How do you prevent metadata and canonicals from drifting across multiple front ends connected to one headless CMS?

Separate content identity from presentation destination.

The CMS should store the content entity once, while each publishing channel has an explicit URL mapping and SEO policy. Do not put one universal hard-coded canonical URL on a content record that can legitimately render on multiple sites.

Define a primary web canonical when the same public content is syndicated across front ends. The rendering layer should compute the canonical from the destination's domain, route, locale, and publication role using validated configuration.

Keep tenant/site settings isolated: hostname, brand name, organization entity, default social image, locale, robots policy, and URL conventions. Shared code is fine; shared client metadata is not.

If a secondary front end intentionally duplicates content, decide whether it should canonicalize to the primary source, be noindexed, or provide a differentiated experience. Do not leave the relationship accidental.

Generate sitemaps per public site and include only that site's canonical URLs. Structured data identifiers should use the appropriate canonical entity/page URLs for the destination.

Add integration tests that render the same CMS record through each front end and verify title, canonical, hreflang, robots, schema, and links are correct for that site.

Headless architecture increases flexibility, but it also moves SEO responsibility into routing/configuration. Make canonical ownership explicit in the publishing model.