How We Manage Content and AI Search

We were on a pitch call recently and the first question the potential client asked us was around AI and content.

The context was that they’d seen an increase in the number of users searching for information in AI tools, such as ChatGPT, Claude and Gemini, and then not visiting the organisation’s websites.

They were concerned that they weren’t going to get the right information to help them with their task, and wanted to understand how we would approach this.

It was quite a big question with under 10 minutes left on the clock for the pitch, however, we were able to give a brief perspective from a content and a developer outlook.

While we were able to give them an outline of both content design and technical approaches, we felt like this was a space that needed more discussion and exploration.

We are also aware that what we are writing here in May 2026 is likely to change quickly as the world of AI develops. As a result we will be working to keep this insights post updated.

What do we know?

According to McKinsey, 37% to 50% of consumers now use AI tools for searching.

We know that it has created a lot of discussion in the content design community and that people are trying to figure out how to adapt and manage what is happening.

There’s also been lots of posts online about ‘garbage in, and garbage out’.

We’ve also heard stories of people only using information from an AI search, without further research, that has had a negative impact on their life. This aspect is causing a lot of concern in the content community, and rightly so.

We also know that organisations are openly blogging about how they are taking a content design approach to try and solve this challenge. We are for this approach, however, we know that teams are being expected to do more for less. We know that teams are being cut. We know that teams managing BAU is a big task. So unless teams can be given the time and space to focus on content management, deleting content that doesn’t meet user needs, or is outdated, then the task is going to continue to expand.

What don’t we know?

I spoke with Jamie Turnbull, an SEO expert who I have worked with for a long time. He said there is still uncertainty on how AI tools search and decide on what information from websites they choose to list.

We also don’t know if the trend will continue with more users searching inside AI instead of through websites. However, I think we can assume that this is very likely to happen.

How can we aim to work within the new perimeters?

Back to basics

The most straightforward thing to do is to get rid of content that doesn’t serve a user or business need. And get rid of it if it’s out of date.

Content audits can be a good starting point, but you really need to think about the criteria. This could be based on age of content, views, who the owner is, or if there’s duplicate versions.

This can take time, especially if you are in an organisation with thousands or millions of pieces of content.

You could even consider using a ROT analysis approach: redundant, obsolete and trivial.

Then put a content maintenance plan in place so you have guidance on how often content will be reviewed.

And with the content you have left, think about the structure. Having a clear structure in place helps users, both human and AI, find the right information.

Embrace the tech

There are 3 areas that we have been exploring, both internally and through conversations with other experts, including Jamie.

Working in markdown, or turning existing content into markdown, is a good starting point to give you:

a clean separation between content and presentation,
straightforward version control through tools such as Git,
consistent formatting when content is passed between humans and AI tools,
easier reuse of the same source across different platforms.

Automating content audits through using AI tools or Open Source toolkits. This moves the auditing process, which can be time consuming and expensive, from an occasional task to continuous oversight.

This is echoed by Affinity’s technical director, James Macintyre, who has been working on automated tools that can test if the content meets Government Digital Services (GDS), accessibility and plain English standards, and flag if there are areas that need to be addressed.

And the third aspect is structured data. As people increasingly find information through AI Overviews, ChatGPT, Perplexity and voice assistants, the way content is labelled has direct consequences for how machines interpret a page.

Schema markup, expressed in JSON-LD, is the standard way of making content machine-readable. Until recently, generating this markup required either developer time or careful manual work. AI tools now generate JSON-LD from existing content, and platforms such as Webflow are building this directly into their authoring flow.

Resources

We’ve put together a list of resources that we’ve been using to help further our thinking on this topic.

User Behaviour AI Learning Group

The group has monthly online meetings. The sessions feature guest speakers who talk through their experiences of working with AI.

There is also a Slack channel on Content Club.

Design in Government Blog

‘When AI answers the questions, what happens to the user journey?’

A great blog post looking at the rise in AI search.

Department for Business and Trade Digital Blog

‘How we’re preventing AI misinformation at DBT’

A blog post focusing on the approaches the DBT team is using counter AI misinformation.

These resources are correct as of May 2026. We will check them on a monthly basis and add in new research, or amend existing information, as needed.