Agent in practice: A practitioner's guide to the content lifecycle

Magnolia Agent in practice: A practitioner's guide to the content lifecycle

Content operations often break at the seams between planning, creation, publishing, and maintenance. This guide is designed to bridge those gaps. By mapping Magnolia Agent’s capabilities directly onto the stages of your daily workflow, you can stop switching between disconnected point tools and start managing a single, connected content lifecycle.

Magnolia Agent streamlines your entire content lifecycle—planning, creation, publishing, and optimization—so our AI tools fit perfectly into the workflow you already use.

Before writing anything, you need to know what to write and what to fix. Ask Agent Chat to sweep your site, and Agent Core ranks every page into a work queue: quick wins with fixable metadata gaps, declining pages losing traffic, and pages that need no attention at all. You get a prioritized table and a proposal to fix the quick wins immediately.

Example prompt: "Are there any pages with missing metadata I can fix quickly?"

Once you know what to work on, Agentic Chat and the generation dialogs handle the heavy lifting: full pages, individual components, or structured stories from a topic and a template.

Doc-to-Page is the capability that shows up most in practice, so it's worth walking through the actual steps. In the Pages app, click Convert doc to page, enter a page title, and choose a layout — a saved set of components tied to a template, which you can either build from a blank template or generate from an existing page's structure. Then either upload a .docx file (converted to text locally in the browser, never sent to an external service) or paste content directly into the built-in editor, and click Create Page.

Image handling runs on a defined resolution order, which is worth knowing before you rely on it:

If content lands in the wrong component or a section runs long, that's a known edge case — the AI is placing content by how it reads the document's structure, and the fix is usually a quick manual move or a tighter layout definition, not a re-run from scratch.

Example prompt: "Convert this PDF brief into a landing page."

Getting content live everywhere it needs to be is its own stage. Translation and localized metadata generation let you take a page live in one market and adapt it — layout, copy, and metadata together — for every other configured language in a single conversation. Content variants let you personalize for different audience segments without duplicating pages.

Example prompt: "Translate this page to French and generate localized metadata."

In practice: Global Blue used this to scale translation across its site without a proportional increase in manual localization work — taking a page live in one market and rolling it out to others in the same conversation.

In practice: Rentokil used this to batch-generate alt text across roughly 35,000 images in 25 languages, and Rentokil applied the same tagging capability to bring an untagged image library up to date for search and asset management — the kind of accessibility and DAM housekeeping that's accurate but tedious enough that it otherwise never gets done at scale.

Once a page is live, the agent keeps it performing, and this is where the conversational format earns its keep — you get a diagnosis, not just a metric. A few real examples of what that looks like in Smart Chat:

GEO optimization runs alongside this, generating the structured data — JSON-LD schema, voice summaries, entity mentions — that AI-powered search engines and voice assistants rely on to find and cite your content.

Example prompts: "Why is this page not getting traffic?" / "Fix all quick wins."

None of the above is locked to what Magnolia ships. Agent Chat connects to any MCP-compatible server — file systems, analytics platforms, back-office systems — so practitioners can call external tools without leaving the editor. On the developer side, the Magnolia MCP Developer's Server gives AI coding assistants like Claude, Cursor, and GitHub Copilot direct, governed access to your Magnolia project. And across every stage above, you choose the model: OpenAI, Azure OpenAI, AWS Bedrock, Google Gemini, or a model you host yourself, without changing your content setup.