Finding content that needs fixing
We are working with the Defra content team to find out-of-date pages, broken links and wording that has drifted.
This is an experiment. We are testing whether AI can help.
The problem
Defra publishes a lot of content, and it keeps growing. Nobody has a clear view of what is out of date, duplicated, or no longer read.
That makes it hard to know what to fix first. It also makes it harder to trust that what is published is still right.
Two related problems come from the same place. Links from GOV.UK service pages to council websites break over time, and there are thousands to check. The style guide also misses terms people use, so wording drifts apart.
What we are building
We are running three tests, one at a time. The first is the priority, and the other two follow if it goes well.
1. Finding content to retire
The check looks for pages that are out of date, duplicated, barely read, or replaced by something newer. It also flags anything that no longer matches policy.
2. Fixing broken links
The AI searches the council website for the page that replaced the broken one. It compares the two and suggests the most likely match. People use these links to report a pest problem or check a bin collection day.
3. Spotting style problems
The check compares content with Defra and GOV.UK style guidance, then suggests plainer wording where language has drifted.
All three give a recommendation with the reasoning attached. A content designer can see why it was made.
What it does not cover
Nothing is changed, removed or unpublished by the tool. Every recommendation goes to a content specialist, who accepts it, rejects it or changes it.
It does not write content.
This is a proof of concept. That means a small test to see whether the idea works, before anyone builds it.
Where it has got to
We have not started building yet. We are gathering what the test needs: the broken link data, the style guides and earlier content audits.
Finding content to retire is the first test. We stay on it until we know whether it works.
How we will know it works
We test the recommendations against examples where we already know the right answer. We measure:
- how often the recommendations are right, and how often they are wrong
- whether a content designer can see why each one was made
- how much time it saves
- whether the same approach would work for other content teams
What comes later
If the tests work, we turn what we learn into patterns other teams can pick up. That includes content teams in other organisations across the Defra group.
Follow this project
Get in touch if your team looks after a large amount of published content.