Pages Lost Rankings But Nothing Changed: What Split Them

By Ugur Saritepe · October 2, 2026

Twenty-five of my pages lost clicks in a fortnight and not one of them had been edited in three months. The table I used to pick which one to fix could not tell me which ones had a page problem. One query pull could, and it said the page I had chosen was not the problem.

Why I went looking

game-scout.app is my games site and my test bed. On 2026-10-01 I asked for a shortlist of pages to work on:

now lets check first our best list pages. which ones are not created/changed recently then we can pick example for what to work now

That is a normal maintenance question and it has a normal answer: list the pages, sort by what they lost, pick the top one. What follows is what happened when I checked whether the sort order meant anything.

First, is the set actually falling

My own read of the last few weeks was that everything had risen and was now sliding back. That turned out to be an artefact of which two weeks I was comparing. Daily clicks across all the list pages, 2026-08-01 to 2026-09-29, 60 rows and not truncated:

window                  clicks   shape
2026-08-01 .. 08-14        622   33 to 57 a day
2026-08-19 .. 09-01      1,161   the climb, 137 on 08-30
2026-09-02 .. 09-15      1,537   the peak, 169 on 09-05
2026-09-16 .. 09-29      1,100   trough of 49 on 09-17, then 93/118/109/76/80

The current fortnight is 71.6% of the peak fortnight and 94.7% of the one before the peak. The last five days of data ran at average positions 7.74, 7.84, 7.49, 7.65 and 7.60, the best stretch in the whole 60-day series against 9.2 to 12.1 in early August. So the set is not sliding; it came off a spike and the most recent week is already turning back up. That settles the site-level question and the rest of this post is about the pages.

The list Google tells you to make

Google's own traffic-drop guide is specific about how to find the pages worth working on: review the Pages table below the chart to find patterns that might explain where the drop is coming from, sorted by clicks difference. I did exactly that, with one extra filter: only pages with no real edit in the last 90 days, so that nothing in the comparison is explained by me having touched it.

That filter matched 82 of the 165 pages my tracker watches on this site. Every one of them records no edit in 90 or more days. These are the seven biggest click losers, with everything the table knows about them. score is the internal link score, in is inbound internal links, health is how many of six on-page checks pass:

page                        score  in  health  clicks      position
best-detective-games           30  12  6/6     343 ->  87  6.34 -> 7.76
best-coop-rpg-games            58  17  6/6      75 ->  26  8.26 -> 8.23
best-fishing-games             25  11  6/6      94 ->  47  8.09 -> 7.99
best-golf-games                 9   6  6/6      33 ->   8  9.88 -> 9.25
best-open-world-rpg-games      63  21  5/6      28 ->   3 10.93 -> 14.55
best-animal-games              10   7  4/6      44 ->  24  6.96 -> 7.52
best-police-games              17   7  4/6      61 ->  44  7.04 -> 7.09

I picked best-detective-games. Biggest absolute loss on the site, real demand still arriving, six checks out of six passing, so a clean content-refresh case with no technical noise to confuse the read. Then I wrote this, before any of the query data came back:

I dont think any of these failing because of title or meta. We did not change anything and they started to move up all of them, now they are slowly decreasing. I am not sure if there any difference between page that working well right now

Nothing on the table separated the two sides

So I pulled the other tail: the 15 pages in the same cohort that gained the most clicks over the same two fortnights, with the same no-edit filter. If any attribute the tracker stores predicts the outcome, the two lists should look different. They do not.

                              25 biggest losers   15 biggest gainers
internal link score, mean                  28.0                 22.8
inbound internal links, mean               10.3                  9.8
failing 1+ of 6 health checks           8 (32%)              4 (27%)
edited in the last 90 days                    0                    0

The losers score higher on internal links than the gainers, and the three highest-scoring pages in the whole sample, at 63, 62 and 58, are all losers. Health fails at almost the same rate on both sides. The worst-scoring page on health in the entire sample, best-house-design-games at four checks of six with both title and meta description failing, gained 6 clicks and improved 1.40 places. And nothing was edited on either side, so staleness cannot be the discriminator when it is the one thing both lists share.

Every column on my page table appeared on both sides of the winner-loser split. A page report tells you what a page is like. It does not tell you why a page lost, because the cause is often not on the page.

The position column points the wrong way

The worst offender is the column that looks most like an answer. Eighteen of the 25 biggest click losers improved their average position in the window they lost the clicks.

best-lego-games lost 16 clicks while improving 2.20 places, on impressions that barely moved (1,408 to 1,292). Its click-through rate fell from 3.98% to 3.10%. The listing changed around it, not its ranking.
best-hunting-games lost 10 clicks on 478 more impressions while improving 0.83 places.
best-crime-games lost 6 clicks on 103 more impressions while improving 0.08 places.

This is arithmetic, not mystery. Average position is a mean weighted by impressions, so losing the impressions you ranked best for and keeping the ones you ranked worst for drags the average down, and losing the ones you ranked worst for pushes it up. The number moves for reasons that have nothing to do with whether anyone clicked. Google says the same thing in one line of the guide above: you shouldn't focus too much on your absolute position. I had been reading it as a page-health column anyway. The longer autopsy of that metric is a separate post.

One level down, the two pages stop looking alike

The thing that does separate them is not on the page table at all. It is the per-query view of each page, pulled for the same two fortnights. Here is the page that lost three quarters of its clicks against a page in the same cohort, same no-edit status, same six-of-six health, that gained 11.

best-detective-games        343 -> 87 clicks   page average 6.34 -> 7.76
  ps5 detective games        23 ->  1        3.02 -> 7.19
  detective games ps5        16 -> 10        4.30 -> 6.70
  best detective games ps5   11 ->  8        3.60 -> 6.40

best-trucking-games          82 -> 93 clicks   page average 7.46 -> 6.93
  best truck simulator ps5   12 -> 11        3.79 -> 3.71
  best truck game ps5         5 ->  4        3.39 -> 3.21
  best trucking game on ps5   2 ->  2        3.63 -> 3.59

Both pulls returned every row, 139 on detective and 32 on trucking, with no truncation, so these are the real sets and not a sample. The detective page's three biggest earners all sat in the top four and all three now sit between 6.4 and 7.2. Its single best query went from 23 clicks to one, and its click-through rate on that query from 19.0% to 1.6%. The trucking page's three biggest earners sat between 3.4 and 3.8 and still do. Not one trucking query moved more than half a place in either direction.

Note what the page average did while that happened. Detective's moved 1.41 places, from 6.34 to 7.76, while three quarters of its clicks left. A head query falling four places is invisible in a page average that is also being pulled around by two hundred long-tail queries nobody clicks.

The pages that held their clicks held a top-four position on their main query. The pages that lost clicks lost that position. Everything else I measured appeared on both sides.

What that changed about the decision

I was about to rewrite best-detective-games. The query data does not say a rewrite is pointless in general; it says this page did not get worse. Its content is what it was in July, its checks all pass, it still earns 2,362 impressions a fortnight, and it is still being shown for the same queries. What changed is the position it is shown at, and nothing in the page explains that.

So the decision moved from "refresh the biggest loser" to "find out what now occupies positions three to six on detective games ps5on a phone, and decide whether that is reachable at all". Those are different weeks of work, and the page table pointed at the wrong one. That is the part worth taking away: the sort order told me where the loss was, which is what it is for, and I read it as telling me where the fault was, which it never does.

What this does not tell me

Four limits, because the finding is only as good as what is behind it.

I have not identified what moved above us. Nothing in this run looks at the live results page. The head-query drop is measured; the cause of the head-query drop is not.
The query rows cover less than half the page. On the detective page, 48 queries earned at least one click in the earlier fortnight and they account for 143 of its 343 clicks, 41.7%. Search Console withholds low-volume queries, so the head-query story explains well under half of what moved, and I cannot see the rest at all.
This is two tails of one comparison, not an experiment. The gainer list is the top 15 by click change rather than all 57 non-losers, and the window straddles the spike in the first chart.
At least three different things are inside the 25. A head-position loss, impressions holding while click-through rate falls, and impressions falling while position improves are three distinct problems. This read separates them from the page attributes. It does not apportion them.

One more, on method, because I got it wrong in the session this post is built from. My first query pull used a 25-row limit with a device split, and about fifteen queries appeared to have vanished from the page entirely. Re-running it with no truncation showed that 44 of the 48 were still there, holding impressions and losing only their clicks: one query went from 9 clicks to 0 while keeping 25 impressions at position 6.68. A comparison that returns nothing for the current period means not in the rows I fetched, which is a different claim from zero. Four queries really did leave the set; the other forty were a row limit.

How to run this on your own site

Four steps, about fifteen minutes. The point of the order is that each step can end it.

Settle the baseline before the page question. Pull daily clicks for the section you are worried about over 60 days, not two fortnights. If the recent window contains a spike, every comparison against it reads as a decline and the pages are not the story.
Make both lists, not one. Sort by click change descending as well as ascending, and filter both to pages you have not edited. A column that appears on both lists is not a cause, and you cannot know that from the losing list alone.
Pull queries for one loser and one gainer. Same two windows, no row limit you might hit, dimensions by query and filtered to the one page. Compare the three queries that earned the most clicks in the earlier window. This is the step that separates a page problem from a results-page change, and it takes one pull per page.
Only then decide what to change. Head queries still getting impressions at a worse position means the page did not get worse. Impressions holding with click-through rate falling means the listing changed around you. Impressions gone across every query means look at demand or indexing, and the catalogue-page read covers that case.

What we have not done

Our own tracker has had the title and meta-description checks failing on best-police-games for the whole of this window, on 2,248 impressions, and we have not fixed them. The session this post comes from named that page as the quick mechanical alternative, and then I went looking at the query data instead and never came back to it. It is convenient for the argument above that the health checks did not predict anything, and it is still a thing we should have done.

All 82 pages in this cohort are stale by our own tool's definition, and we have not edited one of them since this read. The finding is that staleness did not separate the two sides, not that staleness is fine. Pages that were edited and then moved are a different case, and we measured that one separately. The rest of these reads are in the AI-assistant SEO series.

FAQ

Why did some of my pages drop and others didn't?

On our site the answer was not on the pages at all. We took 82 list pages with no edit in 90 days and compared the 25 biggest click losers against the 15 biggest gainers over the fortnights to 2026-09-29. Internal link score averaged higher on the losers (28.0 against 22.8), on-page health failed on 32% of losers and 27% of gainers, and the page with the worst health in the whole sample gained clicks. The one thing that differed was what happened to each page's top query: the losing page's three biggest queries fell from the 3.0 to 4.3 band to the 6.4 to 7.2 band, and the holding page's stayed between 3.2 and 3.8.

Can a page lose rankings without anything changing on it?

Yes, and on our site that is what happened. None of the 82 pages had been edited in 90 or more days, so nothing on either side of the comparison had changed. One page lost three quarters of its clicks and another in the same cohort gained 11, with no edit to either. When nothing on the page moved, the thing that moved is above you in the results, and you cannot see that in a page report.

Does average position tell me which pages are in trouble?

Not reliably. Among our 25 biggest click losers, 18 improved their average position while losing clicks. One page lost 16 clicks while improving 2.20 places; another lost 10 clicks on 478 more impressions while improving 0.83. Average position is a mean weighted by impressions, so when a page loses the impressions it was ranking worst for, the average gets better while the traffic gets worse. Google's own guidance says the same thing in one line: impressions and clicks are the measure of success, not absolute position.

How do I tell a page problem from a results-page change?

Pull the page's queries, not the page's totals, and compare the same two windows. If the queries that used to earn the clicks are still returning impressions at a worse position, the page did not get worse and something above it changed. If impressions held and only the click-through rate fell, the listing changed around you rather than under you. If impressions themselves fell across every query, demand or indexing is the place to look next. The page's own average position will not separate these three.

A page report tells you which page lost. Only the query rows tell you whether the page is why. Create a free account, connect Search Console, and the MCP serverlets your assistant pull your own click rows per page and per query, for two windows at once, in a single message. Making both lists and then pulling two pages' queries took me four messages. The afternoon I saved was the one I would have spent rewriting a page that had not got worse.

Create an account

Claude running a scheduled TrafficTriage daily check: a Search Console report on a real site, summarized with clicks, impressions, and the one query cluster that lost rank

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.