A site opens in your browser but its TLS chain is incomplete
Inspect the certificate chain sent by your public HTTPS endpoint when a browser works but a checker fails, then verify the repaired chain without bypassing TLS.
By Crawlito · 6 min read
In short
- A browser can appear to work while a server omits an intermediate certificate.
- Inspect the chain served for the exact hostname and TLS endpoint that fails.
- Install the appropriate certificate bundle and verify normally; disabling verification is not a repair.
On this page
- Record the hostname and the failing client
- Inspect the certificates sent by the endpoint
- Find the service that terminates TLS
- Install the appropriate bundle through the supported workflow
- Verify from a client that does not hide the defect
- Separate chain delivery from other trust problems
- Make certificate renewal observable
- Frequently asked questions
If a site opens in your usual browser but a checker reports a certificate-chain failure, inspect the certificates the public endpoint actually sends. A browser may already have an intermediate certificate that another client lacks. Install the correct chain at the TLS termination point, then repeat normal certificate verification without bypass flags.
This is a specific possibility, not proof that every browser-versus-checker disagreement has the same cause. Hostname mismatch, expiration, an outdated trust store, or a different network path can produce a similar complaint. Preserve the exact error before changing the server.
Record the hostname and the failing client
Capture the complete public URL, error text, time, and client version. Separate a failure on the main page from a failure on an asset or redirect destination.
A page may load from www while its API uses another hostname. If only that API fails, changing the main site's certificate configuration may have no effect.
The DNS and TLS troubleshooting guide helps place the failure before the HTTP response. A robots.txt rule cannot repair a TLS handshake that never reached the application.
Repeat an ordinary request without disabling verification:
curl -v -o /dev/null 'https://www.example.com/guides/setup'Use a public page without credentials for a shared diagnostic. Verbose output from authenticated requests can contain private headers, so avoid publishing an unreviewed transcript.
Record whether the client receives any HTTP response. A certificate verification error is different evidence from a valid TLS connection followed by an application 403.
Inspect the certificates sent by the endpoint
Use the exact hostname for both the connection and Server Name Indication:
openssl s_client -connect www.example.com:443 \
-servername www.example.com -showcerts </dev/nullThis shows the presented certificates and diagnostic information. Interpret verification results using the local client's trust configuration; the command's output is not a universal verdict for every runtime.
The nginx HTTPS documentation explains the missing-intermediate scenario and why some browsers can conceal it. It also describes assembling the server certificate followed by the appropriate intermediate certificates.
Compare the served chain with the bundle provided for the actual certificate. Do not add unrelated certificates until an error disappears.
Keep the leaf certificate's hostname coverage, issuer, and validity period in the record. A complete chain does not correct a certificate issued for a different domain.
If the public hostname resolves to several endpoints, inconsistent results may mean those endpoints have different configurations. Establish that difference before assuming a client cache explains everything.
Find the service that terminates TLS
Map the public request through the CDN, load balancer, reverse proxy, and application. The component serving the public certificate may be different from the origin web server.
For a CDN-hosted site, installing a bundle on the origin does not necessarily change the public edge certificate. Conversely, a healthy edge certificate does not prove that the CDN can establish its configured connection to the origin.
Use the provider's supported inspection path for each boundary. Do not expose a private origin or switch the site to an insecure mode merely to make a diagnostic easier.
Create a small record:
| Endpoint | Certificate owner | Observation |
|---|---|---|
| Public hostname | CDN or load balancer | What external clients receive |
| Origin hostname | Hosting or web-server team | What the delivery service receives |
| Separate asset host | Asset provider | Whether its chain differs |
The owner column makes the fix assignable. Without it, teams can repeatedly renew a correct certificate on the wrong component.
Keep evidence from each boundary separate rather than combining several successful tests into one unsupported “TLS is fine” conclusion.
Install the appropriate bundle through the supported workflow
Obtain the certificate bundle from the issuing authority or the certificate manager that controls this deployment. Follow the receiving server's documented chain format and ordering.
For nginx, the documented server certificate bundle begins with the leaf certificate and then the intermediate certificates. Other managed services may ask for these fields separately.
Keep private keys private. A public certificate-chain investigation does not require sharing the corresponding private key in a ticket or report.
Validate the configuration before reloading the service, and retain the last known working configuration for recovery. The exact command and deployment mechanism depend on the hosting environment.
If the certificate is automatically renewed, repair the renewal installation step too. A correct manual upload can be replaced by an incomplete bundle during the next scheduled renewal.
Do not disable certificate verification in the checker or application as the production fix. That changes what the client accepts while leaving the server's incomplete delivery unresolved.
Verify from a client that does not hide the defect
Repeat the original failing request after deployment. Confirm that ordinary verification succeeds and that the expected page body arrives.
Then inspect the served chain again. A browser screenshot alone is insufficient because the original problem was precisely that the browser appeared healthy.
Use an additional supported client or clean environment when practical. Record its runtime and trust-store context rather than presenting two different clients as identical tests.
Check the full redirect path. The initial hostname may now be healthy while a redirected hostname still has a certificate problem.
The HTTP status and redirect guide becomes relevant once the TLS connection succeeds. A working handshake followed by a wrong destination is a separate issue.
For an intermittent incident, collect samples at different times or from the relevant delivery endpoints. One repaired server in a pool does not establish that every ordinary request will reach a repaired server.
Separate chain delivery from other trust problems
If verification still fails, read the new error rather than repeating the same installation step. A hostname mismatch, expired certificate, or unavailable trust anchor requires its own diagnosis.
Also check whether an organizational proxy intercepts TLS for one client. Compare the certificate issuer seen on the affected path with the expected public issuer, using approved network diagnostics.
Do not treat an old client trust store as proof that the public server omitted an intermediate. The served-chain evidence should support that conclusion independently.
A useful incident note distinguishes the observations: the server previously omitted a required intermediate, the bundle was corrected, and the original supported client now validates successfully. If another client remains incompatible, name that remaining boundary.
Avoid claiming that every AI crawler now accepts the site. Different clients have their own networking behavior, and a local request does not impersonate the actual crawler's infrastructure.
Make certificate renewal observable
Keep a record of the service responsible for renewal and deployment. Confirm that a renewed certificate reaches the public endpoint with the expected chain, not merely that a certificate file exists on disk.
Include the public hostname and any separate content hosts in the renewal checklist. A new image CDN or documentation subdomain can introduce another certificate lifecycle.
A practical acceptance record contains the certificate identity, deployment time, normal verification result, and a successful content fetch. None of those fields requires exposing private keys or visitor data.
For an agency handoff, assign the certificate-manager owner and the web-server owner explicitly when they differ. That prevents the next renewal incident from becoming an SEO plugin investigation.
If the hostname itself does not resolve normally, inspect the earlier DNS boundary. The DNSSEC migration guide covers one distinct failure that certificate installation cannot repair.
Frequently asked questions
Why can my browser work when curl fails?
The clients may differ in cached intermediates, trust stores, network paths, or verification behavior. Inspect the actual chain and error before choosing the explanation.
Is adding a no-verify option a valid fix?
No. It bypasses the check instead of correcting the certificate delivery or trust problem.
Does a valid certificate guarantee crawler access?
No. It removes one transport obstacle. The subsequent HTTP response, access rules, and page content still need their own checks.
After normal HTTPS verification succeeds, run the free Crawlito check on the repaired public URL to review its page-level readiness.
Filed under Access and robots.txt. Topics: tls, troubleshooting, http-status