15 things every Contentful enterprise project gets wrong in the first 6 weeks

Three of the most expensive bugs in any Contentful enterprise build are not bugs at all — they are missing primitives. They look like content problems for the first six weeks, then surface as 404 storms in week seven.

• 1. No dynamic link references. Editors paste hard-coded URLs into rich text and CTAs. Refactor a slug, and every link to it silently breaks. Production answer: a Link content type that references the target entry, not a string. Solved in the starter with a typed linkReference resolver across every block.
• 2. No page hierarchy. Pages live in a flat list. There is no parent/child relation, no breadcrumb derivation, no way to ask "what's under /products?". Production answer: a self-referential parent field on the page content type, with depth-bounded traversal. Solved in the starter via a single getPageTree(locale) call wired into nav and breadcrumbs.
• 3. No CMS-driven navigation. Header, footer, mega-menu — all hardcoded into components. Marketing wants to add a top-level item; engineering ships a release. Production answer: navigationMenu / navigationGroup / navigationItem content types with optional mega-menu media. Solved in the starter — restructure the entire site nav from Contentful, no deploy.

Every Contentful project starts with one locale and "we'll add more later". Later is always a six-week sprint, because the foundation has to land before any real content can be translated.

• 4. No locale fallback chain. A Danish entry with no Danish copy returns null instead of falling back to English. The page renders empty. Production answer: a fallback chain baked into every CMS query, with the fallback declared per-locale in code, not per-field by editors. Solved in the starter — every cmsFetch applies the chain transparently.
• 5. No localized slugs. URLs stay English in every market. SEO suffers in Denmark, France, and Germany — exact-match local-language slugs are a measurable ranking signal. Production answer: localized parent and child slug fields, joined per-locale at build. Solved in the starter with a localized URL builder shared by sitemap, hreflang, and the language switcher.
• 6. No CMS-driven dictionaries. UI strings ("Read more", "Submit", "Cookies") live in JSON files. Marketing wants to A/B-test "Get a quote" vs "Request pricing"; engineering ships a release. Production answer: a localizedDictionary content type with {token} interpolation. Solved in the starter — every UI string is editable in Contentful.

The SEO checklist is not exciting. It is also non-negotiable. Every gap here costs measurable organic traffic, and most are invisible until a Search Console regression report lands six months in.

• 7. No redirect engine. Every URL change is an engineering ticket. Marketing waits a sprint to retire a campaign URL; SEO bleeds in the meantime. Production answer: a redirect content type with regex/wildcard support, evaluated at the proxy edge. Solved in the starter — editors ship a redirect in 30 seconds, no deploy.
• 8. No auto slug-change redirects. Editor renames a page; the old URL dies silently. Production answer: a webhook on entry update that diffs the slug field and writes a redirect entry automatically. Solved in the starter — slug refactors are safe by default.
• 9. No JSON-LD, hreflang, or llms.txt. Rich results get skipped, locale alternates get ignored, AI crawlers get nothing. Production answer: per-page JSON-LD, locale-aware generateMetadata with hreflang, and a /llms.txt route with an AI crawler policy. Solved in the starter — discoverability across Google, Bing, and the LLM crawlers comes pre-wired.

Editors don't read your README. They click around for thirty seconds, hit something that surprises them, and decide the CMS is broken. Three things consistently break that thirty seconds.

• 10. No Live Preview. Editors save, switch tab, hard-refresh, swear. Production answer: Contentful Live Preview wired end-to-end with the preview-mode token, draft fetch, and inspector overlays on every block. Solved in the starter — preview is on day one, not month four.
• 11. No tag-based ISR invalidation. A content change either does nothing or nukes the entire CDN. Production answer: revalidateTag on a per-entry-type tag, dispatched from a Contentful webhook. Solved in the starter — the webhook is registered automatically on pnpm setup.
• 12. No schema-driven forms. Forms are hardcoded into components. Marketing campaigns wait on engineering. Production answer: a form content type with a typed field schema and a single renderer. Solved in the starter — editors ship a new lead-capture form in an hour, not a sprint.

These three don't break in week six. They break in month nine, when a junior engineer ships a regression nobody catches because the contract was implicit. By then the surface area is too large to retrofit cheaply.

• 13. No type-safe content models. Stringly-typed CMS access, runtime errors on field renames, zero compile-time safety. Production answer: graphql-codegen against the live CDA schema, regenerated on every pnpm setup run, with a drift detector that fails CI when types and CMS diverge. Solved in the starter.
• 14. No hardened CSP. A default Next.js CSP allows everything; security review pushes it to the back of the backlog. Production answer: a strict Content-Security-Policy header with nonces, locked down for Contentful, the analytics provider, and the form endpoints. Solved in the starter — passes Mozilla Observatory A+ on day one.
• 15. No documented migration path. The CMS schema drifts. Six months in, nobody can spin up a new environment that matches production. Production answer: cf:migrate, cf:seed, cf:reset commands with idempotent migration scripts checked into the repo. Solved in the starter — a new environment is five minutes, not five days.