
For custom rule syntax and field behavior, use Cloudflare's WAF custom-rules documentation.
A useful verification pair is a normal browser checkout returning 200 and a bare command-line client being challenged or blocked. Confirm that Security Events shows the rule ID on the test request. For example, test an endpoint you control:
curl -i https://example.com/checkout
curl -i -A 'python-requests/2.x' https://example.com/login
The first request should follow the normal browser flow. The second lacks browser cookies and JavaScript behavior, so it should match your test policy if the path is in scope. Don't treat a challenge result from an API endpoint as successful protection if the legitimate client can't complete it.
A WAF rule change takes effect at the edge quickly, but cached pages can confuse verification. Roll back by disabling the custom rule or changing its action to Log, then send a fresh request and confirm it has a new cf-ray. The video below demonstrates the monitoring context for reviewing these events.
Origin Integration for Apps and APIs
The request path is client, Cloudflare edge, bot scoring and WAF, then the origin on a VPS, bare metal host, or virtual machine. Unless the origin restores the visitor address, its access log records Cloudflare's edge address. That can make source-based investigation misleading and can also leave a direct-origin bypass if the host accepts public traffic outside Cloudflare.
For NGINX on Linux, configure the IP restoration module with Cloudflare's published ranges. Keep the list maintained from Cloudflare's official IP address list, not from memory.
# Include one set_real_ip_from directive for each current Cloudflare range.
set_real_ip_from 173.245.48.0/20;
set_real_ip_from 2400:cb00::/32;
real_ip_header CF-Connecting-IP;
real_ip_recursive on;
The sample ranges above are illustrative, not a complete list. Replace them with the current full list before deployment. Check the public response and compare it with the origin access log:
curl -sI https://example.com
The response should include a cf-ray header and server: cloudflare. After real-IP restoration, the origin log should record the visitor address rather than a Cloudflare edge address. Restrict origin ingress to Cloudflare ranges and your administrative jump hosts, so a bot can't bypass edge rules by connecting directly to the origin.
The same arrangement applies whether the origin runs on a VPS, dedicated server, or Proxmox VM. ARPHost's VPS hosting and bare metal servers are examples of where an origin workload can run. A host WAF or Imunify360 can add another layer behind Cloudflare, but it doesn't replace the edge bot score. See VPS web hosting security bundles for that host-side layer. For proxy headers and upstream setup, use this reverse proxy configuration guide.
Native apps won't execute Cloudflare's browser detections. Pair bot controls with API Shield, mutual TLS, or signed webhooks. For example, a payment provider's POST /hooks can be excluded from browser challenges, while the origin rejects the request if its signature header is missing. A path skip is safe only when the application checks the identity of the sender.
Production Failures and How to Read Them
After enforcement, browsers may return 200 while synthetic monitors go red and checkout tickets arrive. First separate an edge action from an origin response. A Cloudflare block or challenge includes a cf-ray and a Cloudflare interstitial; an origin 403 has no Cloudflare challenge page, and its response body comes from the application or web server.
In Security Events, filter by the custom rule ID and group by path, user agent, and score. If a library user agent gets a denial on POST /api/v1/orders at the same time a browser gets 200 on /checkout, the rule probably isn't scoped to the right path. Compare the edge event with the origin log before editing rules:
203.0.113.40 - - "POST /api/v1/orders HTTP/1.1" 403
198.51.100.24 - - "GET /checkout HTTP/1.1" 200
| What you see | Likely cause | Fix |
|---|---|---|
| Uptime probe always challenged | Monitor has no browser JavaScript | Allowlist the probe IPs or exempt the health path |
| Search traffic drops | Verified bot blocked or challenged | Allow cf.bot_management.verified_bot on public pages |
| Webhook retries fail | Provider posts server to server | Skip the webhook path and verify its signature at the origin |
| Only some offices loop on challenges | TLS inspection proxy breaks the challenge | Add a narrow skip for that app path, or raise the band after confirming users |
| One IP looks like a bot storm | Shared NAT mixes users and scripts | Block the abusive path pattern, not the whole known NAT address |
Review health, admin, and webhook exceptions periodically. Alert when the challenged share jumps after changing a rule from Log to Block or Managed Challenge. Escalate a widespread verified-bot miss, a score distribution that collapses after a Cloudflare model change, or API clients the application team can't identify.
When to Keep Tuning and When to Escalate
Keep tuning in-house when one team owns the zone, someone can review Security Events daily, and every API client has an accountable owner and authentication method. Escalate when several applications share the zone, false positives return despite narrow path exceptions, or nobody owns the score rules after the initial change. Don't leave a production rule in enforcement without an operator who can assess its impact.
A useful operational handoff covers uptime and service alerts, patch and configuration management, vulnerability scanning, and the firewall and DDoS layers protecting the same origin. Those tasks support the host and network; they don't replace Cloudflare's score. Keep DDoS attack detection in the incident plan alongside bot events so teams can distinguish request-level abuse from network-level pressure.
ARPHost managed services include 24/7 monitoring of uptime, resources, and services with automatic alerting, plus server setup, patch and configuration management, security updates, network security, and intrusion detection. We manage Linux and Windows servers; an existing customer can contact support@arphost.com to discuss an operational review.
ARPHost, LLC provides VPS, bare metal, private cloud, and managed infrastructure for origins that need reliable operations behind Cloudflare. If you need help monitoring the host, alerts, and security layers that bot rules depend on, visit ARPHost, LLC.

Leave a Reply
You must be logged in to post a comment.