How to get started with Wagtail AI: a practical guide

What is the Wagtail AI package?

Before diving in, it is worth being clear about what Wagtail AI actually is. It is an officially supported package that adds AI-powered features to the Wagtail CMS editor. It is completely separate from Wagtail core, which means it is entirely opt-in. If your organisation wants to keep AI away from your site, you simply do not install it.

For organisations that do want to explore it, the package adds capabilities like:

• AI-assisted image alt text and titles
• Meta description and headline suggestions
• In-editor prompting and text editing
• Content feedback via a Checks panel, based on a style guide or custom prompt
• Related page suggestions

Everything is designed around a single principle: humans stay in the loop. The AI surfaces suggestions; editors decide what to do with them.

Before you install the package, though, it is worth taking a step back. The most common mistake organisations make is treating AI adoption as a purely technical decision and skipping straight to implementation. Getting a couple of things in order first will save a lot of headaches later.

1. Sort your AI policy first

If your organisation works in health, advocacy, regulation, or charity, your audiences trust you with sensitive information, and that trust needs to inform every decision you make about AI. Having a clear position before you start means your developers know what they are configuring for, your editors know what is expected of them, and your leadership is not caught off-guard.

Your policy does not need to be a weighty document to get started. At a minimum, it should cover: what you are comfortable using AI for, which providers are acceptable given your data and privacy requirements, and who is accountable when AI plays a role in your content.

If you are not sure where to begin, we have built an AI Principles and Policy Builder specifically for charities and public sector organisations. It is a guided tool that produces a tailored first draft based on your values, your audiences, and the legislation relevant to you, including UK GDPR, the Equality Act, and Charity Commission guidance. It is not a substitute for legal advice, but it gives you a practical, editable starting point. Get in touch to request access to our AI policy builder.

2. Choose your AI provider

One of the most important things to understand about Wagtail AI is that it is completely model and provider agnostic. You are not locked into OpenAI, Anthropic, or any other specific platform. As long as a provider has an API, it will work with the package, which gives you real flexibility to choose something that fits your policy and your values.

If data sovereignty is a concern, there are strong alternatives to the big commercial providers. Apertus, for example, is a Swiss model built with copyright compliance and data protection in mind, and is worth considering for organisations with strict requirements in this area.

Environmental impact is another factor worth weighing up. By choosing open-source models hosted with providers like Neuralwatt, Scaleway, or Mistral; you can track the energy usage and carbon footprint of your AI requests at a per-query level. For organisations with sustainability commitments, that kind of transparency is useful rather than just a nice-to-have.

3. Get the package installed

For developers, installation is straightforward and well documented. The Wagtail AI installation guide walks through everything you need. Most teams find it takes a couple of hours to get up and running.

For content or digital leads, this is the moment to loop in your development team or your Wagtail partner. If you work with Torchbox, speak to your client partner and we will help you get set up. If you have an in-house team or work with another agency, the documentation linked above is the right place to point them.

4. Prepare your content guidelines

The quality of what you get out of AI depends heavily on what you put in. The good news is that once the package is installed, content and digital leads can configure the AI prompts and Checks panel directly in the Wagtail UI, without needing to go back to your developers. To make the most of that, gather your tone of voice guidelines, style guide, accessibility standards, and any other brand documentation you have before you start.

With your style guide in place, the AI can give editors specific, qualitative feedback on their drafts against your own standards rather than generic suggestions. The more precise your documentation, the more useful and on-brand the output will be. This is also a good moment to update anything that has drifted out of date.

5. Start small and low-risk

Once everything is configured, the best approach is to start with tasks that are time-consuming, low-stakes, and easy for an editor to review quickly. Image alt text, page meta descriptions, and article titles are all ideal entry points.

Alt text is a particularly good place to begin. It is one of the most consistently neglected accessibility requirements on the web, and one where AI provides a clear improvement. Research on Wagtail content presented during Global Accessibility Awareness Day found that around 40% of real-world alt text is just a single word.

Cost is also worth addressing for any decision-makers who need reassurance. Because the AI only runs when an editor actively clicks a button, the usage is minimal. Generating 50 image descriptions every day for a full year would cost in the region of $5 to $10 in total AI fees. Energy use per request is similarly low, and trackable if you are using a provider that reports on it.

6. Roll out with clear guardrails

As you expand beyond the initial pilot, it is worth thinking about permissions and governance. Wagtail lets you restrict which user groups can update and configure the AI prompts within the CMS, so the settings are controlled by the people who understand your policy rather than being open to every editor by default.

A common concern at this stage is SEO. The short answer is that AI-assisted content will not hurt your search rankings, provided editors are reviewing suggestions rather than publishing AI output wholesale. Google's position on AI-generated content is that it penalises low-quality, undifferentiated content, not AI assistance. Because Wagtail AI keeps humans in the loop by design, this is not something you need to worry about if you are using it sensibly.

For thinking more broadly about how to design AI-assisted workflows for your content team, the Google People + AI Guidebook is a well-regarded resource worth bookmarking.

Ready to get started?

Visit the Wagtail AI page for a full overview of the package and real-world case studies.

For installation, developers can head straight to the technical documentation.

If you are a Torchbox client and want to discuss getting this set up, speak to your client partner. If you are new to Torchbox and want to explore your options, please get in touch, we'd love to hear from you.