Content Decay: How to Tell a Losing Page From a Broken Report
Every content site has this conversation. A page that brought 900 sessions a month now brings 300. Nobody edited the page, nobody edited the tags, and the monthly report shows a hole where a reliable performer used to be.
The reflex is to rewrite the page. That is sometimes correct. But before anyone spends a day on it, a duller question has to be answered. Is the page losing traffic, or is the measurement losing accuracy? The second answer turns up more often than teams expect, because default settings and history limits decide what you are allowed to see.
Why pages lose traffic in the first place
Four causes cover almost every case.
Intent drift. The query still exists, but the people searching it want something different from what the page gives them. A guide to a pricing question turns into theory once the pricing model changes and nobody updates it.
A different result page. A competitor publishes a better answer, an AI overview takes the top of the page, or Google starts preferring a different content type for the query. You changed nothing, your average position may even improve, and your clicks still fall.
Seasonality. The page is fine. The month is not. Comparing November against October tells you about the calendar, not about the page.
Structural damage. A navigation change, a lost internal link, a redirect chain, a template update that dropped the heading, or a canonical pointing somewhere unexpected.
Only the first of those four is a content problem. The rest are diagnosis problems, and none of them shows up as a single month-over-month number.
Rule out the measurement fault first
Work through these in order. Most false alarms die in the first two.
1. Did collection change? A consent banner update, a new CMP, a tag reorganisation, or a container change that altered which tags fire will all move page-level numbers. If the drop starts on the day of a release, stop here. On a UK or EEA site with a high denial rate, coverage moves with visitor consent behaviour, so a CMP change can shift page numbers while real readership is flat.
2. Did reporting change? A new data filter, a hostname filter, an internal traffic rule, or an added report filter all remove rows. GA4 data filters apply from the moment of creation and never to historical data, so a filter added in September looks exactly like traffic disappearing in September.
3. Is the page simply below a threshold? Low-volume pages lose rows to data thresholds, and demographic breakdowns lose them faster. A page with 200 sessions a month is a poor candidate for a month-over-month argument.
4. Is the comparison window even queryable? This is the one that catches experienced people.
5. Are you reading unsettled data? GA4 needs roughly 24 to 48 hours to finish processing, and attribution credit keeps moving for up to 12 days. A comparison ending yesterday is a comparison against a moving target.
The retention setting decides which comparisons exist
GA4 standard properties keep user-level data for 2 or 14 months, and that period also governs key events data. Event-level data can go to 14 months, with 26, 38 and 50 months reserved for 360 properties. Age, gender and interest data are always cut to two months regardless of what you select.
Two months is the default on a new property. That means a property nobody has configured cannot answer a year-over-year question at all, and a 90-day window may not be fully available either. The Explorations documentation states the default plainly, and the setting lives under Admin, then Data collection and modification, then Data Retention. Raising it covers data already collected, not what has expired.
Search Console is more generous and completely inflexible. It keeps 16 months, and because the GA4 Search Console reports read from it, those reports cap at 16 months as well.
The honest position: year-over-year analysis is possible in Search Console and often impossible in GA4 on a standard property. If a decay review depends on a 13-month comparison, build it on the search data and use GA4 for the engagement and conversion half of the story.
Three report surfaces, three different questions
GA4 has three places where page-level decline shows up, and they answer different questions.
Engagement, then Pages and screens. This covers every page view, whether it was the first page of a session or the tenth. The dimension is Page path and screen class, where the page path is the part of the URL after the domain. Filter on that dimension to isolate one page, and use a path exploration to see what people did before or after it.
Engagement, then Landing page. This covers the first page of a session only, and it is session scoped. A visitor who arrives on page A in one session and page B later is counted in both rows. Use it when the question is about entry rather than readership.
Reports, then Search Console, then Google organic search traffic. This one is unpublished by default, which is why so many properties never use it. Find the Search Console collection under Library in the left navigation and publish it. It then shows landing pages with organic clicks, impressions, click-through rate and average position from Search Console, next to GA4 active users and engaged sessions.
That third report is the closest thing either product has to a purpose-built decay view, and it ships switched off.
Read the search numbers correctly
Search Console reports four metrics and they tell different stories. Average position is the average ranking of your topmost result per query. In the chart it is calculated across the whole site. In the table it is calculated for the specific URL in that row.
Click-through rate is clicks divided by impressions, and GA4 returns it as a fraction. A value of 0.0588 means about 5.88%. Read it next to position and never on its own, because a page at position five with a two percent rate has a title problem, and one at position two with the same rate has a different problem.
The pattern worth hunting is impressions holding steady while clicks fall. That usually points at the result page changing, not the page getting worse, and new copy will not fix it.
Query-level history beyond 16 months means exporting to BigQuery, which needs a Google Cloud project with billing enabled. There is a free usage level, and usage above it is charged. Data from before the export is configured is not included. For most small teams the 16-month window plus the GA4 side is enough, and adding a billed cloud project to chase a ninth quarter is a poor trade.
Score pages on value, not sessions
Ranking pages by sessions produces a to-do list of your biggest pages, not your most valuable ones. A page with 400 sessions and a six percent conversion rate earns more than a page with 4,000 sessions and a 0.4% rate, and only one of them deserves the work.
Build one table with page path, sessions, engaged sessions, key events and revenue where it exists, for two comparable periods. Comparable means the same range a year earlier, or the previous period plus a seasonality check. Sort by lost key events or lost revenue, never by lost sessions.
An exploration gets you there: a free-form table with page path and screen class as the row dimension, sessions, engaged sessions and key events as metrics, and two date ranges in separate tabs. Explorations hold up to 20 dimensions and 20 metrics, and any query over 10 million events may be sampled, so keep the window tight.
Then pick exactly one action per page.
Refresh. The intent is still there and the page is thin, dated or incomplete. This is the common case.
Consolidate. Two or three pages chase the same query. Merge the strongest into one and point the others at it with a permanent server-side redirect. Google uses a 301 or 308 as a signal that the target should be canonical. A 302, 303 or 307 does not, and the target may still be indexed, so a temporary redirect is the wrong tool for a permanent merge.
Redirect elsewhere. The page has no unique value and a close relative does. The same redirect rules apply.
Leave it alone. Seasonality, a documented result page change, or a strategic page that converts well at low volume. Write down the reason, or the same page returns to the list next quarter.
Remove it. The page is wrong, obsolete, or diluting the focus of the site. Removing or updating the content is the durable fix. The Search Console removals tool only hides a URL for about six months, so it is a pause button rather than a decision.
Refresh means changing the answer, not the date
Google's own guidance is blunt about the shortcut. Its people-first content page lists, among the warning signs, the question of whether you are changing the date of pages to make them seem fresh when the content has not substantially changed. A new date on unchanged words claims the substance moved. It did not.
A real refresh changes what the page knows. Add the detail the current intent needs and the old version lacks. Replace examples that are now wrong, correct factual errors, and add first-hand evidence: a real screenshot, a real number, or a decision you made and the reason behind it. Tighten the structure so the answer arrives before the background. Rewrite the title and description if the click-through rate says they are the problem.
Before and after, run the self-assessment Google publishes. Does the content provide original information, reporting or analysis? Does it describe the topic substantially and completely? Would you bookmark it? Does it hold up against the other pages ranking for the same query? That list beats any word count target, and Google states in the same place that it has no preferred word count.
Measure the result without fooling yourself
Three habits keep a refresh review honest.
Annotate the change in GA4 with the date and what moved, so the next person reading the chart is not guessing at what happened that week.
Keep a control set. Pick ten pages you did not touch and compare them over the same window. If the whole site fell 15% and the refreshed pages fell 5%, the work succeeded even though both numbers are red.
Wait long enough to measure. GA4 needs 24 to 48 hours to settle and attribution credit can shift for up to 12 days. Search Console needs a recrawl before it reflects new copy. Judging a rewrite on day three produces confident nonsense.
Judge on engaged sessions, key events and revenue per session. A refresh that trades 400 sessions for 250 sessions at three times the conversion rate is a success that a sessions chart will report as a disaster.
A cadence that survives a busy quarter
Monthly, look at the top 20 pages by revenue or key events and flag anything that moved more than 20%. Quarterly, run the full comparison, score the movers and give one action per page to one named owner. Annually, look at the pages that have not changed in two years and decide whether they still deserve to exist.
The step most teams skip is the log. A page refreshed in March that declined again in June is a different problem from a page nobody has touched since 2023, and you can only tell them apart if the decisions were written down.
If your traffic reports disagree with your traffic, book a free tracking audit and we will separate the measurement faults from the real decline, page by page.