Umbraco AI: My first look

I’ve been researching the new Umbraco AI offering and, with my nerdy CMS hat on, it looks really exciting. As with much of Umbraco’s documentation, the information is a little fragmented and development-focused, so I’ve tried to summarise it into a more end-user-friendly guide.

Firstly, in Umbraco’s own words:

Umbraco’s AI strategy is built around our core principle of keeping the CMS stable and AI flexible. Instead of hard-coding AI into the core, we provide a platform where AI is modular, optional, and configurable. Use AI where it adds value, extend it when needed, and stay in control as technology, teams, and your needs evolve.

It's a smart approach.

Especially while we're all in our own AI discovery phase - working out how it can support our businesses and workloads in a practical way - the ability to choose exactly how and when you use AI within your website CMS is a game changer.

How does the Umbraco AI integration work?

Umbraco has provided a foundation package that can be installed on top of the standard back-office installation (v17 onwards). Businesses can then choose exactly how to configure, control, and scale the application.

AI providers

Umbraco provides a range of AI provider and model integrations to choose from. You can even use different providers for different AI functions.

The initial release supports:

• Anthropic

• OpenAI

• Google

• Amazon Bedrock

• Microsoft Foundry

Governance

You configure your own guardrails (rules) that are applied to all AI tasks. These can include:

• Brand rules for words or phrases that shouldn’t be used

• Personal data usage restrictions

• Legal or compliance-related terminology controls

You can also create different contexts where specific tones of voice or rules apply in different scenarios, such as marketing, brand, or legal content.

Prompts

This is where the magic really happens.

Specific prompts can be written and attached to relevant fields within the editor. These prompts are configured alongside your chosen contexts, ensuring the AI behaves appropriately for each task.

Examples include, but are certainly not limited to:

• Generating article summaries

• Creating metadata

• Generating image alt text and descriptions

• Translating content

• Drafting service page content

This gives businesses complete control over which AI-powered tasks are available and where they can be used, while making it incredibly easy for content editors to use them consistently.

What does this mean for content teams?

For me, this is where the real value lies.

Most organisations aren't looking to hand content creation entirely over to AI. What they want is a way to help their teams work more efficiently, maintain consistency, and reduce the time spent on repetitive tasks.

By embedding AI directly into the CMS, content editors can generate summaries, metadata, translations, and first drafts without jumping between different tools. Because prompts, contexts, and guardrails are centrally managed, the content produced is also far more consistent and aligned with brand guidelines.

In practical terms, that means teams can spend less time on administration and more time focusing on strategy, creativity, and creating genuinely useful content for their audiences.

Copilot

You can also opt in to using Umbraco Copilot within the editing interface.

Copilot is a chat-based assistant that reads and understands the pages you're working on. You can type prompts, speak out loud, upload documents, or paste links alongside your own instructions to carry out editing tasks on the fly.

Every response gives you the option to accept, edit, or reject the output, ensuring that a human remains in complete control at all times.

Costs

The costs are paid directly to the AI provider, so there’s no markup from Umbraco. It works in much the same way as using these AI tools independently.

You pay as you go, and the number of credits used depends on the complexity of the task. As a simple example, metadata generation is relatively inexpensive, while translation tasks typically consume more credits.

Your monthly costs will depend on:

• The number of AI prompts you make available

• The number of pages they are used across

• The type and complexity of the prompts being used

You can also set a maximum credit limit for individual prompts as a guardrail, so you can be confident that editors are working within the cost parameters you decide.

Interested?

I’m currently working with the team at Frontmedia to implement Umbraco AI for a number of existing clients and will be sharing case studies in due course.

In the meantime, if you're keen to be one of the first to explore what's possible, or if you have any questions, get in touch.