A site usually dies by leaking traffic, not with a sudden cliff. One page starts losing clicks to a fresher competitor, another stops answering the question people now search for, a pricing article still says last quarter’s number, and a bunch of tidy little problems stack up until the whole content library looks tired.
Most teams miss that stage because they only watch the obvious charts. By the time sessions fall hard, the page has already been wrong for weeks or months. AI is useful here because it does not get bored checking the same page for the tenth time. It can sit on top of your existing content, compare it against the live search results, and tell you which pages are quietly going stale before the traffic drop becomes expensive.
Content decay is usually a drift problem
Content decay is not always dramatic. A post can remain indexed, still attract impressions, and still look “fine” in a monthly report while its usefulness slips away. Search intent shifts. Competitors add a better explainer. Product names change. Pricing changes. Internal links break. The page keeps its shape, but the answer inside it no longer fits the query that brought the visitor there.
An AI content decay detector should watch for this failure mode. It should ask whether the page still matches the current landscape around it, not just whether a page exists or ranks.
A practical detector would compare:
- the page’s original topic and wording against today’s top-ranking results
- competitor coverage, including sections your page does not touch
- product changes, plan names, feature sets, and pricing tables
- terminology shifts, where the industry has renamed the thing you wrote about
- broken internal and external links
- fresh questions people now ask that the page never answered
This is a different job from a basic SEO audit. A technical crawl can tell you a link is broken. A content decay system has to decide whether the broken link is part of a larger relevance problem. A page with three dead links and no answer to current search intent is not “a bit old.” It is a candidate for refresh or replacement.
AI should score pages, not just flag them
The useful output is a decay risk score per page, built from multiple signals and weighted by impact, not a giant spreadsheet of warnings.
A decent scoring model would pull from four layers:
- search performance, such as declining clicks, impressions, CTR, and rank movement
- content freshness, such as outdated facts, stale terminology, or missing subtopics
- competitive pressure, such as newer pages that cover the topic more completely
- technical friction, such as broken links, crawl errors, or poor mobile performance
If a page falls out of the top three for a core query, loses a quarter of its organic traffic, and starts missing related long-tail questions that now show up in search console data, it should be pushed up the queue. It should not sit in a “watch list” somewhere.
The score does not need to be mystical. In practice, a 0 to 100 scale is enough.
- 0 to 29, low risk
- 30 to 69, watch closely
- 70 and above, refresh now
The threshold is not the point. The score converts a messy bundle of clues into a decision. Editors do not need a philosophy lecture. They need to know which page is about to go soft and which one can wait another month.
What the system should actually compare
Many AI SEO discussions stop at “compare your content to competitors.” That is too vague to be useful. A good decay detector should compare specific things.
Search results
The page should be checked against the current SERP for its target query and close variants. If the result set has changed, the page may have decayed even if rankings have only slipped a little. New People Also Ask questions, related searches, video results, and different content formats all signal that the query has moved.
If your page still answers the question from two years ago, while the SERP now rewards a more practical or more commercial answer, the page is already behind.
Competitor coverage
The detector should extract the headings, subtopics, examples, and media types used by the top-ranking competitor pages. If all the best results now include a pricing table, a step-by-step workflow, or a downloadable template and your page has none of that, the gap is obvious.
This gap affects search performance even when the article still reads well. Search engines do not reward elegance alone. They reward pages that match the current shape of the query.
Product and pricing changes
Many sites embarrass themselves here. A page about a tool, service, or feature set can drift into nonsense because the product team moved on and nobody updated the article. Old plan names remain in the copy. Prices are wrong. A feature listed as standard is now an add-on. A screenshot shows a UI that no longer exists.
AI is good at spotting these mismatches if you feed it the current site, the CMS content, and a source of truth for the product. It should treat those changes as content risk, not just editorial housekeeping.
Broken links and stale references
Dead links are often the smallest visible symptom of a larger problem. If a page still points to a now-missing guide, an old support doc, or a deprecated external source, the article has lost maintenance discipline. This affects trust.
A detector should not only report the broken link. It should ask whether the broken link is part of a page that needs a full refresh brief.
New questions people now ask
LLMs earn their keep here. They can compare a page against current search intent and spot the questions that no longer appear in the original outline. A page about WordPress speed, for example, may need to answer caching plugin conflicts, Core Web Vitals, and mobile-specific issues that were not central when the post was first written.
If the page does not address the questions users now ask, it is drifting into irrelevance even if it still ranks for the old phrasing.
The refresh brief beats the new article
The worst habit in content teams is the reflex to publish something new whenever a topic gets attention. This often leads to sites with 200 mediocre pages instead of 50 useful ones and 150 thin duplicates.
A better system uses AI to produce a refresh brief for high-risk pages.
A good brief should include:
- the page’s decay score
- the main reasons it scored that way
- the queries it still owns and the ones it has started to lose
- the competitor pages that now outrank it
- the content gaps that need to be filled
- the broken links or outdated facts that need fixing
- the new sections, examples, or data points the page now needs
- a clear recommendation: refresh, merge, redirect, or retire
AI saves time here in a way a human checklist does not. It can turn six messy signals into a usable editorial plan. A writer does not have to trawl through Search Console, a few SERPs, the product docs, and the CMS history just to find out what changed. The machine can hand over the brief and leave the actual writing to the human who understands the business.
A practical workflow that would work
If you were building this properly, the stack would be simple enough to maintain.
Data sources
Connect the detector to:
- Google Search Console for impressions, clicks, CTR, rank, and queries
- GA4 for engagement and conversion behaviour
- your CMS, such as WordPress, for page text, titles, metadata, and publish dates
- a SERP source for live competitor comparisons
- crawl data for links, metadata, and internal linking structure
- product or pricing data, if the site sells anything that changes
A basic process
1. Crawl the site on a fixed schedule, daily or weekly. 2. Pull the latest search and analytics data. 3. Run each page through a relevance check against the current SERP. 4. Compare page content against product, pricing, and terminology sources. 5. Score the page. 6. Generate a brief for pages above the threshold. 7. Push the brief into the content workflow. 8. Measure the refreshed page after publication and feed the result back into the model.
A pilot is the sane way to start. Pick one cluster of pages, maybe your top 50 or top 100 URLs, and see whether the detector surfaces the problems humans already suspect but never get around to fixing. If it misses obvious decay, tune it. If it catches the pages your team had on a nagging list, it is doing its job.
Why refreshes often beat new publishing
There is a stubborn belief that growth means more pages. Sometimes that is true. Often it is just an excuse to make more content because publishing feels productive.
Refreshing existing pages tends to produce better returns for a few plain reasons. The page already has age, backlinks, internal links, and some level of trust. A smart update can lift a page faster than starting from zero. Backlinko has cited a study showing that updating and republishing old posts increased organic traffic by 111.37% on average. Ahrefs has long pushed the same practical idea, which is not surprising because it matches how sites actually perform in the wild.
The economics are hard to ignore. A refresh usually costs less than a brand-new article because the research base already exists. You are not inventing the whole piece again. You are repairing, expanding, and aligning it with the present.
The “200 excellent pages” argument beats the “keep publishing forever” habit for these reasons. A tight library of strong, current pages usually outperforms a bloated archive of content that all kind of says the same thing. The site gets easier to maintain. The signal gets cleaner. Searchers stop landing on dead answers.
What good looks like on a working site
A serious content maintenance system is not flashy. It is boring in the best way.
It tells you which pages are slipping. It tells you why. It tells you what to fix. It tells you what can be left alone.
This is enough to stop content decay from becoming a slow-motion outage.
For most sites, the winning move is not a bigger content machine. It is a better maintenance loop. Keep the strongest pages current. Cull the dead weight. Refresh the pages that still matter. Publish new content when there is a real gap, not because the calendar feels empty.
If the site has 20 pages worth keeping up and 180 pages worth auditing, the answer is not more noise. It is better triage.
