The pages to refresh first are those that lost the most clicks against the rest of your site over the same days, among pages with enough impressions to judge. Pages picked because they fell tend to rebound anyway, so hold some back if you want to measure the refresh.
- Penlid's decay check judges a page's last 28 days against the 84 before them, after subtracting the site-wide move over the same window (Penlid, methodology page, read 30 October 2026).
- Pages below 3,000 impressions in the trailing 90 days are marked unknown rather than classified (Penlid, methodology page, read 30 October 2026).
- In Penlid's simulations, unchanged pages picked for a bad last 28 days were still called "improved" 5.50% to 9.20% of the time, against at most 3.10% for pages not picked that way (Penlid, methodology page, read 30 October 2026).
- Google's helpful-content guidance asks whether you are changing the date of pages to make them seem fresh when the content has not substantially changed (Google, Search Central, read 30 October 2026).
Which pages are candidates for a refresh?
Start with pages whose clicks fell against the rest of your site over the same calendar days, among pages with enough impressions to judge. Penlid's content decay check works that way: it judges each page's last 28 days against the 84 before them and subtracts the site-wide move, so a January dip that hit everything reads as healthy
. For the subtraction by hand, see how to tell content decay from a site-wide dip.
A page below 3,000 impressions in the trailing 90 days is not classified: the methodology marks it unknown and the tool answers "can't tell", because beneath that volume daily noise swamps the signal. Keep those pages off the ranked list; if one needs work for another reason, such as visibly outdated facts, that is a separate decision (our suggestion).
Decay is not the only source of candidates. Google's help page on common Performance report tasks names two more: low-CTR pages, for which it suggests updating the title and description to better represent the content, or adjusting the content to align better with the queries that show the page, and pages with a significant change, found by sorting on the Difference column. Striking-distance queries make a third list, and pages not worth saving are a pruning question.
Which candidate should come first?
Put first the candidate with the most clicks at stake: the clicks it would have had over the last 28 days had it moved with the rest of your site, minus the clicks it got. That ordering is our reasoning, not a Penlid or Google rule. A page's gap is its percent change minus the site's percent change; the clicks at stake are then its baseline clicks a day, times the size of a negative gap as a fraction, times 28.
Two simpler orderings can mislead. Percent fallen ignores size, so a steep fall on a tiny page can outrank a gentler fall on a busy one. Raw clicks lost includes the fall every page shared, so the biggest page can top the list while doing exactly what the site did. If some pages matter more than their clicks suggest, such as pages that sell, weight the list by that.
Why is a recovery after a refresh hard to credit?
A page picked for a bad stretch tends to recover whatever you do, so a refresh can take credit for a rebound that was coming anyway. The methodology puts it this way: A page that has just had a bad four weeks is likely to have a better next four whatever you do to it
. That is regression to the mean: an unusually high or low measurement tends to be followed by one closer to the average, as Wikipedia describes it (see why pages that dropped often recover on their own).
Penlid measured the problem on simulated sites. Pages whose last 28 days sat at least one standard deviation below their own average, and which were then changed in no way at all, still came back "improved" at the rates below; pages not picked that way were falsely called improved at most 3.10% of the time. Drift means pages wandering apart on their own.
| Pages in the simulation | Called "improved" |
|---|---|
| Picked for a fall, 4 clicks a day, no drift | 5.50% |
| Picked for a fall, 40 clicks a day, no drift | 5.75% |
| Picked for a fall, 4 clicks a day, 15% drift | 9.15% |
| Picked for a fall, 40 clicks a day, 15% drift | 9.20% |
| Not picked that way, at most | 3.10% |
Penlid's pre-trend warning, which flags a page already moving differently from its comparison pages, appeared on only 10.9% to 22.7% of those pages: a useful flag, not a safeguard.
How do you refresh pages so you can measure the result?
Set aside a small test group, kept out of your queue and chosen by a rule that ignores recent clicks; refresh half of it and leave the other half alone. That is our reading of the methodology's remedy, not a Penlid prescription: choose which pages to change without looking at how they were trending
, which it spells out as every page in a set, or alternate pages
(see how to choose pages without bias). Every other page in a list sorted by publish date fits that rule; the decay queue does not, because it is picked by trend, so read its results with caution.
Leave the held-back pages untouched, and keep at least three of them: Penlid draws a changed page's comparison pages from the untouched pages on your site, leaves out any page changed from 56 days before the change to 28 days after it, and with fewer than three untouched pages left the result is "not measurable" rather than a number.
Log each change as you make it: the URL, the date, what you changed and which pages you left alone. The content operations ledger guide covers what else to record. If AI drafts the new text, check its facts against primary sources and have a person approve it before it goes live.
Give it time: Google's page on debugging traffic drops says changes to your site may take time to show an effect and advises waiting a few weeks before analyzing again. Penlid's did-it-work check sets each changed page against pages you left alone over the same days. A rise afterward is timing evidence, not proof; where you picked pages for decline, read an "improved" as at least as good as the pages left alone, and possibly some recovery that was coming anyway
.
Does changing a page's date count as a refresh?
A new date on unchanged content is not a refresh, and Google's list of questions for spotting search engine-first content includes it. In its people-first content guidance, Google asks Are you changing the date of pages to make them seem fresh when the content has not substantially changed?
In our reading, a refresh worth logging is one where you can say what changed; the date follows.
Example (illustrative numbers)
These numbers are invented to show the arithmetic; they are not data from any site. Suppose the rest of the site, with the page being judged left out as the tool does, averaged 10% fewer clicks a day over the last 28 days than over the 84 before. Penlid's pages say the site's move is subtracted but print no gap formula, so this is our version: daily averages, subtracted in percentage points.
| Page | Clicks a day, prior 84 days | Clicks a day, last 28 days | Gap against the site | Clicks at stake in 28 days |
|---|---|---|---|---|
| A | 400 | 360 | 0 points | 0 |
| B | 60 | 36 | −30 points | 60 × 30% × 28 = 504 |
| C | 30 | 12 | −50 points | 30 × 50% × 28 = 420 |
| D | 3 | 1 | not judged | not ranked |
A page's change is its last-28-days column ÷ its prior-84-days column − 1, and its gap is that change minus the site's −10%. Page B: 36 ÷ 60 − 1 = −40%, so its gap is −40 − (−10) = −30 points. Page C: 12 ÷ 30 − 1 = −60%, a gap of −50 points. Page A: 360 ÷ 400 − 1 = −10%, exactly the site's, so its gap is 0. Page D had 2,310 impressions in its trailing 90 days, under the 3,000 line, so it is not judged whatever its percent says.
The yardstick decides the order. By percent fallen, D comes first (1 ÷ 3 − 1 is about −67%). By clicks lost a day, A comes first (400 − 360 = 40), though all of its fall is the site's. By clicks at stake, B comes first: its gap is smaller than C's, but 60 × 30% = 18 clicks a day lost against the site beats 30 × 50% = 15. So the queue is B, then C; A needs nothing from this list, and D waits for more impressions. B and C were both picked for falling, so a rise after refreshing either could include recovery that was coming anyway.
What this cannot tell you
- Why a page fell. A rival, a shift in what searchers want, a lost link or a broken page can all open a gap, and a refresh fixes only some of them (our reasoning).
- Whether a refresh will help. No Google page we read gives a refresh schedule or a staleness threshold, and the ranking systems guide lists freshness systems without a rule for updating old pages; Penlid's 28-day, 84-day and 3,000-impression numbers are its own design.
- One yardstick for the whole site. Every page is judged against one site-wide figure, so sections that move differently, such as one seasonal and one evergreen, can make a page look better or worse than it is (our reasoning).
- Simulated rates. The 5.50% to 9.20% figures come from simulated sites; they show how large the effect can be, not what it is on yours.
- Old files and thin pages. A page under 3,000 impressions in the trailing 90 days gets "can't tell", and verdicts are read against the last day in your export, not today, so an old file ranks pages as they stood then.
Questions people also ask
How do you decide which old pages to update first?
Look for pages whose clicks fell by more than the rest of your site's did, drop any with too few impressions to judge, and sort what remains by the clicks each one lost against the site. A page that fell no more than the site did is following the site, so it is not a decay candidate.
How often should you refresh a page?
No Google page we read sets a schedule, and Penlid's pages give none either, so let evidence set the timing instead of the calendar. The decay check looks at the last 28 days, so a second export a week after the first mostly re-judges the same days; a monthly look is a reasonable rhythm (our reasoning, not a rule).
Does refreshing an old page bring its traffic back?
Sometimes the page does better afterward, but a rise alone does not show the refresh did it. A page chosen because its last four weeks were poor is likely to do better next, with or without a refresh, and in Penlid's simulations unchanged pages picked that way still read as improved 5.50% to 9.20% of the time. Compare the page with pages you left alone over the same days before crediting the refresh.
Should you change a page's date when you refresh it?
Only when the content changed substantially, in our reading. Google's people-first content guidance lists changing page dates to look fresh, with no substantial change to the content, among the questions to ask about search engine-first content, and it lists adding a lot of new content or removing a lot of older content mainly to make a site seem fresh too. Note what changed beside the date so the edit can be checked later.
How long should you wait before judging a refresh?
Google offers ranges, not one number. Its traffic-drop guide says some site changes take effect in a few days and others could take several months, and suggests a few weeks before analyzing again; its recrawl page says a recrawl can take from a few days to a few weeks. Penlid compares the 28 days before a change with the 28 days after it, which is Penlid's own choice, not Google's (see how long to wait after a content update).
Sources
- Creating helpful, reliable, people-first content, Google, Search Central, read 30 October 2026.
Are you changing the date of pages to make them seem fresh when the content has not substantially changed?
- Performance report (Search results): Common tasks and use cases, Google, Search Console Help, read 30 October 2026.
- Debugging drops in Google Search traffic, Google, Search Central, read 30 October 2026.
- Ask Google to recrawl your URLs, Google, Search Central, read 30 October 2026.
- A guide to Google Search ranking systems, Google, Search Central, read 30 October 2026.
- Regression toward the mean, Wikipedia, read 30 October 2026.
- Methodology: how we measure, and what we can't, Penlid, read 30 October 2026.
A page that has just had a bad four weeks is likely to have a better next four whatever you do to it
choose which pages to change without looking at how they were trending
every page in a set, or alternate pages
at least as good as the pages left alone, and possibly some recovery that was coming anyway
- Content decay check: which pages are losing traffic?, Penlid, read 30 October 2026.
a January dip that hit everything reads as healthy