TrafficTriageOrganic Traffic Still Dropping? One Page, Not the Site
By Ugur Saritepe · October 28, 2026
Organic traffic still dropping, and you cannot tell whether to wait or fix something? On 2026-10-08 I asked my assistant that about my own site. The site total read down 3% over a fortnight. The page table read one page down 177 clicks and every other page up 13%. The answer split the same way: wait on the one page, act on the rest.
The question, as I typed it
game-scout.app is a site I run with about 110 curated game lists. Its clicks had surged in late August, peaked in the second week of September, and come down since. Every compare I opened read red. So I asked.
Let's check our GSC stats. It's still decreasing, do we have anything to do about this or do we have to wait and accept this?
sc-domain:game-scout.app, clicks by week (Sat to Fri), 22 Aug to 6 Oct sc-domain:game-scout.app, by page, 22 Sep to 5 Oct vs 8 Sep to 21 Sep
week Aug 22 Aug 29 Sep 5 Sep 12 Sep 19 Sep 26 Oct 3 (4 days) clicks 571 811 850 644 563 625 358 14-day compare, by page: site 1,233 vs 1,275 best-detective-games 52 vs 229
It is not decreasing anymore. Site clicks have been flat for three weeks at the level they were before the late-August surge. The whole drop you are seeing is one page.
The session pulled its numbers through a Search Console CLI. Every figure below was re-pulled for this post through the MCP server on 2026-10-09, and Search Console data ends 2026-10-06. Three cuts of the same data answered the question, and they are the whole post.
Cut one: the weeks, against the week before the surge
week (Sat to Fri) clicks impressions Aug 22 to 28 571 28,362 the week before the surge Aug 29 to Sep 4 811 37,834 Sep 5 to 11 850 42,231 the peak Sep 12 to 18 644 34,070 Sep 19 to 25 563 36,746 Sep 26 to Oct 2 625 38,350 Oct 3 to 6 (four days) 358 20,969
"Still decreasing" was true against the peak and false against the week before it. Three weeks at 563 to 625 clicks, against 571 in the last ordinary week of August. The 30 days to 2026-10-06 held 2,701 clicks on 157,555 impressions, more than the 30-day block before it (2,592 on 126,014), and more than any 30-day block since June (673 in June, 1,065 in July). A compare anchored on a spike reads every later week as a loss; I ran into the same trap in pages lost rankings but nothing changed, where the recent fortnight was 72% of the peak and 95% of the fortnight before the peak.
Cut two: the page table, sorted by click difference
Google's guide to debugging traffic drops says to open the Pages report and "Select Clicks Difference to order it based on pages that lost the most traffic." I did that over two equal fortnights, then added a step the guide does not: I summed every page except the top loser.
22 Sep to 5 Oct, against 8 Sep to 21 Sep clicks impressions position whole property 1,233 / 1,275 75,842 / 71,679 8.1 / 9.4 best-detective-games 52 / 229 1,842 / 4,385 7.5 / 6.8 every other page together 1,181 / 1,046 pages with at least one click 115 / 88 pages up / down / unchanged on clicks 71 / 44 / 199
One page lost 177 clicks. The site lost 42. Every other page together gained 135, up 12.9%. Seventy-one pages went up and forty-four went down, and the site total netted one page's loss against the rest. Google's Search Console help page on traffic drops puts it in one line: "If only a few pages on your site are responsible for most of the drop, think about why that happened."
gainers clicks losers besides detective clicks best-flight-games 46 to 98 best-coop-rpg-games 41 to 17 best-trucking-games 78 to 94 best-lego-games 47 to 29 best-police-games 36 to 53 best-animal-games 37 to 23 best-hunting-games 19 to 30 best-medical-games 29 to 17 best-vampire-games 4 to 17 best-medieval-fantasy-games 27 to 16 best-mystery-games 1 to 14 best-samurai-games 35 to 26
Two honesty notes on the 12.9%. Twenty-seven of the 115 pages with a click earned their first click in the later fortnight, so part of the gain is new pages arriving, not old pages recovering. And the compare returned 314 page rows marked truncated; the rows sum to 1,233 and 1,275 clicks, the same as the totals, so every click in both windows is inside them. A truncated result where the rows do not reach the total is the trap I wrote up in queries do not add up to the total.
Cut three: the one page
The best-detective-games list had been the site's top page. Over the same two fortnights its clicks went 229 to 52 and its impressions 4,385 to 1,842, while its average position read 6.8 then 7.5. The query split says what that average hides.
best-detective-games, same two fortnights clicks impressions position ps5 detective games 13 / 1 88 / 59 4.0 / 7.9 detective games ps5 13 / 6 139 / 71 5.0 / 7.0 ps5 mystery games 7 / 1 78 / 21 6.4 / 7.1 best detective games ps5 10 / 6 74 / 56 5.1 / 5.5 19 queries with 1 or 2 clicks each 26 / 0 244 / 0 no row in the later window
The two head queries lost both position and slots, and the tail stopped being shown at all. "ps5 detective games" went from position 4.0 to 7.9 and from 88 impressions to 59. Nineteen queries that each sent one or two clicks in the earlier fortnight have no row in the later one; Search Console shows a row only when the page appeared. The query table is capped at 60 rows here, so the tail is a sample, but the shape is not in doubt.
On the live results page, US, 2026-10-09, game-scout.app is absent from the top ten for both "detective games ps5" and "ps5 detective games". What is there: a Reddit thread at 1, a YouTube ranking at 2, two list pages at 3 and 4, PlayStation's own puzzle-games editorial at 5, then a second Reddit thread, two Facebook group posts with no snippet, a store listing and one more list at 10. Not a competitor list that displaced us; a results page that stopped having a slot for us.
And the crawl. URL Inspection on 2026-10-09: "Submitted and indexed", fetch successful, canonical matching, last crawl 2026-10-03 at 13:28 UTC. The crawl before that was 2026-09-17, so the page sat 16 days without a fetch while its clicks fell, and the fetch landed 36 minutes after I had logged a prediction saying I would request indexing and leave the page alone. The page's content file was last committed on 2026-09-12, before the drop, and has not changed since. The three days of data after the recrawl read 6, 2 and 1 clicks.
Wait or fix: the answer split
Wait on the page.Google recrawled it three days before I asked. The prediction I logged on 2026-10-03 reads "owner requests indexing in GSC (manual) and the page is otherwise left untouched so the recovery can be read cleanly", baseline 73 clicks in the fortnight to 09-29, expected 150 or more in the 14 days ending 2026-10-27. Edit the page now and the next read measures the edit and the recrawl together. Google's guide says the same thing twice, once for the page and once for the site: "we recommend avoiding making radical changes if your page is already performing well", and "you'll likely want to wait a few weeks to analyze your site in Search Console again". I have the waiting half of this covered in how to prioritize SEO tasks, where every bet gets a check date before it ships.
Act on everything else, and not on the 71 pages that went up.The act half was already written down. On 2026-10-03 the same review had set three actions beside the detective request. On 2026-10-08, when I asked whether to wait, none of the three had shipped, and the one that had shipped was the page I was being told to leave alone. The honest answer to "do we have anything to do" was yes, and it was the list from five days earlier.
What we have not done
Checked against the repository on 2026-10-09. We have not shipped the harvest proof on best-coop-rpg-games that prediction 17 was logged for on 2026-10-03: the page still reads a last update of 2026-08-04 and no commit since then mentions it, so a prediction with a check date of 2026-10-27 has no change under it. We have not changed the flight list's title to the words its queries use; it still reads "Best Flight Games for PS5 & PS4 (2026) - Ranked", last edited 2026-09-12. We have not switched the daily comparison routine off the format that graded at 0.42% click-through on 2026-10-03; four more comparison articles merged between 10-04 and 10-08. And we have not established why the detective page lost its slots; the thin-page incident of mid-September was rejected as the cause on its own evidence, and the results page shows forums and a video, not a list that took our place.
What this does not tell me
- Why the page fell. Three cuts located the loss and dated the recrawl. They did not name a cause, and this post does not.
- Whether waiting works. Three days after a recrawl is not a readout. The 14-day window closes 2026-10-27 and this post gets a dated follow-up either way.
- One site, two fortnights, one owner. The split was one page against the rest here; a drop spread across a page group would need the group read as a unit.
- The tail is a sample. The per-query table is capped at 60 rows, and 27 of the 115 earning pages are new supply inside the 12.9%.
- One live read. Two queries, US, one day. Other countries and desktop could show a different page.
How to run this on your own site
- Weeks first, including the week before any surge. Performance, Search results, Date view, three months. If the recent weeks sit at the level before the spike, the spike was the event, not the decline.
- Pages by click difference, then the sum without the top loser. Compare two equal windows, order by Clicks Difference, and subtract the first row from the site total. If the remainder is flat or up, the site is not dropping; one page is.
- Count pages up against pages down.Seventy-one against forty-four is a healthy site with one injury. Ten against ninety is a site-wide problem, and Google's guide says to assess the whole site, not the pages.
- Open the loser's queries and its last crawl. Head queries moving down two to four places with impressions halving, and a tail that vanished, is a slot loss. Then URL Inspection: if the last crawl is older than the drop, request indexing and log the date.
- Split the decision per page. Recrawled in the last few days, or inside a logged test: wait until the window closes. Decided weeks ago and never shipped: that is the work. The pages that went up need nothing.
If you want the serving side of the losing page checked while you wait, the free triage report runs eight checks from the URL alone. The site-level version of this question, where a 28-day compare blended two separate events, is in impressions dropped but position improved, and the rest of these measured reads are in the AI-assistant SEO series.
FAQ
Should I wait or make changes when my organic traffic drops?
Decide it per page, not per site. Sort the Pages report by click difference and add up every page except the biggest loser. If the rest is flat or up, the site does not need a change; the one page might. If that page was recrawled in the last few days or has a logged test running on it, leave it alone until the window closes. Act on the changes you had already decided and never shipped. On my site that was three items, and none of them was the page that dropped.
How do I tell whether a traffic drop is site-wide or one page?
Compare two equal windows in the Pages report, order by click difference, and read three numbers: the site total, the top loser, and the total without the top loser. On game-scout.app for the fortnight to 2026-10-05 those were 1,233 against 1,275, 52 against 229, and 1,181 against 1,046. The loser carried more than the whole net drop, and 71 pages went up against 44 down. Google's own traffic-drops guide names the click-difference sort for exactly this question.
How long should I wait after Google recrawls a page before judging it?
Longer than three days, which is what I had. Google's guide says to wait a few weeks before reading Search Console again after a change, and my own logged test on the page sets a 14-day window ending 2026-10-27 against a 73-click baseline. Editing the page inside that window would mean the next read measures the edit and the recrawl together, and neither alone.
The site total said down 3%. The page table said one page down 177 and seventy-one pages up. Only one of those is a decision. Create a free account, connect Search Console, and the Pages view lists every tracked page with its clicks and impressions against the previous fortnight, sortable by the difference, with its last crawl date beside them. With the MCP serverconnected your assistant reads the same table, so "is it still decreasing, or is it one page" is a question you type rather than a spreadsheet you build.
In your AI assistant
Ask your AI about your own site
TrafficTriage plugs into Claude, Cursor, and ChatGPT. Your assistant runs the 8 checks, reads your Search Console data, and can even check your site every morning. Free account, no API key, nothing to install.