Publishing and CMS Development

Usually you should not, and we will say so. WordPress, Ghost and the established publishing platforms cover standard editorial workflow well. Custom work earns its place when your workflow does not fit the tool — multi-title operations with shared content, embargo and legal review steps, syndication rules, print and digital alignment, or rights and licensing built into the content model. If none of those apply, buy the platform.

It depends on whether your traffic or your loyalty is the asset. Metered suits high-volume general news where casual readers convert slowly. Freemium suits publications with a clear premium tier. Hard paywalls suit specialist and trade titles where the audience has no substitute. The more useful question is not which model but when the wall appears, which is where propensity modelling changes the numbers more than the model choice does.

Yes, and it is the part that most often goes wrong. Publishing archives accumulate decades of URLs, many with inbound links and search authority, and a migration without a complete redirect map loses that permanently. We build the URL map before anything moves. We have done this on our own site after a migration that shipped without redirects, and the recovery cost more than the map would have.

Good enough to triage, not good enough to decide alone. Models reliably catch obvious abuse and reliably struggle with context, satire, quoting and coded language — all of which are common under news articles. The workable pattern is automated action on high confidence, a human queue for anything ambiguous, and an appeal path. Fully automated moderation on news comments produces complaints you will have to answer publicly.

Not in anything we build. AI is useful for summaries, headline variants, tagging, translation drafts and archive search — tasks where an editor reviews the output before it publishes. Generated articles published without editorial review damage the thing a publication sells, which is trust, and search engines have become markedly less tolerant of scaled content. We build AI into the workflow, not in place of it.

For a defined scope we target a working MVP in 30 days and a core production build in 90. Publishing extends when archive migration, print system integration or a multi-title rollout are involved. Archive migration in particular is usually underestimated — it is a data and URL problem more than a content one.