AI for Content Editors: How to Create Real Quality Content

AI can write a blog post in ten seconds. Most of the time, it reads like it was written in ten seconds: generic phrases, confident mistakes, and a tone of voice that belongs to nobody.

Enonic’s content team has used generative AI every day since ChatGPT and Midjourney arrived in late 2022. The biggest lesson? What you get out depends on what you put in: the context, the rules, and the content the AI has to work with. That’s why we no longer see AI as a chat window next to the CMS, but as something that works inside it, on top of well-structured content.

This guide is for content editors and editorial leads. It covers what AI is good and bad at, why structured content makes such a difference, and the five-step method we use with Juke, the AI assistant in Enonic’s Content Studio.

What AI Can (and Can’t) Do for Editors

AI is a tool, like a word processor or a spell checker. A language model doesn’t think or know things the way people do. It predicts statistically likely text based on what you give it. That makes it brilliant at some editorial jobs and useless at others.

AI is good at:

• Drafting first versions from an outline or notes
• Rewriting text to be shorter, simpler, more formal, or more fun
• Suggesting alternatives, such as titles, intros, and meta descriptions
• Summarizing long texts
• Translating and localizing
• Spotting unclear sentences, jargon, and inconsistencies

AI is bad at:

• Knowing facts it hasn’t been given, such as your prices, products, or customer stories
• Admitting what it doesn’t know. It would rather invent something plausible
• Original insight, opinions, and first-hand experience
• Judging what your audience actually needs

In short: let AI handle the drudgery, and keep the thinking, the facts, and the final say with people.

See also our guide: AI for Content Editors » 

Real-life Examples

Content Creation: Netflix & NYT

Netflix’s personalized recommendations (sometimes down to the specific thumbnail you see) illustrate how viewing habits can shape the content each person encounters. The New York Times’ Blossom project explored a complementary approach: predicting which stories might perform well on social media and suggesting when to share them. Audience insights can surely guide editorial choices.

Content Distribution: Facebook & Spotify

Facebook personalizes feeds using signals such as user interactions and connections, while Spotify’s Discover Weekly introduces listeners to music based on their preferences and listening patterns. For content teams, the opportunity is to apply similar principles to newsletters, websites, and recommendations. Help audiences discover relevant material instead of presenting everyone with the same selection!

Content Strategy: Amazon & Airbnb

Amazon’s personalized product recommendations demonstrate the value of understanding individual interests, while Airbnb’s work on demand forecasting and pricing illustrates how predictive insights can support planning. You can do it too: Identify audience needs, prioritize topics, and plan campaigns around likely interest.

What We Learned from Prompting in a Chat Window

When we started out, we did what everybody did: ChatGPT in one tab, the CMS in another. We quickly learned that a vague prompt gives a vague answer. Ask for “the best text ever about boats,” and you’ll get nothing of the sort.

Good prompts follow three rules:

• Be as specific as possible
• Give context, such as the target audience, industry facts, and the purpose of the text
• Describe the tone of voice or style

So instead of “write a blog post about spaghetti,” you write something like: “Write a 1,000-word blog post about cooking spaghetti in the rustic Tuscan style, for busy parents in their 30s, in the style of Gordon Ramsay.” Even then, the result needs a human touch to remove generic filler and add your own data, examples, and voice.

That works, but it doesn’t scale. Every editor writes prompts their own way, so the tone drifts. The context has to be typed in again for every task. Text gets copied back and forth between tabs, and the AI never sees the rest of your content. Nobody can tell which instructions were used for what.

The fix is to move the context out of individual prompts and into the CMS itself.

Why Structured Content Makes AI Better

Structured content means your content is stored in defined fields rather than as one big page of text. An article has a title, an intro, a body, an author, and topics. A product has a name, a price, and specifications. An event has a date and a venue. A content model describes which content types you have and which fields each one contains.

This matters for AI in four ways:

• Clear context. The AI knows whether it’s working on a title, an intro, or a meta description, and what each one is for.
• Smaller, sharper tasks. “Shorten this intro to two sentences” gives a far better result than “improve this page.”
• Facts stay facts. Prices, dates, and product names live in their own fields, so they can be checked and reused instead of reinvented by the AI.
• Easier review. When the AI changes one field, the editor can see exactly what changed.

Structure helps on the way out too. Structured content can be delivered to any channel through APIs, and it’s easier for search engines and AI assistants to understand and quote correctly. Structured content becomes the source of truth for people and AI alike.

See also: Why Editors Should Use Structured Content »

The Enonic Method in Five Steps

This is how we combine structured content and AI in our own content work, using Juke, the AI assistant built into Content Studio.

1. Model your content first

AI can only respect structure that exists. If every page is a single rich-text field, the AI has nothing to hold on to. Start by defining content types and fields that match what you actually publish. It pays off for editors, developers, and AI alike.

Read more in What is content modeling? (Fun fact: we drafted that post with ChatGPT back in 2023.)

2. Set the rules once

Instead of every editor writing their own prompts, write your rules down once and give them to the AI. Juke supports site-wide custom instructions: your audience, tone of voice, terminology, words to avoid, and style rules such as plain language. Every request then starts from the same foundation.

Treat the instructions like a style guide. Give them an owner, usually the editor-in-chief or content lead, and update them when your guidelines change.

3. Draft and rework with Juke, field by field

Juke’s Content Operator lives in a chat panel inside Content Studio. It recognizes the fields of the content you’re editing, and it can write, modify, and analyze text right there in the interface. No more copying between tabs.

Typical requests look like this:

• “Suggest five titles under 60 characters.”
• “Shorten the intro to two sentences.”
• “Rewrite this section in plain language.”
• “Write a meta description based on the intro.”
• “Check this text for jargon and unclear sentences.”

4. Translate in layers

Localization is where AI saves the most time. With Enonic’s layer system, each language version inherits from the original and can then be localized. When you localize a content item, Juke’s Content Translator offers to translate it into the language of that layer. You can add custom instructions for how translations should be done.

VID Specialized University uses this to run its multilingual sites. Instead of duplicating content by hand, editors get fast, GDPR-compliant AI translations. As web editor Kjell Håkon Larsen puts it: “On a multilingual website, Juke saves us hours of translation work.”

5. Keep a human editor in charge

AI drafts. Editors decide. Before anything is published, a person should check the facts and add what only you can add: your data, examples, customer stories, and opinions. They should also make sure the text sounds like you. The AI never presses “Publish.” That’s still your job.

What About Images?

We got tired of soulless stock photos, so we started generating our own illustrations with AI image tools such as Midjourney. The same rule applies as for text: the result depends on your prompt and your patience.

A few tips:

• Describe something tangible. Abstract ideas give strange results. To illustrate a content model, we asked for a jigsaw puzzle.
• Specify style and format, such as “isometric 3D, simple design, white background,” and the aspect ratio you need.
• Iterate. Generate several versions, pick the best, and refine it.
• Expect to finish the image in an image editor, for cropping and sharpness.
• Always write alt text. It’s a field like any other, and the image isn’t done without it.

Image generation directly in Content Studio is on Juke’s roadmap.

Creative AI Tools We Use

Here are some of the creative generative AI tools we use in our daily content management:

• ChatGPT Plus: A top choice if you enjoy using prompts for content creation. It’s powered by advanced AI and can help you generate creative content in a jiffy
• Midjourney: Midjourney is a generative AI for creating photos, illustrations, and graphics. It excels at creating images from natural language prompts, providing a unique visual take on content creation
• ElevenLabs: ElevenLabs offers browser-based, AI-assisted text-to-speech software. Its nuanced speech synthesis and customizable voice options make it stand out
• Canva: This AI-enhanced design tool helps content editors create visually stunning graphics for their content. With its intuitive interface and vast template library, Canva is a dream come true for content editors
• Higgsfield: Powerful AI video generation tool (and more) with its own cinema studio, and countless video styles, effects, and models.
• Adobe Creative Cloud: Large and small touch-ups in existing material are handled by Photoshop, Illustrator, etc.

Quality, Risk, and Governance

AI makes it easy to publish more. That’s not the same as publishing better. When everyone can create content in seconds, accuracy, original insight, and a recognizable voice are what set you apart.

Ground rules we recommend:

• Check every fact. Names, numbers, dates, and quotes must be verified by a person.
• Protect your data. Don’t paste confidential or personal information into public AI tools. An assistant inside your CMS, running on enterprise infrastructure, is a safer place for editorial work. Juke runs on Google Vertex AI, which is built for security, compliance, and data protection.
• Decide on disclosure. Agree on when and how you tell readers that AI was involved. The EU AI Act also sets transparency rules for some AI-generated content.
• Measure quality, not volume. Track whether AI-assisted content performs and converts, not just how much of it you publish.
• Keep accessibility in mind. Plain language, clear headings, and alt text matter just as much when AI writes the first draft.

Get Started with AI in Enonic

The Enonic method boils down to this: give the AI good structure and clear rules, let it take the drudgery, and keep people in charge of facts and judgment. That’s how you get real quality content, not just more content.

Read more about AI in Enonic, or see Juke in action in Our New AI Will Boost Your Content Productivity. The quickest way to get going is Enonic SaaS, where we host and manage the platform for you. Compare the plans »

First published 31 May 2023, updated 9 October 2026.