Your traffic report looks fine. The line is flat, nobody is panicking, and the page you published two years ago still brings in visits.
And yet something underneath it is already moving.
That is the frustrating part about content decay. By the time it shows up in a traffic chart, it has usually been happening for months. The content decay signals that come first are quieter: fewer impressions, a slipping average position, queries that used to send visits and now send nothing.
Here's the good news. Those signals are all visible in tools you already have open. You do not need a new stack or a bigger team. You need to know which numbers move first, and what a real warning looks like next to a false alarm.
That is what this guide does. It teaches you to spot content decay while the warning is still upstream, before the drop, so you have time to decide what to do about it. Fixing the page is a separate job for a separate day.
What content decay looks like before traffic drops
Content decay is the slow decline of a page's organic traffic and rankings over time. The word that matters most in that sentence is slow.
Real decay usually unfolds over months, sometimes years. A sharp fall overnight is a different animal. That is more likely a technical failure, a search-system update, or something that changed on the page itself.
So the first thing to look at is not the size of a drop. It is the shape of one.
The most common early shape is a plateau. Traffic looks steady, so nothing feels urgent. Underneath, the page is slipping in the rankings and competitors are catching up. The traffic holds for a while because a few strong queries are carrying it. Then those go too, and the fall arrives all at once.
Two more things worth getting straight before you look at any chart.
Decay is page-level, not site-level. Your whole site can grow while one page quietly loses the queries it was built for. The reverse happens too. A sitewide dip is not proof that every page on it is decaying.
Google visibility and AI visibility are separate observations. A page can slip in Google and vanish from AI answers for completely different reasons, on completely different timelines. A page can still sit on page one and never appear in ChatGPT or Google AI Overviews. Watch both. Do not use one as a stand-in for the other.
Your next step: pick one page that matters to your business and open its last 16 months of data. Not three months. Sixteen. You are looking for shape, and three months is not enough to see one.
Which Search Console metrics move first?
If you only learn one set of content decay metrics, learn these four, in this order.
Impressions
An impression means a link to your site appeared in Google Search, Discover, or News. It is an exposure signal, not a visit.
This is usually the earliest thing to move. A page can lose impressions well before it loses clicks, because it is showing up for fewer queries, showing up less often, or sitting far enough down the results that exposure fades.
Watch the page's overall impression trend, and watch impressions on the queries you actually care about. A broad page-level decline means more than one noisy query having a bad month.
One caution. A fall in impressions on its own can mean lower demand, a different search mix, an algorithm change, or a technical problem. It is a strong reason to look closer. It is not a verdict.
Average position
Search Console defines position as where your link sits in the results, with 1 at the top. The number in your Performance report is an average of the topmost position your site held across every search where it appeared.
That is not the same as a fixed rank. Position varies by query, location, device, and search history, and Google itself warns the metric can mislead without context.
Use it for direction, not precision. The signal you want is a quiet deterioration across the query groups that matter, while the traffic line still looks calm.
Query footprint
This one gets overlooked, and it might be the most honest of the four.
Compare the queries the page ranks for now against the set it used to own. Warning patterns look like this:
- important topic queries have dropped out of the report entirely;
- impressions have drifted from high-intent terms toward peripheral or branded ones;
- the page is visible for fewer variations of the problem it answers;
- a couple of terms still deliver clicks while the wider cluster fades;
- a competitor is picking up visibility for the same intent as your exposure contracts.
In Search Console, filter to the page with the Pages dimension, then look at the Queries dimension. That turns query mix from a memory exercise into something you can measure.
Do not treat every change here as decay. Google changes how it matches language, and your audience, product, or season can legitimately shift the terms. What you are watching for is a sustained loss of the queries the page was built to win.
CTR
CTR is clicks divided by impressions. It moves independently of ranking, and three combinations are especially useful as early warnings:
- Impressions down, CTR down. A classic decay pattern. The page is losing visibility and losing its share of what is left.
- Impressions down, CTR up. The page has probably lost positions or query coverage, but the people who still see it are choosing it more often. Treat this as a visibility warning, not proof the result itself is unappealing.
- Impressions flat, CTR down. The page is still being shown, but something around it changed. A new SERP layout, new result types, a stronger competitor, a shift in intent, or a title that no longer fits. On its own, this is not decay.
There is no universal CTR benchmark worth chasing. Compare the page against its own history, and segment by query, device, and country before you draw a conclusion. A blended CTR can move just because the query mix moved.
Your next step: for one page, put impressions, position, and CTR on screen together for the last 16 months. Three lines, one page. Most of what you need is in that view.
Why stable traffic can hide a declining page
Clicks matter. They are just late.
A click is the outcome of exposure, position, and how appealing your result looks. Which means it moves after all three of those have already moved. By the time clicks fall visibly, the earlier signals have usually been flashing for a while.
This is why stable traffic fools people. A page can hold its clicks because a small number of strong queries are compensating for a long tail that has already gone quiet. The headline number looks healthy. The foundation under it is narrower every month.
Ask a different question of your declining pages. Not "did traffic fall?" but "is this page still visible for everything it used to be visible for?"
Use clicks to size the business impact and to confirm a trend you already suspect. Do not use them as your alarm. An early warning content decline shows up in visibility and query coverage first, and clicks were never built for that job.
Your next step: take your five most valuable pages and count the queries generating meaningful impressions for each one, today versus a year ago. A shrinking count on a flat traffic line is exactly the pattern this section is about.
What engagement and freshness signals reveal
Search Console tells you what happened before the click. Analytics tells you what happened after it.
In GA4, a session counts as engaged when it lasts longer than 10 seconds, includes a key event, or includes at least two page views. Engagement rate is the share of sessions that qualify. Bounce rate is simply the inverse.
Average engagement time measures the time your site was actually in focus in the browser, not just how long a tab sat open.
For a single page, a sustained fall in engagement rate, engagement time, or key events can be an early content-fit warning. It is most interesting when your Search Console numbers still look stable, because that combination hints the page is no longer answering the question well even though it is still being found.
Be careful here. That 10-second rule is a way of classifying sessions, not a standard for how long a good article should hold someone. Engagement can shift because of traffic source, device mix, tracking changes, page speed, layout, or consent behavior. Some pages answer the question fast and short on purpose, and that is a feature.
Compare a page with itself, and with pages like it.
Then read the page. Numbers tell you something changed. A quick read tells you why it might have. Signs worth noticing:
- facts, prices, product details, statistics, examples, or screenshots are visibly out of date;
- the answer no longer matches what the query means today;
- the intro promises something the page delays or never delivers;
- a reader would probably search again to get a better answer somewhere else;
- newer competitors cover the same intent more completely;
- the page is written around search phrases instead of the reader's actual task.
Google's guidance on people-first content frames these as self-assessment questions rather than rules, and it makes one warning explicit: do not change a publish date to make unchanged content look fresh. There is no universal shelf life for an article. Age is context. Staleness is a mismatch between what the page says and what someone needs now.
Your next step: open one suspect page and read it as a first-time visitor with the query in mind. Five minutes. You will feel a mismatch faster than you will measure one.
Which outside changes can look like content decay?
This is the section that saves you from acting on a false alarm.
Not all declining pages are decaying pages. Plenty of things produce the same chart shape without being decay at all. Before you label anything, rule these out.
Seasonality and falling demand. A page can lose impressions because interest fell, not because the page got worse. Google's own traffic-drop guidance suggests looking at the last 16 months and comparing the last three months against the previous period or against the same period a year earlier. For seasonal topics, year over year is the only comparison that tells you anything. Then check the wider trend for the topic. If interest is falling everywhere, your page cannot hold onto demand that no longer exists.
Algorithm and system changes. Google's ranking systems change, and core updates move pages. The tell is pattern and timing. One page weakening slowly inside its topic cluster looks nothing like many unrelated pages moving together on the same date. Google is also clear that a drop tied to an update does not automatically mean something is wrong with your content.
Indexing, crawl, and serving problems. A page cannot earn visibility if Google cannot crawl, index, or serve it. A stray noindex, a server error, a robots problem, a 4xx or 5xx response, or a changed canonical can all look like a decline. If a page left the index, that is an indexing event first. Check the Page indexing report and run URL Inspection before you blame the writing.
Page experience. Core Web Vitals targets are LCP within 2.5 seconds, INP under 200 milliseconds, and CLS under 0.1. Those are user experience thresholds, not decay thresholds. Google says there is no single page experience signal and relevance can still win. A layout change or an intrusive interstitial can hurt satisfaction and coincide with a decline, but describe that as a page experience issue, not stale content.
Lost links. A sustained drop in referring pages or domains is useful corroboration when it lines up with falling impressions and rankings. It is not proof of anything on its own. Links vanish because pages get removed, edited, or redirected, and third-party link tools show discovered links, not a complete census of the web.
Your next step: before you call any page decayed, answer three questions. Is demand still there? Did anything change technically? Did many pages move on the same date? If all three come back clean, your content decay signals are worth trusting.
How to watch AI visibility before citations disappear
Everything above lives in Google's measurement. AI search needs its own layer, because none of it shows up in Search Console.
Google says AI Overviews and AI Mode surface relevant links, and that a page has to be indexed and eligible to appear in ordinary search with a snippet before it can appear there. These features may also use query fan-out, running several related searches across subtopics and sources. There are no separate AI-only eligibility requirements that replace normal SEO fundamentals.
The measures worth trending:
- Mention rate: how often your brand gets named in an answer. A recognition signal.
- Citation rate: how often a page of yours is linked or used as a source. An evidence signal.
- Page-level citation activity: which URLs are being used, how often, and for which prompts.
- Share of voice: how much of the citation space you hold next to competitors.
- Prompt coverage: the share of your recurring question set where you appear at all.
- Competitor substitution: a rival showing up more often for the same tracked question while you show up less.
- Answer history: how the answer itself changes over time, including which source it leans on and how it describes you.
Do not collapse mention rate and citation rate into one number. A brand can be mentioned without a single page citation, and a page can be cited without the brand being the recommendation. They answer different questions, so trend them separately.
The early warning here is not "did citations fall." It is narrower and it arrives sooner. Are you appearing in fewer of your recurring prompts? Is a competitor replacing you in the same ones? Are fewer of your pages being used? Has the answer started describing the topic differently? Those shifts can show up before any aggregate citation number moves.
One measurement caveat worth holding onto. Bing's AI Performance documentation is explicit that its citation data measures visible citation activity, not rankings, authority, or quality, and that it does not represent traffic, clicks, or user behavior. Its grounding queries are grouped, generalized phrases rather than individual user questions, and sparse activity may mean nothing appears at all. So when a citation metric is your only evidence, say "AI visibility weakened." Do not say "traffic fell."
This is measurement work, and doing it by hand across engines gets old fast. DeepSmith's AI Visibility tracks mention rate, citation rate, and share of voice across the engines on your plan, checks your prompt set on a schedule, keeps the full answer history behind every number, and attributes citations down to the page. That gives you the same page-level and prompt-level view for AI answers that Search Console gives you for Google, which is the view this section asks for.
Your next step: write down ten questions your buyers actually ask an AI engine, and check them on a schedule. Ten tracked prompts and a date column beats a perfect framework you never run.
How to read the signals together
Here is the rule that ties all of this together. No single one of these content decay metrics identifies decay on its own. Combinations do.
| What you see | What it may indicate | What not to conclude yet |
|---|---|---|
| Impressions down, position down, important queries lost | Shrinking visibility and a plausible early decay pattern | Not decay until demand, technical health, and update timing are checked |
| Impressions down, CTR up | Seen less often, but still attractive to the people who see it | Do not call the title or the answer weak from this alone |
| Impressions flat, CTR down | SERP context, competition, intent, or presentation changed | Do not read this as a ranking loss |
| Clicks stable, position or query footprint slipping | A slow plateau, with a few strong terms masking wider loss | Stable traffic is not proof of health |
| Engagement down, search visibility stable | Possible traffic-quality or answer-fit deterioration | GA4 definitions and channel mix can create this |
| Lost referring domains plus weaker impressions and rankings | Link support declining alongside visibility | A lost-link count does not prove causation |
| Indexing or crawl anomaly plus abrupt traffic loss | A technical or access event | Do not label an abrupt failure decay |
| Many unrelated pages moving together on one date | A search-system or sitewide event | Do not blame each page's editorial quality |
| Demand falling across the whole market | Seasonality or topic decline | Do not use the drop as evidence the content aged |
| Fewer AI prompt appearances, more competitor citations | Weakening AI visibility for that question set | Not automatically a Google or traffic loss |
Read down the middle column and you have your watchlist. Read the right column and you have your discipline.
A credible early warning has three properties. It is persistent, so it holds across more than one period. It is page-specific, so it is not just your whole site moving. And it is corroborated, so more than one independent signal agrees, after demand, technical health, and update timing have been considered.
Resist the urge to invent a threshold. There is no defensible universal cutoff for impressions, position, CTR, engagement, backlinks, or AI citations, and inventing one ("a 10% ranking loss means decay," "a page is stale after six months") just gives you a confident wrong answer. Some teams do use a lagging trigger for prioritization, like flagging pages down more than 20% year over year in a quarterly review. That is a practical way to sort a backlog. It is not a definition of decay, and it arrives long after the signals in this guide.
What actually works is boring and repeatable. Same pages. Same date windows. Same query and prompt sets. Every quarter.
Start with one page this week
You do not need a monitoring system to begin. You need one page and one honest look.
Pick the page that would hurt most to lose. Pull 16 months of impressions, position, and query footprint. Check whether the query set has narrowed. Check whether AI answers still mention or cite you for the questions that matter. Then ask the three ruling-out questions before you conclude anything.
That is the whole practice. Catching early warning content decline is not about watching more numbers. It is about watching the right ones early, and being honest about what they can and cannot prove.
If tracking your AI visibility by hand is the part that keeps slipping, DeepSmith runs your prompt set on a schedule and keeps the citation history next to your content, so the AI half of this stops depending on someone remembering to check. Start a free trial and watch one page for a month.
One page. One month. That is a real start.
