When a client asks us to "add SEO" at the end of a build, the honest answer is that the expensive decisions were made months earlier.
Rendering is the first SEO decision
A client-rendered catalogue page can be indexed, but it competes with a server-rendered one that loads in a third of the time. On the Aurora engagement, moving 214,000 catalogue pages from client rendering to incremental static regeneration was the single change that moved organic sessions 61%.
Cache policy is a ranking factor by proxy
Core Web Vitals are measured on real users. That makes your cache hit ratio, your image pipeline and your font loading strategy into ranking inputs. We enforce a Lighthouse budget in CI for exactly this reason: LCP under 2.0 s on a throttled mobile profile, CLS under 0.05, no render-blocking third-party script above 20 KB.
Structured data as a contract
Schema.org markup is generated from the same data that renders the page — never hand-written into a template. If the product price changes, the structured data changes with it, because both read the same field. Hand-maintained JSON-LD drifts within weeks.
Internationalisation done properly
Eleven locales means eleven canonical decisions and a full hreflang matrix, including x-default. Getting this wrong does not produce an error message; it produces the wrong page ranking in the wrong country, quietly, for months.
The checklist we apply
- Server-rendered content for anything meant to rank
- One canonical per URL, generated not authored
- hreflang matrix with
x-default, validated in CI - Sitemaps generated from the database with real
lastmodvalues - Structured data derived from source data
- Performance budget that fails the build



