A library has two ways of sending people to a book. The catalog is one: you look up a subject, you get a call number, you walk to the shelf. The reference desk is the other: you ask the librarian a question and she answers it out loud, mentioning whichever book she trusts on the subject. Those are different systems with different rules, and a book can do well in one and go unmentioned in the other for years.
Your blog posts live in both systems now. Search is the catalog. AI answers are the reference desk.
Content decay is what happens when a page slides out of one, the other, or both, and it goes unnoticed because nothing announces it. There’s no error, no broken page, no alert. The numbers simply get smaller each month until one day the post brings in nobody at all.
Content decay is unfair to every writer who did the work once and did it well. You spend a week on a post, it pulls in readers for two years, and then the platforms reshuffle and nobody sends you a note. I don’t blame the writer for missing it. The reports we all read were built to celebrate growth, and a slow bleed never shows up as anything worth reporting.
What is content decay?
Content decay is the slow loss of traffic to a page that used to perform, caused by the world moving while the page stands still. Competitors publish something better. The question people ask changes shape. Your examples age out. A search engine adjusts what it rewards. Each of those is small. Together, over eighteen months, they take a page from steady traffic to nothing.
The word decay describes the shape of the loss.
A broken page fails all at once and you find out fast. A decaying page keeps working, keeps loading, keeps looking fine, and loses ten percent a quarter. By the time it registers as a problem, you’ve lost most of a year of readers you already earned.
I think this is the cruelest kind of loss an archive can suffer, because it hits the posts that worked. The weak ones never had readers to lose. The strong ones did, and they bleed them while the author is busy writing the next piece, trusting the old one to keep pulling its weight.
The first clock: losing ground in the catalog
The familiar clock is search position. Google Search Console is where you read it, and the comparison that matters is the last six months against the six before that. Sort by page. Look for the ones where clicks and impressions both fall together. That pairing is the signal. Falling impressions mean Google shows you less often, and that usually starts before the clicks drop.
The two numbers apart tell you different stories. Impressions holding steady while clicks fall means the page still ranks and stopped getting the click. The title and description are the problem, and the fix takes ten minutes. Impressions falling means you’re losing position. That’s a content problem and takes longer. Chasing rank on a page that already sits on page one with a collapsing click rate is wasted effort. That page needs a better promise in its title.
One cause of decay is self-inflicted. Check it first. Two of your own posts competing for the same query will suppress each other, and the older one usually loses. I found several of these on my own site, and the fix is consolidation: merge the weaker page into the stronger one and redirect it.
The second clock: disappearing from the reference desk
The newer clock is the one almost nobody tracks.
A page can hold its search ranking while AI answers stop citing it, and no report you currently read will tell you it happened. When somebody asks ChatGPT or Perplexity or Google’s AI summary a question in your subject, the system pulls a handful of sources into the answer. If you’re not among them, you’re invisible to that path. And that path is taking a growing share of the questions people used to type into a search box.
What the system pulls depends less on your ranking than on the shape of your writing.
Retrieval works on passages. A model pulls a section, sometimes a paragraph, and uses it to answer. A section that depends on three paragraphs above it to make sense is a section nobody can lift. So the same page can be excellent for a human reader start to finish and useless to a machine reading it in fragments.
I’ve written about this in more depth in writing to get found by AI instead of just ranked, and about why the model is less settled than the confident coverage suggests in the cracks in AI search.
Both come down to this. Answer the question in the first sentence, then explain. Write every section so it survives on its own.
How do you find decaying blog posts on your site?
Pull twelve months of Search Console data, split it in half, and compare.
Export both halves and put them side by side in a spreadsheet with a column for the change in clicks and a column for the change in impressions. Sort by the click change. The pages at the bottom of that sort are your audit list, and it’ll be shorter than you expect.
Then apply a second sort that’s nothing to do with the numbers: which of these pages do you need? A recipe post that lost forty percent may not matter. The page that explains what you do, that answers the question your buyers ask before they call you, matters enormously even if it lost less. Decay on a page that carries commercial weight is a different problem from decay on a page that carries none.
The AI half of the audit has no tool yet, so it’s manual and slower.
Take your ten most important questions, ask them of two or three AI systems, and write down whom they cite. Do it again next quarter. That’s a crude instrument, and it’s more than most people have. I’d take a crude answer to whether the reference desk knows I exist over a polished dashboard that can’t tell me at all.
What my own audit turned up
My site has just over three thousand published posts and pages. That’s enough content to have decay everywhere, and enough to make the pattern visible. Two findings surprised me.
The first was concentration. One long guide carries close to eighty percent of the commercial citations my site earns in AI answers. The service pages, the ones that exist to sell the thing I do, appear in none of them. That gap is structural. The guide answers questions, and the service pages announce offerings. Answer engines quote the first kind and ignore the second.
The second was mechanical. Going through the site I found long unbroken blocks of text where a single paragraph ran for two hundred words and covered three ideas. Human readers tolerate that. Passage retrieval doesn’t, because there’s no clean unit to lift. I ended up splitting several thousand of these into paragraphs that each carry one idea. That changed nothing about the writing and a great deal about how machines read the pages in pieces.
I also pulled the site’s 404 log and found several hundred distinct broken URLs, some from posts I’d moved years ago. Every one of those was a page somebody wanted and didn’t get. Fixing them recovered readers I’d already earned and then lost.
What does an answer-ready page look like?
The page below is what all of this produces. Every part is doing a job for one clock or the other.
Read it as a checklist against your own decaying posts.
The title phrased as a question somebody would type. The answer in the opening sentence instead of four paragraphs down. Subheadings that stand alone, with no pronouns pointing back up the page. A visible update date backed by an actual update. Links to the siblings that keep a cluster alive. An FAQ that matches the questions people ask, with schema that agrees with what the page visibly says. A named author with credentials a reader can confirm somewhere other than your own site.
That last one matters more than it used to. Search engines and answer engines both try to work out whether a source is a real entity with a track record, and corroboration that exists only on your own domain is the easiest kind to fake. I go further into this in why AEO makes websites stronger.
How do you fix a decaying page instead of replacing it?
A refresh beats a rewrite from zero. The reason is arithmetic. An old page carries links, history, and whatever authority it accumulated.
A new page carries none of that and starts at the back of the line. Fixing three old posts in an afternoon usually returns more than writing one new one that may never rank.
I can’t abide the advice to keep publishing and never look back. It sounds productive, and new content feels like progress. Meanwhile the writer’s best work rots in the archive, and the cost lands on the one person who can least afford to throw away readers they already earned.
Open the decaying page and read it as a stranger would. Ask whether the advice still holds. Kill the dead links, the aged screenshots, the examples that date the piece. Then add one thing the original never had, because a refresh that only updates a date is a lie. Move the answer up. Break the long paragraphs. Add the FAQ if it’s none.
Then check the page’s technical footing, since a page can decay because it got slow while your competitors got fast. My guide to the three numbers that measure how your website feels covers what to look at, and my broader take on blog SEO for authors covers the fundamentals underneath all of this.
What should you do with a post that is not worth saving?
Some posts deserve retirement. Pretending otherwise leaves you with an archive nobody wants to maintain. The test is whether the page has a job. If it covers a topic you no longer address, ranks for nothing, earns no links, and would embarrass you if a client read it, it’s no job.
How you retire it matters. Merge anything useful into a stronger page and redirect the old URL there, so the links and history transfer instead of evaporating. Delete without a redirect only when nothing points at the page and nothing ever will. A deleted URL with inbound links is a reader hitting a wall.
Writers resist retiring posts, and I think pride keeps a lot of them sitting on pages that embarrass them. A dead page with your name on it does more harm than no page at all. It tells a stranger you stopped caring, and that stranger may be the reader or client you most wanted.
How often should you audit for content decay?
Once a quarter, block an afternoon. Pull the Search Console comparison, list the pages that are sliding, cross-reference against which pages carry commercial weight, and fix three or four. Run your ten questions past the AI systems and note whom they cite. That’s the routine, and it takes less time than writing one new post.
The habit counts for more than the thoroughness. A blog you look at four times a year doesn’t accumulate the kind of rot that takes a month to dig out. My site went years without this discipline and the cleanup cost me weeks. Do it small and often.
The catalog and the reference desk
Both systems are asking a version of the same question about your page: is this the best answer available right now? The catalog answers it with position, the reference desk answers it by deciding whether to mention you at all, and neither will tell you when the answer changes. That’s what makes decay a discipline instead of a task. You check because nothing will tell you.
Start with your ten most important pages and one afternoon of Search Console data.
If the site itself needs the work and you’d rather hand it to somebody who has done this for decades, that’s what I do. Everything else in this series lives in the Technology of Writing hub. The archive you already wrote is the cheapest traffic you’ll ever get back, and it’s sitting there waiting for somebody to look at it.
Too much good writing gets written off as a failure when all it needed was an afternoon of attention. The platforms won’t warn you and the dashboards won’t flag it, so the job falls to you, and a writer who skips it is handing readers to whoever bothered to update last.
