dns.pizza research
The State of CAA 2026: only 12.9% of the top 100,000 domains publish a CAA record
A CAA record is the only way a domain can tell the world's certificate authorities which of them it trusts. We read them on 100,933 domains of the Tranco top 100,000. Nearly nine in ten publish nothing at all — and most of the records that do exist name a crowd rather than a shortlist.
Published November 1, 2026 · data collected June 2026–November 2026 · download the data
12.9%
publish a CAA record at all
80.5%
of allowlists name two or more CAs
80.2%
of allowlists include Let's Encrypt
98.8%
of 6,778 comparable domains served a certificate their own CAA allows
Key findings
- Nearly nine in ten top domains authorize every CA on earth. 87,863 of 100,933 domains (87.1%) publish no CAA record. With no record, every publicly trusted CA — well over a hundred of them — is free to issue for the name, and the domain owner has no say. CAA has been a standard since 2013 and mandatory for CAs to check since September 2017.
- The bigger the site, the likelier the record. Adoption runs 33.5% in the top 1,000, 18.9% from 1,001–10,000 and 12.1% in the long tail — a clean rank gradient, and the reverse of what DMARC adoption does.
- An allowlist is usually a crowd, not a shortlist. Of the 12,579 domains that name at least one CA, 80.5% (10,126) name two or more; only 19.5% (2,453) commit to a single one. The sharpest case is SSL.com: it appears on 4,732 allowlists and is the sole authorized CA on 4 of them. A CA that is almost never chosen alone looks like it arrives in a bundle rather than by decision — a hypothesis this data suggests but does not prove. Either way, each extra name is another CA whose compromise becomes your problem.
- Let's Encrypt is on 80.2% of allowlists. It leads DigiCert (66.3%), and between them they cover 89.6% of allowlists — a distinct-domain union, not the sum of the two bars, which would be a meaningless 146.6%.
- 244 domains publish CAA with no issuance rule in it. 1.9% of all CAA publishers set only advisory properties — an
iodefreporting address or the RFC 9558contactemail— with noissueorissuewildproperty at all, so a CA looking for a rule about who may issue finds none. One caveat we can measure but not resolve: 1 of those sets the RFC 8659 §4.1 Issuer Critical bit on a property this report does not evaluate. A CA that does not support that property must then refuse issuance entirely — so what that record does depends on which CA reads it. It stays in this no-issue/no-issuewildcount, which describes the tags present, with no claim made about its effective issuance policy. - Hardly anyone uses CAA's sharpest edges. Just 269 domains explicitly forbid wildcard certificates (
issuewild ";"), and 11 publish a deny-allissue ";"with no wildcard CA to override it — the correct, free setting for a domain that serves no HTTPS. (15 publish the deny-all property; 4 of those also name a wildcard CA, which RFC 8659 §4.3 lets override it.) 26.1% of publishers (3,407) give CAs aniodefaddress to report a refused request to. - Where CAA exists, it matches reality. Across the 6,778 domains where we hold both an allowlist and a trusted certificate we can attribute, 98.8% were observed serving a certificate from a CA their own CAA authorizes. Only 82 disagree — and a stale record after a CA migration explains a mismatch at least as well as anything untoward. This is not a mis-issuance finding; it is evidence that the records people do publish are the records they actually run on.
- Adoption is a registry story too. Among the 20 largest TLD samples we publish, .nl leads at 27.2% against 2.8% in .jp — a 9.6× spread across namespaces running the same protocol.
CAA posture by popularity tier
Share of domains in each Tranco rank tier, by which properties their CAA record carries. The three segments partition the corpus exactly.
Source: dns.pizza DNS record checks · snapshot November 1, 2026
Which CAs the allowlists name
Share of the 12,579 domains that name at least one CA. Rows overlap — a domain naming four CAs appears in four of them — so they are not a total.
Source: dns.pizza DNS record checks · snapshot November 1, 2026
Adoption by tier and TLD
| Segment | Domains | Has CAA | Restricts | One CA only | Has iodef | No wildcards |
|---|---|---|---|---|---|---|
| All domains | 100,933 | 12.9% | 12.7% | 19.5% | 26.1% | 269 |
| Top 1,000 | 1,002 | 33.5% | 31.4% | 32.2% | 39.9% | 3 |
| 1,001 – 10,000 | 9,327 | 18.9% | 18.3% | 25.7% | 29.3% | 27 |
| 10,001 – 100,000 | 90,604 | 12.1% | 11.9% | 18.2% | 25.1% | 239 |
| .com | 46,890 | 12.8% | 12.5% | 12% | 25.6% | 84 |
| .net | 6,516 | 11.7% | 10.9% | 22.8% | 25.3% | 21 |
| .org | 4,178 | 13.8% | 13.6% | 24% | 27.3% | 18 |
| .ru | 3,855 | 8% | 8% | 38% | 23% | 7 |
| .io | 1,932 | 18.3% | 17.8% | 9.2% | 30.8% | 4 |
| .jp | 1,810 | 2.8% | 2.7% | 19.1% | 31.4% | 1 |
| .de | 1,788 | 16.8% | 16.7% | 17% | 41.5% | 17 |
| .br | 1,409 | 13.6% | 13.4% | 10.9% | 17.8% | 2 |
| .uk | 1,348 | 12.2% | 12.2% | 12.3% | 40% | 11 |
| .in | 1,264 | 23.2% | 23% | 53.7% | 3.4% | 0 |
| .fr | 992 | 22.7% | 22.6% | 22% | 30.2% | 4 |
| .cn | 972 | 4.6% | 3.6% | 54.3% | 31.1% | 0 |
| .it | 865 | 13.8% | 13.6% | 35.6% | 18.5% | 0 |
| .co | 852 | 12.2% | 12.2% | 23.5% | 20.2% | 0 |
| .edu | 829 | 19.1% | 18.7% | 5.8% | 47.5% | 9 |
| .pl | 767 | 20.9% | 20.9% | 12.6% | 19.4% | 4 |
| .nl | 648 | 27.2% | 27.2% | 14.2% | 52.8% | 25 |
| .es | 555 | 16.8% | 16.8% | 19.4% | 38.7% | 0 |
| .au | 534 | 18.5% | 18.5% | 11.5% | 41.4% | 2 |
| .tv | 529 | 8.7% | 8.7% | 19.6% | 19.6% | 0 |
“Has CAA” and “Restricts” are shares of the segment's domains; “One CA only” is a share of those naming at least one CA; “Has iodef” is a share of CAA publishers. “No wildcards” is a count of explicit issuewild ";" denials. TLDs need at least 200 measured domains to appear, and the 20 largest qualifying samples are shown.
Authorized certificate authorities
Each row counts distinct domains whose issue allowlist names that CA, out of the 12,579 domains that name at least one. A domain authorizing four CAs appears in four rows, so these numbers overlap and must not be added up.
| Certificate authority | Domains | Share | Sole CA | Named in issuewild |
|---|---|---|---|---|
| Let's Encrypt | 10,092 | 80.2% | 1,079 | 57.2% |
| DigiCert | 8,346 | 66.3% | 329 | 67.5% |
| Sectigo | 7,126 | 56.6% | 130 | 76.6% |
| Google Trust Services | 6,888 | 54.8% | 437 | 71.5% |
| SSL.com | 4,732 | 37.6% | 4 | 96.1% |
| Amazon | 4,370 | 34.7% | 98 | 41.1% |
| GlobalSign | 1,986 | 15.8% | 192 | 42.5% |
| GoDaddy | 875 | 7% | 24 | 35.5% |
| HARICA | 414 | 3.3% | 48 | 59.7% |
| Entrust | 311 | 2.5% | 3 | 21.9% |
| Certainly (Fastly) | 250 | 2% | 1 | 24% |
| Other / unrecognised | 704 | 5.6% | 53 | 31.5% |
CAs need at least 200 domains to get a row; 10 smaller recognized CAs covering 629 domains are folded out. “Other / unrecognised” is a distinct-domain count of domains naming at least one identifier we do not map to a known CA — it is not a CA and not a sum. “Named in issuewild” counts domains that name the CA in an explicit issuewild property; under RFC 8659 §4.3 a domain with no issuewild at all still allows that CA to issue wildcards, so this is a count of explicit tags and not of effective wildcard permission.
Methodology
dns.pizza continuously resolves DNS records — including CAA — while generating domain intelligence reports across the Tranco top 100,000. This report aggregates the latest conclusive CAA lookup per domain: 100,933 domains measured between June 10, 2026 and November 1, 2026.
One vote per domain: the latest DNS report for each domain, excluded afterwards if its CAA lookup errored. The corpus is every domain our rank table places at 100,000 or better. That is 101,638 distinct domains rather than exactly 100,000, because the ranking has ties — 96,825 distinct rank values are shared among them — which is also why the “Top 1,000” tier holds 1,002 domains. Domains themselves are unique, so no domain is counted twice. Of 101,637 ranked domains with a DNS observation, 704 are excluded from every denominator because the CAA lookup itself errored (ESERVFAIL, EBADRESP) — an unreachable record is unknown, not absent. An empty answer is a real absence and counts as no-CAA; 734 domains returned an explicitly empty CAA set.
Three denominators do different jobs. Adoption shares are of all 100,933 conclusive domains. “Of publishers” shares are of the 13,070 with any CAA record. Per-CA shares are of the 12,579 that name at least one CA — which is 15 fewer than the 12,594 publishing an issue property, because a deny-all issue ";" record names nobody and belongs in no CA's row.
Property tags are matched case-insensitively per RFC 8659 §4.1, and parameters after a semicolon (issue "pki.goog; cansignhttpexchanges=yes") are stripped before the CA is identified. A value of ";" is RFC 8659's “authorize nobody” form and is counted as a denial, not as a CA.
CAs are counted as brands, not as identifiers. Amazon documents four CAA identifiers (amazon.com, amazonaws.com, amazontrust.com, awstrust.com), Sectigo five and DigiCert seven. Counting raw identifiers would split one CA across several rows and understate it, so identifiers are mapped to the same brand labels the certificate authority market share report uses, and every published figure is a distinct-domain count per brand. Identifiers we do not recognize keep their own identity when counting how many CAs a domain names, rather than folding together: collapsing them would make a domain naming two unknown CAs look as though it had committed to one.
Wildcard and denial counts are explicit-tag counts. RFC 8659 §4.3 makes issue govern wildcard issuance whenever no issuewild is present, so a domain allowing one CA and setting no issuewild does permit that CA to issue wildcards without appearing in any wildcard column here. Read those figures as counts of what domains explicitly wrote down, which makes them a floor on effective policy. The same rule is why 11 rather than 15 domains are reported as total denials: 4 deny-all records also name a wildcard CA that overrides them.
Per-CA counts overlap and are never summed. A domain may authorize several CAs, so the rows are not a partition. Where the report quotes a combined share it is a distinct-domain union computed in SQL: the two most-named CAs together reach 89.6% of allowlists, where adding their two bars would claim 146.6%. Every count in this report comes out of a single SQL statement, so no two figures are read from different database snapshots.
The CAA-versus-certificate cross-reference is the narrowest cut here. It covers only domains that name at least one CA we recognize and serve a trusted certificate whose issuer we can attribute to a known brand — 6,778 domains. A further 120 are dropped because we could not attribute their issuer. That choice is visible rather than asserted: scoring those as mismatches instead would report 2.9% disagreement in place of 1.2%, which would be a measure of our issuer map rather than of anyone's CAA. Agreement is measured at the brand level, so a cross-signed or subordinate issuer counts as its parent. The CAA reading and the certificate reading are selected independently — each is the latest usable observation of its own kind — so they need not come from the same moment, and the figure is what we observed rather than a claim about what any domain serves right now.
Limitations: the corpus is the top-ranked slice of the web (Tranco aggregates several popularity signals; it is not a literal visit count), so adoption here reads higher than the web at large. CAA is checked by CAs at issuance time and we read it afterwards, so a record added after a certificate was issued looks like disagreement. We read CAA at the registered domain only, not at hostnames below it, and RFC 8659 tree climbing means a subdomain may inherit a policy we do not see. 476 domains publish at least one property tag this report does not evaluate — mostly RFC 9558 contactemail, plus a handful of outright misspellings (ideof, bissue) that no CA will honor either. Among records with no issue or issuewild we check the critical flag on those unevaluated tags, since RFC 8659 §4.1 turns an unsupported critical property into a refusal to issue. That flag is a byte, not a boolean — the Issuer Critical bit is 128, while this corpus also carries reserved values of 1, 2, 3, 4, 5, 9 and 10 that are not critical — so it is read with a bit mask. 391 domains set that bit on some property. We report the 1 advisory-only case as a caveat rather than reclassifying it: whether such a record blocks issuance depends on whether the CA reading it supports the property, which is not something a DNS snapshot can answer. Finally, a CAA record is a constraint on issuance, not a security grade: a single-CA allowlist on a neglected domain is worth less than a broad one that matches a real certificate lifecycle.
The data is free to cite and reuse (CC BY 4.0, credit "dns.pizza"). For questions or the raw per-domain data, get in touch.
Who can issue a certificate for your domain?
Run a health check to see your CAA record — or confirm you're one of the 87.1% that haven't set one.