AI in the CMS: steering the ecosystem
Navigating the signal and the noise, opportunities and pitfalls in AI-powered content management
With the industry at large embracing Artificial Intelligence capabilities left and right, today we get to take a look at what this might look like within Wagtail! There are many such capabilities we could leverage in a CMS. Take this image editing form – can you spot all of the different ways in which machine learning and large language models (LLMs) could be used to help CMS users?

Automated tagging, descriptions, focal points – so many opportunities to increase the automation of common tasks. But there are clear pitfalls here too, tucked out of sight by the AI hype machine. We can navigate this: we can make room for AI integrations, where warranted.
AI guiding principles
When exploring which common CMS tasks could be supported by AI, we thought we needed clear guiding principles. We have the Zen of Wagtail when it comes to product decisions, and our open source roots also set high standards of transparency and openness. For AI, here’s what we’ve arrived at for now:
- No AI dependency in Wagtail core: all opt-in via packages from Django’s vast ecosystem.
- Responsible approach to AI: high alignment with our values; ethical, sustainable, transparent, privacy-preserving.
- Model and provider agnostic: compatible with a wide range of open source models, not just flagship proprietary ones.
- Only the right AI: focus on the use cases where there is a definite improvement.
- Human in the loop: preserve the user’s autonomy and agency.
Those principles are designed to help us avoid concepts like AI slop generation, or introducing more dependencies on this or that tech giant’s products that might unilaterally decide your site’s content is fair game for their AI training. As site implementers and CMS users, you’re in control, you have many options to take ownership of AI use and its required infrastructure (and its environmental footprint, which only some providers report on).
AI-enhanced features
With those principles in place, here are some of the concepts where we think leveraging AI could be beneficial:
- Content summaries: for example page descriptions, or even titles.
- Image descriptions: captions, alt text, ideally contextual (adjusted for surrounding content).
- Content classification: auto-filling categories or tags.
- Related content suggestions: encouraging linking to existing pages where suitable.
- Translations: making it simpler to create multilingual sites or even one-off pages.
- Qualitative content feedback: think readability improvements or style guide alignment.
- Quantitative feedback: repeatable metrics based on the page’s content.
- Improvements suggestions: based on automated accessibility and SEO checks.
- PDF to HTML conversion: a common requirement to make content more accessible.
A lot of those tasks often fall through the cracks when done manually – for example, alt text quality is very low in real-world sites (40% of alt text is just one word?!). There is clear room for automation with AI, but also a clear need for human supervision over the results (AI is particularly good at producing believable nonsense and reinforcing stereotypes).
Example: alt text
Focusing on a specific example, our accessibility team identified lacklustre image descriptions as a common issue in their 2024 accessibility statistics of Wagtail sites. So common we even compared real-world alt text with AI-generated ones:

Those findings led to the creation of new contextual alt text capabilities, informed by our commitments to accessibility standards like ATAG 2.0. The ideal interface encourages creating alt text contextually, but also reusing any existing image descriptions that might be appropriate. And with AI in the picture, it can create multiple suggestions for the user to decide on:

Those techniques have been around for a while, but tended to be limited to specialized machine learning services or flagship large language models. Now they are available in open source models that have a much clearer track-record when it comes to ethical factors like bias and environmental footprint. Mozilla are even experimenting with alt text generation running locally in the browser!
This is a perfect case of thoughtful AI that fits well with Wagtail: fully opt-in, feasible with open source and low-footprint models, leaving the user in control of their content.
New capabilities
To guide our ecosystem towards those thoughtful uses of AI, the upcoming Wagtail 7.1 release comes with additional opportunities for AI-powered customizations in the editor experience. Here are a few examples:
- Supporting client-side UI customizations in block editing interfaces.
- New APIs to extract page content from live previews. To compute custom metrics, or prompt the AI for feedback.
- New experimental docs for client-side components. So package maintainers have a better sense of which advanced UI elements are ready to reuse.
We expect those improvements will combine well with the admin’s extensive extension capabilities to power those new features.
New pitfalls
Although the above capabilities are promising and have been carefully selected, there are still clear risks we’ll have to navigate:
- Usability issues – favoring quality over quantity when CMS users interact with the AI
- Copyright concerns if the LLM was to create new content based on training on copyrighted materials.
- Environmental footprint, if the features only worked with heavyweight models.
- Ethical and bias issues, depending on how the models are trained.
We’ll be open about those challenges, make sure to learn from existing standards (Responsible AI, Frugal AI) but also do our own research where warranted.
Our direction
We want a thriving ecosystem of packages with AI-powered features. With open source, transparent, lightweight models that don’t burn the planet in the process. Over the next few months, we’ll be exploring what those new capabilities make possible. Join us! This is impactful work for our industry, and we’d love to showcase some of it at Wagtail Space 2025 in October (sign up / propose a talk if you haven’t already!).
And come talk to us if you’re interested in sponsoring those features, or if you want to build a package to leverage those AI-friendly customization options.