A content health check looks at live content – a website, a help centre, an intranet or the knowledge base behind an assistant – and rates it against criteria such as accuracy, clarity, accessibility and usefulness. Each round records ratings alongside the findings that explain them. Because the criteria and usually the sample stay the same, one round can be compared with the last, so a team sees whether things are getting better or worse rather than a single snapshot.
The name borrows from medicine. Like a routine check-up, it’s preventive and comparative: it looks for early signs of trouble against a known baseline instead of waiting for a crisis. Like screening, it’s better at showing where to look more closely than at explaining causes. The practice belongs to the same family as the qualitative content audit, the heuristic review and the content scorecard. What sets it apart is repetition with consistent criteria.
Why it matters for content
Content rarely fails all at once. A price changes, a linked form moves or a policy is revised, and the page explaining it stays as it was. A page may pass a spell check yet describe a service that no longer exists. A health check brings different kinds of evidence together so a team can decide whether to correct, investigate, merge or retire something, instead of treating every problem as a rewrite.
Checks can work page by page or follow the tasks people come to do, such as applying for a grant or cancelling a subscription. Following a task exposes problems that sit between pages, like a missing step or a dead end, which a page-by-page review can miss. Repeated results help teams maintain content, prioritise and test whether improvements last.
How AI changes the picture
Recurring checks used to be limited by the hours available for reading. Language models can apply written criteria across thousands of pages, which makes it practical to check a whole estate regularly rather than a sample once a year. They can also compare content with a source of truth, matching support answers against a revised procedure and highlighting the passages that now disagree. A useful flag shows the evidence behind it, so a reviewer can see why a page scored badly and judge whether the assessment holds.
What needs checking is changing too. Assistants and AI search draw on help articles, policies and knowledge bases to answer questions, so a weak source may affect many answers that nobody reads individually. Health checks increasingly need to cover that source material and, where possible, samples of the answers built from it.
An automated checker is an instrument, and instruments need calibrating. Test it against examples where the right answer is known, and be clear about criteria it can’t assess from the text alone: whether content is useful may depend on research, service data or what people do next. Automating a judgement doesn’t establish that the criterion or the result is sound.
Questions people ask
- How is a health check different from a content audit?
A health check is a kind of audit, focused on quality and repeated over time. A full content audit often starts with an inventory of everything that exists and ends with decisions about what to keep, merge or remove, typically before a redesign or migration. A health check keeps watching how well it’s doing.
- Which criteria should we use?
Choose criteria that reflect what your content is for and that reviewers can apply consistently. Most frameworks cover accuracy, usefulness, clarity, accessibility, structure and currency, with additions for voice, brand or regulation where they matter. Published frameworks and scorecard templates save starting from scratch. Whichever you use, write down what each rating means, so that two reviewers looking at the same page would reach a similar score.
- Does a high score prove that content works?
It shows that content meets the criteria as assessed. Whether it helps people still needs evidence from use, such as task completion, search behaviour, support contacts or research with readers. A health check is diagnostic; testing with real people is how you confirm the diagnosis.
- Who should run it?
Whoever is responsible for keeping live content useful, with subject experts checking accuracy. A small team might review its ten most important pages each quarter; a large publisher might combine automated assessment across the estate with expert review of a risk-based sample. A second reviewer, or a periodic comparison of scores between colleagues, helps keep individual bias in check.
Learn more
- How to evaluate content using a content scorecard – Content Strategy Inc.’s method for scoring content by walking through user scenarios, with the caution that expert review is diagnostic and still needs testing with real customers.
- Taming the content monster – Samantha Campbell’s account of scoring a 282-page Atlassian content hub against criteria adapted from Nielsen’s usability heuristics, which found that many of the most visited pages scored medium to low.
- Content Health Check – Contentious’s tool for recurring, AI-assisted checks of live content against the 15 criteria of its Content Health Framework.