Short version: publishing isn’t a one and done job. Updating old articles is the highest return work sitting in most content libraries right now, because search engines and AI answer engines both reward pages someone actually maintains. We’ve refreshed over 3,000 articles across 50 client sites in the last two years. Average time on site is up 36%, and search traffic has grown double digits year over year.
Most people treat their archive like a filing cabinet. You write the thing, you publish the thing, you move on to the next thing. Then two years later a reader lands on a post recommending software that shut down in 2023, and you look like you stopped paying attention. Because you did.
The Part Nobody Planned For
Your article now has a middle man.
Google’s AI Overviews show up on a huge chunk of searches. Estimates for 2026 range from about a third of queries to over half, depending on whose study you read and what queries they sampled. When an Overview appears, clicks to the regular results drop hard. ChatGPT, Perplexity and Gemini pull answers straight off your pages too.
So here’s the new failure mode. Your page holds its ranking, your traffic quietly falls off a cliff, and your rank tracker tells you everything is fine.
Freshness drives a lot of this. Ahrefs looked at millions of AI citations and found the sources these systems quote run noticeably fresher than what ranks organically. Pages that used to sit untouched for five years and keep earning now have a shelf life.
An old library doesn’t just get less accurate. It gets less quotable. And quotability is half of visibility now.
Why This Matters
Things break. Products get discontinued, companies get acquired, prices change, tools rebrand. Content pointing at dead software annoys readers and dents your credibility. Worse, an AI engine can grab your obsolete recommendation and repeat it to thousands of people who never see your site, never see the correction, and just think you gave bad advice.
Specifics get quoted, mush doesn’t. A number, a date, a named source, a clear definition. Those are the things an answer engine can lift with confidence. Hedged waffle never gets picked up, which is one reason so much AI generated content sounds interchangeable and gets cited by nobody. A lot of refreshing is just turning vague sentences into claims someone can stand behind.
Your writing got better. Go read something you published four years ago. You’ll wince. Tighten it up.
Corrections build trust. Fix mistakes fast and say what you fixed. Readers respect it. A visible “last updated” line does real work.
Search still moves. Revisiting intent, structure, metadata and internal links on an old post still lifts rankings. Now it moves your AI visibility at the same time, which is a nice two for one.
How To Actually Do It
Start with triage. “Update everything” sounds noble and falls apart by week three.
Sort the library first
Four buckets:
Refresh properly. Pages ranking around positions 4 to 20 for queries you care about, plus anything with real money attached that’s lost traffic. This is where the work pays back fastest. A good refresh can move a page onto page one in weeks. Writing a brand new article on the same topic takes months.
Fix the facts only. Long tail pages that still earn a trickle and just need to stop being wrong.
Merge. Three thin posts on the same topic compete with each other and all three lose. Combine them into one strong page, redirect the rest.
Kill. Some pages are dead. Delete or redirect them and stop feeling guilty about it. Not everything deserves a rescue.
Fix wrong things immediately
Doesn’t matter what bucket the page is in. Wrong is wrong, fix it today. And when a reader emails you about an error, credit them publicly when you correct it. Costs you nothing and people love it.
Sort out dead product mentions
Replace with something current, or clearly flag the original as gone and explain what happened to it. “Discontinued in 2024, closest current alternative is X” beats quietly deleting the paragraph. If your post recommends tools, keep it honest about what’s genuinely worth paying for now rather than leaving 2022 recommendations sitting there collecting dust.
Put the answer at the top
This one changes the most posts.
Open with two or three sentences that answer the question in your headline. Then supporting detail. Then background and context last. Most AI citations come from the first chunk of a page, so burying your best material under 400 words of throat clearing costs you.
Write headings that look like questions people ask. “How to migrate a WordPress site to static hosting” works. “A New Way To Think About Hosting” does not.
Then the technical bits
- Title tags and meta descriptions, with the year in there when the query implies currency
- A heading structure that matches what’s actually on the page
- Internal links going both ways, out from the refreshed page and in from newer posts
- Image compression and alt text that says something useful
- Mobile and Core Web Vitals check
- Schema markup. Article schema with author, datePublished and dateModified. FAQ schema where you genuinely answer questions. Update dateModified when you change something real, and leave it alone when you don’t.
Date it honestly
Bumping the date without touching the content is a trick that works for about five minutes. Readers notice, and so does everyone else. Update the date when you’ve done the work, and say what changed.
Track the right things
Time on page, bounce rate and conversions, same as always. But add AI visibility to the list. Go ask ChatGPT, Perplexity and Gemini questions in your category and see whether your pages come back. Watch branded search volume too, because that tends to climb when you’re being cited even if your clicks don’t move.
Build A Rhythm
This works when it’s boring and regular, not when it’s an annual panic.
Monthly, scan for breakage. Dead links, discontinued products, changed pricing.
Quarterly, properly refresh your top 10 to 20 money pages.
Once a year, full audit. Including the merging and deleting decisions everyone avoids.
Seasonal pages break the rule. Update those before the peak hits, not after.
What It’s Worth
Outdated content isn’t just embarrassing, it costs you. Bad information misleads buyers, loses sales and chips away at your reputation. The difference now is that the bad information gets summarized and repeated by machines to people who never visit you and never see the fix.
Your site is a living thing and it needs looking after. Three thousand articles and fifty sites later, I’m more convinced of this than when I started.” A maintained library appreciates. An unmaintained one loses value faster every single year.
What’s your refresh cadence, and are you tracking AI citations yet? Tell me what’s working.