Once the code is deployed, check the result on the production route. The build may have passed while the public domain continues to serve an old cache; one CDN zone may have updated while another has not; a data migration may have run only partially; an environment variable may have added noindex. In the first few minutes after a release, a short set of production checks selected according to the change’s risk is enough.
Confirm the new version and availability
Find a safe release marker: the application version in a service endpoint, an expected interface element, a changed header, or a static asset identifier. Do not rely on CI/CD status alone. The request should go to the public domain along the same path visitors use.
Check from a private browsing session and, if the infrastructure is distributed, from a second independent source. Clearing the local cache does not clear the CDN. The headers Age, ETag, Last-Modified and service indicators help determine which version was served.
If the new version is not visible, do not immediately redeploy. Find where the delay is: the artifact, application, load balancer, CDN, or browser. Republishing may hide the cause and make rollback more difficult.
Check availability before SEO markup
Request the homepage, changed routes, and several critical pages. Check DNS, TLS, connection time, final HTTP status code, and response size. Errors 5xx, widespread 404, redirect loops, and empty HTML require the highest priority. There is no point checking meta tags until the document can be retrieved reliably.
Compare responses for a regular browser and a search engine user agent if you use a WAF or anti-bot protection. Check the mobile version. Record a brief spike in errors during the version switch with its timestamp; prolonged or recurring unavailability is grounds for a rollback or urgent fix.
The incident response guide describes how to localize a network incident. Follow it if the problem extends beyond the specific release.
Make sure indexing is not blocked
Retrieve the production versions of robots.txt, page headers, and the source HTML. Look for an accidental sitewide Disallow, noindex, a directive for a specific crawler, and X-Robots-Tag. Check the value of nofollow separately: it affects following links on the page but does not, by itself, prevent the page from being indexed. Check the homepage and affected templates, since the directive may be added conditionally.
Compare these with the pre-production configuration. A common mistake is for a staging protection rule to end up in the built artifact or a shared CDN setting. If important sections are blocked, there is no need to wait for a search console report: the actual response already confirms the defect.
After fixing it, repeat the request and clear the relevant cache. Save the headers from “before” and “after” along with the timestamp; they will be useful if a search engine crawled the site during the problem window.
Check representative URLs for each template
Use a sample prepared before the release: a regular page, an edge case, a locale, pagination, and a missing object. Check the status code, title, H1, main text, canonical, hreflang, and internal links. For a JavaScript interface, compare the source with the rendered DOM and network requests.
Pay particular attention to shared values. The same title or canonical on different pages points to a template defect. Missing content only for objects without an optional field points to an untested data branch. Assess the common cause across the entire page family; otherwise, fixing one URL will leave the template defect in place.
If audit results seem questionable, follow the protocol in the article on rechecking SEO findings: reproduce the request and distinguish unavailability from the actual absence of an element.
Check routes, sitemap, and the bulk delta
For renamed pages, check each important “old → new” pair. The redirect should be direct, permanent, and point to a relevant final URL. The new URL should return 200, use a self-canonical, and appear in internal links. The old URL should not be in the sitemap.
Open the sitemap index and affected child sitemaps. Check the XML, domain, protocol, count, and several new or removed URLs. If the sitemap is generated by a background job, wait for that job to finish and record the time. A successful application deployment does not prove that a separate generator has updated.
The article on sitemap errors describes cleanup and troubleshooting. Do not resubmit the sitemap to search engines while the public file still contains the old state.
Run a short bulk crawl
After the manual checks, run a limited crawl of the affected sections. Compare it with the baseline: URL count, status code distribution, indexability, canonical, metadata uniqueness, and depth. This crawl catches branches that were not included in the manual sample.
Do not mix old known issues with release regressions. The report should show the delta: what appeared specifically after publication. A single new 404 from a removed test URL and a widespread disappearance of links have different priorities.
Uptime Monitor reChecker checks availability on a schedule and stores the HTTP status code, response time, and outage history. Connect it to monitor the site after a release, and continue checking content and canonical with manual requests.
Also check background processes that the release has only started: static page generation, sitemap rebuilding, cache clearing or warming, image queues, imports, and updates to the site’s internal search index. A job starting successfully does not mean it completed successfully. Check its status and errors, and review several results from the beginning, middle, and end of the set. With partial execution, seemingly random 404 may actually follow the queue order.
If the release changed structured data, check the JSON-LD on live pages and compare its facts with the visible content. Price, availability, name, author, and date should not differ because of different cache update times. Syntax validation matters, but production checks should also confirm the semantics of the values.
Decide: monitor, fix, or roll back
A rollback is justified when a confirmed regression broadly blocks access or indexing, breaks routes, removes main content, or affects critical pages, and no quick, safe fix is ready. The decision should account for migration reversibility and whether the old version is compatible with the new data.
A non-blocking defect can be fixed in the next release if its URLs, impact, owner, and due date are recorded. Do not leave it as “SEO later”: after a release, it gets lost among product tasks. Monitor a temporary external outage using logs and multiple locations, without changing the application unless a connection is confirmed.
Repeat the same test after every action. A rollback is complete only when the public route once again shows the expected state.
For a hotfix, create a new version marker and repeat the entire short check cycle, including the URL where the defect was noticed and adjacent templates. An urgent change can affect shared configuration. Record which version ultimately remained in production so a later check is not performed against an artifact that has already been replaced.
Keep a release log
Record the version, publication time, checked URLs, results, deviations found, and decisions made. Attach responses or screenshots of key signals. Note which checks were not completed because of authentication, generator delays, or an unavailable external service.
Repeat critical checks a few hours later or during the next working period, and review search consoles. Their data does not update instantly and confirms consequences later. Immediately after a release, rely on production requests.
Close the first check cycle by recording the time and version, results for critical URLs, robots/sitemap status, and the decision to “monitor,” “fix,” or “roll back.” Schedule the next check with background generators and search console updates in mind. This gives the team a specific time to reassess delayed effects.