Ask an AI assistant what to do about a persistent headache and you'll get a detailed answer in seconds. You probably won't click through to the article it came from. You likely won't know which experts contributed to it. And you almost certainly won't know how current it is. That borrowed confidence is risky anywhere. In health, it's dangerous.
This is the new reality for health content. Google's AI Overviews, ChatGPT, Perplexity, and a growing list of assistants now sit between patients and the content health systems publish. They read the page, digest it, and hand back a summary. For content teams, that changes the question. It used to be "Is this article good and accurate?" Now it's "Is this article still accurate — can anyone tell — and is anyone actually reading the page?"
Stale content is worse than no content
A person reading an old article has context. They can see the publish date. They can see the page design from three redesigns ago. They notice the guidance mentions a program that no longer exists. All of that tells them to be skeptical. It isn't ideal, but it's a layer of protection: our internet BS-meter has been trained for years to spot content we should ignore.
An AI summary strips that context away. Answer engines pull from whatever is live and indexed, and the answer they give rarely carries the publish date, the framing, or the page around it. A four-year-old article about screening guidelines can come back as two confident sentences with nothing to signal its age. Sources are often cited. They're rarely clicked.
So an outdated article isn't just a missed opportunity. It's a liability. In healthcare, where guidance changes and accuracy carries real stakes, that matters more than almost anywhere else.
Governance is now part of the user experience
Content governance has long been treated as busywork. Someone runs an audit once a year, flags a spreadsheet of stale pages, and the list slowly gets worked down between higher-priority projects.
That model made sense when the main risk was a mildly embarrassing page. It doesn't hold up when everything you publish becomes raw material for answers you don't control. Governance now decides what AI can draw from. Update or unpublish a page, and once the index catches up, the old version stops showing up in answers.
That puts governance inside the UX conversation, not after it. The job for UX and content strategy teams is to build freshness into the system so it happens by default, instead of depending on someone remembering to check. The key is treating content authors as users in their own right. Design the workflow for them, and it sticks.
What it looks like in practice
We recently worked through this on a health system's newsroom redesign: a content hub for patient stories, health guidance, research news, and media. Four patterns stood out.
1. Every article gets an expiration date
Articles and media are built with a scheduled unpublish date. When that date arrives, the system takes the content down automatically. No one has to remember.
On its own, automatic removal would be a blunt tool, so it's paired with a review window. A CMS view lists everything set to expire in the next 90 days, recalculated every time someone opens it. On the 15th of each month, the content team gets an email linking straight to that view, and every item links directly to its article so an editor can update it, extend it, or let it go.
The result is decay managed as a continuous process instead of an annual cleanup. And it solves the real problem: stale content never lives long enough to become an AI answer.
2. Freshness is visible on the page
When an article is republished after an update, it shows an "Updated" date instead of the original publish date. It's a small detail, but it tells readers the content is maintained, not abandoned. It also gives crawlers a clear, consistent freshness signal.
3. Expert review is built into the design
Articles include an "Expert reviewed" card with the reviewing clinician's photo, name, and title, linked to their profile. Health content is judged heavily on experience, expertise, authoritativeness, and trust (E-E-A-T, the framework in Google's search quality rater guidelines), and this puts the strongest trust signal where both readers and machines can find it.
The detail worth copying: if no one has reviewed the article, the card doesn't appear. No empty placeholder, no generic "medically reviewed" badge. The signal only shows up when it's real, which is exactly what makes it worth trusting. This isn't about dressing up unreliable content. It's about a framework that produces and maintains trustworthy content by design.
4. Taxonomy works for people and machines
Content is organized in two visible layers. Primary content types tell readers what kind of piece they're reading and power curated landing pages. A second layer of patient-facing health topics, such as heart health, cancer care, and women's health, adds depth for filtering and search.
Underneath sit tags most readers never see: audience, content intent, service line, location, and more. They power search and related content across the site, and they give structure to what would otherwise be a flat list of articles. They also make each article's purpose explicit to any machine reading it. That matters, because structure is what helps content be understood out of context, and out of context is exactly how AI encounters it.
The tradeoff: simple beats clever
Not every good idea made it in, and that's part of the lesson.
We considered a new publishing status, something like "Needs update," that content would move into automatically before it expired, so editors could filter for it right in the CMS. It was a tidy idea that would have made "needs attention" a first-class state.
We didn't build it. The CMS already had a scheduled-content view that could sort by unpublish date, and one filtered view on top of it met the need with far less complexity. The same thinking shaped the email: a fixed send date on the 15th, no conditional logic for weekends, and a link to the live view instead of a list of every item.
Governance systems fail when they're too heavy for the team that has to run them. The version that gets used every month beats the elegant version that gets worked around.
What governance can't fix
It's worth being honest about the limits. Answer engines that read live pages will stop citing content once it's gone. But models trained on older snapshots of the web may have already absorbed what an outdated article said, and nothing you unpublish today reaches back into them.
No content team can fully solve that. What you can control is what gets found from here on: make sure the current, correct, expert-reviewed version is the one future crawls and training runs encounter. That's one more reason to start now rather than after the next audit.
Where to start
If you run content for a health system, a few questions are worth asking this quarter:
- Does every piece of content have an end date? If not, stale content is live by default.
- Does someone see what's expiring before it expires? A review window turns removal into a decision.
- Can a reader tell the content is current? Show an updated date when it applies.
- Are your trust signals real? Show expert review only when it happened.
- Is your content structured for machines, too? Taxonomy and tagging are how content stays understandable out of context.
AI isn't going to stop summarizing health content. The teams that do well will be the ones that treat freshness as part of the design, not a cleanup job.