Automating your website's content workflow
How the steps between an idea and a published page can be automated, and which steps must stay with people.
The short answer
Publishing a page on a website involves far more than writing it: a brief, some research, a draft, review and corrections, a title and description for search, images with alt text, structured data, links to and from related pages, publishing, and then keeping the page current for years. Most of those steps are mechanical and most are often skipped, which is why so many business sites have pages with no descriptions, images without alt text, no internal links and content that has not been looked at since launch. The mechanical steps automate well: AI drafts the metadata and the alt text for a person to approve, proposes internal links, generates structured data from the content model, and the pipeline checks every rule and publishes. Drafting itself can be assisted but not delegated, because the expertise, the position and the facts must come from people who know the business. The pipeline then enforces quality before anything goes live: length, forbidden phrases, links that resolve, images that exist, accessibility checks.
The workflow, step by step
| Step | Who or what does it | Automation |
|---|---|---|
| Idea and brief | A person who knows the audience | Templates; suggestions from search queries and support questions |
| Research | A person, with AI to gather and summarise sources | Source collection and summaries, always checked |
| Draft | A person, with AI to structure, tighten, propose | Drafting assistance from the brief and house style |
| Review and facts | A person who knows the subject | Checklists; flagged claims that need a source |
| Title and description | AI drafts; person approves | Length and style enforced |
| Images and alt text | Person chooses; AI proposes alt text and crops | Optimisation, formats and sizes automatic |
| Structured data | Generated from the content model | Fully automatic |
| Internal links | AI proposes related pages; person confirms | Suggestions from the content graph |
| Quality checks | The pipeline | Automatic: rules, links, images, accessibility |
| Publishing | The pipeline | Automatic on approval |
| Staying current | AI flags stale pages, broken links, changed facts; person reviews | Scheduled checks and reminders |
Setting it up
- Model the content with the fields each page type needs, including metadata, images with alt text, topics and related pages.
- Write the rules that matter to you and encode them as pipeline checks.
- Add drafting assistance for metadata and alt text, with approval.
- Add link suggestions from the content graph.
- Generate structured data from the model automatically.
- Publish through the pipeline with previews and checks.
- Schedule staleness checks: pages not reviewed in a year, links now broken, facts likely changed.
- Keep the person on brief, facts, review and final approval.
What the person keeps
The decision about what to write and for whom. The knowledge only the business has: how the work is actually done, what goes wrong, what the honest answer is. The review of every fact and claim. The final approval. The voice. Everything else, the descriptions, the alt text, the links, the structured data, the checks, the publishing, the reminders, is work that people were never good at remembering and that machines never forget.
What this means for you
Automate the steps around content that people skip: metadata and alt text drafting with approval, link suggestions, structured data, quality checks and publishing through the pipeline, and reminders when pages go stale. Keep people on the brief, the facts, the review and the voice. The site stays complete and current without nagging, and the content stays yours.
Frequently asked questions
Can AI write our website content?
It can draft from a brief, restructure, tighten and propose alternatives, and it does those quickly. What it cannot supply is what makes content worth reading: your expertise, your position, your examples, your facts. Content drafted by AI from a thin brief and published unchanged is generic, sometimes wrong and increasingly recognisable. Content where a person who knows the subject briefs, corrects and finishes, with AI doing the mechanical parts, is faster to produce and still yours.
Which steps are worth automating first?
The ones people find tedious and skip: writing meta descriptions, adding alt text, proposing internal links to related pages, generating structured data, checking that links resolve and images exist, and flagging pages that have not been reviewed in a year. Each is small, each is often left undone, and together they are much of what separates a maintained site from a neglected one.
How does the pipeline help with content?
It treats content like code: every change is previewed, checked and published through the same automated steps. Checks can enforce your rules, description length, forbidden words, house style, links that resolve, images with alt text, accessibility, and fail a publish that breaks them. The rules you would otherwise remind people about are enforced automatically, and the content that reaches the site meets them every time.