Presented by Cantarus and Umbraco | May 2026
Featuring Phil Whittaker, Staff Engineer (AI) at Umbraco; James Jemmett-Page, Senior Consultant at Cantarus; Jon Whitter, Head of Development Services at Cantarus and Umbraco MVP; and Dani Barker, Marketing Services Manager at Cantarus.
This is the second instalment in our quarterly AI in Membership webinar series, co-hosted with Umbraco. While the first session focused on AI governance, organisational maturity and where to start, this time the conversation turned to practical application, exploring how AI capabilities within a CMS can deliver tangible value for membership organisations.
The panel discussed the content challenges many organisations face, demonstrated Umbraco’s built-in AI functionality and looked ahead at what's coming next on the product roadmap. Here's a summary of the key themes from the webinar.
Before getting into the tooling, the panel addressed a reality familiar to most content teams. Producing even a straightforward piece of content, whether that's a blog, membership update or policy summary, often involves far more work than simply writing it.
“In one word, painful,” said Dani Barker. “You’d need to do all of that research manually. You’d have to pull your notes together. Once you’ve got a draft, you’d need to be sharing that Word doc to colleagues, lengthy review and approval. There’s arguably too many people involved.”
Generative AI tools like ChatGPT, Claude and Copilot have undoubtedly sped up the drafting process, but they haven't removed the bottlenecks that come afterwards.
As Dani put it:
“It’s not necessarily solved all of these problems. If anything, it’s just moved the bottleneck. You still get a Word document that you need to share around your colleagues, review and approve and edit, put into the CMS, review and approve and publish. It’s only really accelerated that initial part of the process.”
The efficiency gains, then, are less about drafting content faster and more about embedding AI capability directly into the content management workflow, where the reviewing, approving and publishing actually takes place.
That's exactly where Umbraco’s AI tooling is designed to help.
Phil Whittaker, who leads AI development at Umbraco, demonstrated how AI-assisted content generation works directly within the Umbraco 17 back office.
At its simplest, that means generating content such as page summaries or description fields without leaving the CMS. More powerful, though, is the ability to control how the AI behaves through what Umbraco calls profiles, prompts and contexts.
Profiles connect to specific AI models and define settings like temperature, determining how creative or constrained the output should be. Organisations can set up multiple profiles for different purposes: one for legal content, another for marketing copy and another for editorial, each backed by a different model or configuration.
Prompts sit on top of profiles and provide task-specific instructions, whether that's generating a summary or rewriting a paragraph for a particular audience. Because prompts also use the context of the page being edited, the AI works with the existing content rather than generating text in isolation.
Contexts allow organisations to inject additional information into the prompt: brand guidelines, product terminology or organisational values. Phil demonstrated how the same summary field produced noticeably different output before and after a brand voice context was applied:
“This is really important to be able to do. This is really important to be able to control exactly how you want your organisation to come across and really to take control of the AI, to keep the AI on a tight leash.”
Jon Whitter explained that the same approach allows organisations to apply different rules across different parts of their website.
“It gives you the option of doing that differently in different areas of the content, having more legal, professional product pages or white papers, but your blog a bit more personal. You’ve got that level of control.”
James Jemmett-Page noted the membership-specific relevance:
“There’s a use case when you’re drafting content for different audiences. For example, if you have a student membership cohort and you want to utilise a slightly differentiated tone of voice. The tool enables you to do that with quite a high level of precision.”
The panel also demonstrated Umbraco AI’s guardrail functionality, which prevents sensitive information from being sent to an AI model. During the demo, an attempt to send an email address to the model was immediately blocked and logged in an audit trail.
For membership organisations, where content teams often work with personally identifiable information (PII), member data and commercially sensitive material, that level of protection is significant.
As Jemmett-Page observed:
“One of the biggest risks when onboarding a new employee is that they might make a mistake early on and end up sending personally identifiable information off to an LLM. That functionality would really help to speed up that onboarding process.”
Another recurring theme was what Phil described as “model sovereignty”: the principle that organisations shouldhave full control over which AI models they use, where their data goes and how much they pay for it.
Umbraco AI integrates with all the major model providers (Google, Amazon and Microsoft Azure), allowing organisations to choose their provider and specific model for each use case. This stands in contrast to CMS platforms that lock users into a single vendor’s AI offering or resell tokens at a markup.
As Phil explained:
“We come across vendors that are doing the same thing; they’re making a markup and limiting the models that you have access to. That is not our approach. Umbraco is very much about being open and honest and transparent.”
The practical benefits of this flexibility cover cost management, portability and data residency. Organisations can use a frontier model for complex tasks and a cheaper model for more routine work like generating alt text. If a provider raises prices or deprecates a model, the organisation can switch without rebuilding their AI integration. The ability to choose providers that meet specific regional data handling requirements is a consideration Phil flagged as increasingly important, noting Umbraco’s interest in European model providers like Mistral.
James underlined the commercial reality behind this:
“A huge amount of investment has gone into AI infrastructure. These companies are going to want to recoup those investments. Not all of these LLM providers are going to survive. It’s likely they’re going to be thinned out relatively brutally over the next couple of years. Portability is going to be incredibly important.”
The webinar also looked at upcoming additions to Umbraco’s AI capabilities.
One of the biggest additions in development is an AI co-pilot that sits alongside the content editing experience. Rather than clicking through different menus, editors will be able to interact with the CMS conversationally to update page content, pull in information from external documents and complete common content tasks.
As Phil explained, the goal is to make it:
"As frictionless as possible to work with AI within Umbraco, inside of Umbraco and outside of Umbraco as well."
Another upcoming feature is a built-in evaluation framework, allowing organisations to test prompts and compare the performance of different AI models before rolling out changes.
As Phil said:
"When your provider of choice deprecates the model you were using, you can switch the model, run the tests, and see that everything's still being generated to the quality you were happy with on the previous model."
Further down the roadmap, Umbraco plans to introduce workflow automation that chains multiple AI actions together. Saving a page, for example, could automatically generate a summary, page title and meta description in a single step.
Phil also described how AI could help organise content behind the scenes.
"Having something that is contextually aware of what your content actually is, evaluating the quality of your headers, giving you suggestions, as well as organising things. Upload an image, let the CMS work out what folder it should go in, automatically alt text it, tag it."
The panel also drew an important distinction between AI in Umbraco and Umbraco in AI.
The first refers to using AI directly within the CMS. The second is about making Umbraco content available to external AI platforms such as Microsoft Copilot, Claude and ChatGPT through Model Context Protocol (MCP) servers.
This would allow organisations to edit content, access analytics and work with personalisation data from their preferred AI assistant, rather than only through the CMS interface.
Later on the roadmap, Umbraco plans to introduce agent-based tools to help developers implement websites more efficiently by recommending best practice for areas such as content modelling, site configuration and information architecture.
For agencies like Cantarus, Jon sees this as a way to improve both speed and consistency.
"When a task comes in, we've got all these agents and skill sets that will greatly elevate our ability to deliver good quality, best practice implementations for clients."
James also shared that he recently joined the Umbraco AI Advisory Board, a working group made up of content editors, developers, website administrators, senior stakeholders and agency professionals. The board’s role is to ensure the product aligns with user needs across different roles, from foundation owners who define tone of voice and guardrails through to content editors and reviewers working with the features day-to-day.
“One of the first exercises we’re working through is to look at the various personas that work with AI in Umbraco, really just to ensure that the functionality is fully considerate of each specific role and that we minimise the friction for each of those user types,” said James.
The panel also highlighted an area where AI can deliver immediate practical value: technical SEO and accessibility.
AI is particularly well suited to taking existing content and repurposing it for different formats. That's exactly what's needed for tasks such as generating metadata, schema markup and structured data, helping search engines and assistive technologies better understand and use your content.
Phil said:
“Technical SEO and technical SEO writing is quite hard to do really well, but it’s something that’s very valuable. Agents and AI are perfect for that, because you can set up your models, your structures, your prompts to enable you to do that in the best possible way and take a lot of the weight off.”
Jon expanded on this with the example of JSON-LD schema, the structured data that powers rich search results, recipe cards on smart displays and other machine-readable content formats. For content editors, producing this markup is typically either confusing or entirely invisible. AI-driven automation could handle it as part of the publishing workflow, drawing on the content already on the page to generate the appropriate schema without manual intervention.
Accessibility is another area where AI could make a meaningful difference. Generating alt text, maintaining clear heading hierarchies and applying consistent metadata are all tasks that AI can help automate, reducing the reliance on individual editors remembering every best practice.
In response to audience questions about training and support, James confirmed that Cantarus will be providing consultancy and guidance to help clients deploy Umbraco AI once version 17 is in place. He was clear, however, that organisations will need to do preparatory work of their own.
“You will need to start doing some thinking yourselves internally, because things like guidance on tone of voice, the guardrails that you want to put in place, the choice of model, ultimately that will need to be led by you, by your organisation. But we will be providing consultancy to help guide you through that process.”
Umbraco AI is available as a free, open-source package. While it isn't installed as part of Umbraco 17 by default, it can be added by your implementation partner without any additional licensing costs.
The biggest takeaway from the session was that AI delivers the most value when it's embedded into the content workflow, rather than treated as a separate tool. Features such as AI profiles, brand-aware prompts, model flexibility and built-in guardrails give organisations a way to adopt AI without sacrificing quality or control.
For organisations planning or already undergoing an Umbraco 17 upgrade, now is a good time to start preparing. Defining your tone of voice, establishing governance and deciding which AI models best suit your organisation will make it much easier to introduce AI when the tooling is ready.
We help organisations implement AI in a way that's practical, secure and built around the way your teams work.