Bot Blocker Index

← Index

Cloudflare Bot Management

cdn-bot-manager · Cloudflare · public (NET)

Challenged a default bot: 48.8% of 43 · newest measurement 2026-11-15

Marketing claims

No claim captured yet.

Claim archaeology — what has been marketed, and when

Monthly Wayback snapshots of the 1 page listed below, read back to 2019. A number reaches this timeline only after the sentence around it was read and judged to be Cloudflare Bot Management describing its own product. Those judgements were made by an AI model (claude-opus-5) and have not yet been checked by a person. Every one is published with the quote it was made from and the reason it was made, so you can check it against the archived page yourself. Numbers that turned out to be customer case studies, uptime SLAs or traffic-composition stats are listed further down — recorded, not published. Nothing is inferred between two dated snapshots.

Claim publishedRead, claim not thereNo readable evidence
Decision latency
20192020202120222023202420252026
One cell per calendar month, from monthly Wayback snapshots of the pages listed below. A filled block is a month whose archived snapshot carried that number. A gap is a month where we read a page that has carried this claim and the claim was not on it. Hatching is a month we cannot speak for — no snapshot exists, the archiver was served a bot wall instead of the page, or the page that carries this claim was not among the ones archived that month. Values are never carried across a gap.
Decision latency — every change, with its citation
PeriodMonthsWhat the archived page saidArchived
Jan 2019 – Feb 202238No decision latency claim in 35 months where we read a page that has carried it; 3 further months had no readable snapshot—
Mar 2022 – May 2023150.3ms
…required. Lowest Latency Cloudflare Bot Management delivers ultra low-latency bot defenses, with median latency of 0.3 milliseconds or less. Our bot management performance keeps applications in the fast lane. Rich Analytics and Logs…
first · last
Jun 2023 – Nov 202642No decision latency claim in 4 months where we read a page that has carried it; 38 further months had no readable snapshot—

No publishable claim was found on these pages for: detection accuracy, coverage, false-positive rate. Numbers we found and held back are listed below.

Found on these pages, not published as a claim

Our extractor pulled these numbers out of the same snapshots. Each was held back: a number bound to a metric is not automatically the vendor claiming a capability, and treating it as one would score Cloudflare Bot Management against something it never said.

NumberSeenWhat it actually isArchived
0.01%2019-04 – 2023-04
49 months
not a vendor claim a named customer testimonial, not Cloudflare describing its own product
…Their mitigation strategies block bots without impacting real users - our false-positive rate is now less than 0.01%." Tony Bruess Engineer Forrester named Cloudflare a “Strong Performer” in bot management Cloudflare was named a…
link
Pages excavated
PageEraMonths probedMonths read
www.cloudflare.com/products/bot-management/
2019-01 – 2026-11
current9554

“Months read” counts snapshots we actually retrieved and searched.

Sanctioned test surfaces

SurfaceTypeTierBasis
Own CF-proxied site (Turnstile free + Bot Management)self-deploy-freeBWe own the site + CF account

A — vendor invites testing · B — our own deployment · C — passive observation

Our test results

No sanctioned runs recorded yet.

Real-world challenge rate (passive, anonymized)

Across 43 sites detected running Cloudflare Bot Management, this is what a default automated client met:

ServedChallengedBlocked

Challenge rate 48.8% — 95% CI 35–63% (±14pp, n=43). A wide interval means a small sample; read it accordingly.

Newest measurement behind this figure: 2026-11-15. No site in this figure was last measured more than 21 days ago.

Site identities are not published. This measures whether a vanilla client is challenged — no bypass performed.

Detection is by CDN fingerprint, so “served” means the site sits behind Cloudflare Bot Management but a default bot was not actively challenged — CDN presence is not proof the paid bot-management product is engaged. The challenged + blocked share is where protection actively fired.

Challenged or blocked, week by week

Each point is one weekly measurement cycle: the share of sites detected running Cloudflare Bot Managementthat challenged or blocked a default automated client, taking each site’s latest run of the week. Weeks with no usable measurement are shown as gaps, and the line never crosses one — no usable measurement was recorded for those weeks, and nothing is inferred across them.

Cloudflare Bot Management: share of sites that stopped a default bot, week by week (measured from a residential IP; challenged or blocked)
Weekly share, n ≥ 10 sitesThin sample, n < 10 (dashed line)No usable measurement recorded that weekWhiskers: 95% intervalDated change to our method (a step here may reflect it)
0%25%50%75%100%No usable measurement recorded for the week of Aug 24. Nothing is drawn or inferred for this week.no dataNo usable measurement recorded for the week of Sep 21. Nothing is drawn or inferred for this week.no data2026-10-13 (UTC): Classifier and attribution change. On this date we changed how we classify responses and attribute sites to a vendor: weaker-tier evidence can now attribute a site to Akamai, we read the cf-mitigated response header, and we apply a two-factor denial rule. Sites that had previously refused our client without being attributable could enter Akamai's sample from this date, and some Cloudflare sites moved between the challenged and blocked columns. We did not measure an effect on other vendors' samples, so this marker is shown only on the vendors it touched. This change affected Cloudflare Bot Management’s sample. A step in the line at this date may reflect our method.method change2026-10-20 (UTC): Retired sites whose vendor we no longer detect. On this date we removed seven sites from the figures. Each had stopped carrying any signature we can read, or had moved to a different vendor, or had left our sweep list, and each retirement carries a date and a recorded reason. Until now such a site kept its last measurement published indefinitely, so a figure could rest on evidence weeks old. A step in a line at this date reflects that removal, not a change in how the vendor behaved. One further site qualified on age but was held back rather than retired, because removing it would have pushed a published figure below the sample size we are willing to publish. Nothing is deleted: the retired sites keep their measurement history, and a site detected again after its retirement date counts again. This change affected Cloudflare Bot Management’s sample. A step in the line at this date may reflect our method.method changeWeek of Aug 10: 63.6% stopped (7 of 11 sites), 95% interval 35–85%Week of Aug 17: 68.3% stopped (28 of 41 sites), 95% interval 53–80%Week of Aug 31: 65.1% stopped (28 of 43 sites), 95% interval 50–78%Week of Sep 7: 69.2% stopped (27 of 39 sites), 95% interval 54–81%Week of Sep 14: 69% stopped (29 of 42 sites), 95% interval 54–81%Week of Sep 28: 70.7% stopped (29 of 41 sites), 95% interval 55–82%Week of Oct 5: 66.7% stopped (28 of 42 sites), 95% interval 52–79%Week of Oct 12: 73.2% stopped (30 of 41 sites), 95% interval 58–84%Week of Oct 19: 71.4% stopped (30 of 42 sites), 95% interval 56–83%Week of Oct 26: 73.2% stopped (30 of 41 sites), 95% interval 58–84%Week of Nov 2: 72.5% stopped (29 of 40 sites), 95% interval 57–84%Week of Nov 9: 72.5% stopped (29 of 40 sites), 95% interval 57–84%Aug 10n=11Aug 31n=43Sep 14n=42Oct 5n=42Oct 19n=42Nov 2n=40
Weekly figures
Week ofSitesStopped95% intervalNote
Aug 101163.6%35–85%
Aug 174168.3%53–80%
Aug 24———no usable measurement recorded
Aug 314365.1%50–78%
Sep 73969.2%54–81%
Sep 144269%54–81%
Sep 21———no usable measurement recorded
Sep 284170.7%55–82%
Oct 54266.7%52–79%
2026-10-13 (UTC)Our method changed: Classifier and attribution changeOn this date we changed how we classify responses and attribute sites to a vendor: weaker-tier evidence can now attribute a site to Akamai, we read the cf-mitigated response header, and we apply a two-factor denial rule. Sites that had previously refused our client without being attributable could enter Akamai's sample from this date, and some Cloudflare sites moved between the challenged and blocked columns. We did not measure an effect on other vendors' samples, so this marker is shown only on the vendors it touched. This change affected Cloudflare Bot Management’s sample. A step in the line at this date may reflect our method.
Oct 124173.2%58–84%
2026-10-20 (UTC)Our method changed: Retired sites whose vendor we no longer detectOn this date we removed seven sites from the figures. Each had stopped carrying any signature we can read, or had moved to a different vendor, or had left our sweep list, and each retirement carries a date and a recorded reason. Until now such a site kept its last measurement published indefinitely, so a figure could rest on evidence weeks old. A step in a line at this date reflects that removal, not a change in how the vendor behaved. One further site qualified on age but was held back rather than retired, because removing it would have pushed a published figure below the sample size we are willing to publish. Nothing is deleted: the retired sites keep their measurement history, and a site detected again after its retirement date counts again. This change affected Cloudflare Bot Management’s sample. A step in the line at this date may reflect our method.
Oct 194271.4%56–83%
Oct 264173.2%58–84%
Nov 24072.5%57–84%
Nov 94072.5%57–84%

Residential vantage only; datacenter runs are never pooled in. Hollow points rest on a small sample — read their intervals, not their position.

These points will not always match the headline rate above, and both are right: a point counts only the sites measured in that week, while the headline keeps each site’s most recent result whenever it was taken. A site missed in one week still counts above, but not in that week’s point.

Cited external benchmarks

We cite these rather than generating real-world bypass numbers against non-consenting production systems. Commercial-vendor figures are self-reported marketing claims, flagged as such above — treat them as the ceiling a paying bypass service advertises.