Impressions Dropped, Position Improved: One Was Real

By Ugur Saritepe · October 6, 2026

Impressions dropped but average position improved? On my own site the 28-day compare read impressions down 46% and position up from 26 to 10.5. Every ranked explanation says that is Google cleaning bot impressions out of the data. On my site it was two events a month apart, and only the first one was harmless.

Where this came from

traffictriage.com is this site, the one I run the product on. A scheduled morning report pulls its Search Console numbers every day. On 2026-10-05 it flagged the week.

sc-domain:traffictriage.com, last 7 days vs the 7 before, totals

Sep 29 to Oct 5:   0 clicks     5 impressions
Sep 22 to Sep 28:  0 clicks   170 impressions   (-97%)
most recent day with data: Oct 2

So on 2026-10-06 I pulled the whole picture: 56 days of daily data, the same window on a second property I own, the country and page splits, 28 URL inspections, and the live results page for four queries. Search Console data ends 2026-10-03, three days behind.

What the 28-day compare showed

28 days to 2026-10-06 vs the 28 before     impressions   clicks   position
sc-domain:traffictriage.com                    545 / 1,011    3 / 5   10.5 / 26.3
desktop                                        421 /   812    0 / 2   10.8 / 29.9
mobile                                         122 /   196    3 / 3    9.7 / 11.5

Impressions down 46%, position better by nearly 16 places, clicks nearly unchanged. That is the exact signature six of the ten pages ranking for this question describe, and all six give the same cause: in September 2025 Google stopped counting the deep-position impressions that rank trackers generated, so impressions fell and average position rose everywhere at once. The advice that follows is to relax. The daily series on my property did not support relaxing.

Event one, 2026-09-01: the deep tail ended

The first thing the compare hid is that the position jump and the impression drop happened on different days. Average position moved on 2026-09-01, from the 30s to about 10, and impressions did not move with it. The country split shows what left.

country        12 to 31 Aug (20 days)     1 to 27 Sep (27 days)
Vietnam          59 impr at pos 67.7         0
Philippines      40 impr at pos 62.0        25 impr at pos  8.4
Indonesia        31 impr at pos 67.0        11 impr at pos 10.2
Bangladesh       23 impr at pos 66.0         6 impr at pos  8.5
Malaysia         19 impr at pos 55.4         3 impr at pos  6.7

Turkey           13 impr at pos 54.5         3 impr at pos  8.3
Mexico           11 impr at pos 57.2         3 impr at pos  8.7
nine countries  212 of 665 (32%), 0 clicks   57 of 881 (6.5%), 0 clicks
United States   123 impr at pos 11.3        325 impr at pos 10.9
whole property  665 impr, 33.3 a day, 34.9  881 impr, 32.6 a day, 10.2

Nine countries had been supplying a third of the property's impressions at positions 42 to 68, with zero clicks on all 212 of them. On 2026-09-01 that stopped. Vietnam went from 59 impressions in twenty days to zero on every day since. The United States, which was earning the clicks, nearly doubled per day. Impressions a day: 33.3 before, 32.6 after. Google defines position as the topmost position of your link, averaged across all queries, so removing 212 rows at position 60 from the average moves it from 35 to 10 without a single page ranking higher. That is the harmless version of this pattern, and it is the only version the ranked explainers describe. I cannot say what those impressions were. Search Console does not say whether a position-67 impression in Vietnam was a person on page seven or something automated, and nothing on my site was built for either. I decomposed one position into its head and tail the same way in is average position accurate.

Event two, 2026-09-28: the head ended

Four weeks later the impressions that were left went too, and this time the pages that earned them were the ones that went.

daily impressions, sc-domain:traffictriage.com
Sep 16 to 27   25  24  27  30  29  29  20  36  35  44  16  15
Sep 28 to Oct 3  4   2   0   0   3   1        (data ends Oct 3)

page                               14 to 27 Sep       28 Sep to 6 Oct
vercel-site-not-showing-up          119 at pos  8.0     6 at pos 5.5
github-pages-not-showing-up          55 at pos 10.1     0
netlify-site-not-showing-up          40 at pos  8.5     1 at pos 9
lovable-site-not-showing-up          30 at pos 10.1     0
replit-app-not-showing-up            26 at pos  7.6     1 at pos 10
whole property, 29 pages            393, 28 a day      10, about 1 a day

Fourteen days at 28 impressions a day, then six days at about one. The United States and everywhere else fell together: US daily 17, 6, 2, 1, 1, 0, 0, 1, 0 from 09-25; the rest 27, 10, 13, 3, 1, 0, 0, 2, 1. Desktop and mobile fell together. And the average position of the ten impressions that remain is 5.8, better again. If I had opened the compare on 2026-10-06 and read "position improved" as the explainers tell me to, I would have read a collapse as a clean-up.

Five checks, in the order Google's own guide puts them

Google's guide to debugging traffic drops gives a cause table: a reporting glitch, an algorithmic update, a site-wide technical issue, seasonality. I ran it on 2026-10-06 in that order, with one check it does not list.

1. The Data Anomalies page. Google logs its own reporting errors at the Search Console data anomalies page. It has no entry for September or October 2026; the latest is a Generative AI report error on 2026-08-13 to 08-17. Not a logged glitch.
2. A second property. game-scout.app sits on the same account and the same Search Console connection. Its daily impressions from 09-25 to 10-03: 5,622; 6,924; 6,567; 5,105; 4,969; 4,562; 4,613; 5,610; 6,551. Its 28-day compare: up 4%. A reporting change moves every property on the same days. This moved one.
3. The ranking updates dashboard.Google's Search status dashboard lists a September 2026 spam update released on 24 September 2026, which "applies globally and to all languages. The rollout may take up to two weeks to complete." My drop began four days into that window. That is a coincidence in time and nothing in my data connects the two beyond it; Google's spam updates page says sites that see a change after one should review the spam policies, and says recovery is measured in months.
4. Serving and index.I inspected the four pages that lost the most. All four are "Submitted and indexed" with matching canonicals; Google fetched the Vercel guide successfully on 2026-09-30, inside the collapse, and another page on 09-25. The sitemap was downloaded on 2026-10-01 with 67 URLs and no errors. The site is behind a Vercel security challenge that answers 429 to every scripted request I sent, including one with a Googlebot user agent; Vercel documents verified Googlebot as allowed through, and Google's successful fetches say it is. Nothing here stops a page from being served.
5. The live results page.Four queries where my 28-day position reads 7.5 to 9.9: "netlify site not showing up on google", "vercel site not showing up on google", "replit app not showing up on google", "github pages not showing up on google". On 2026-10-06, US, traffictriage.com was absent from all four top tens. The 28-day averages are nineteen days of ranking blended with nine days of not.

Checks one and two exclude a reporting change. Check four excludes a serving fault I can see. Check five confirms the loss is real on the results page, not only in the report. Check three is the only dated external event, and it is a date, not a cause.

What I found on the way: 27 posts, one crawled

The index check turned up a second problem that has nothing to do with the drop. Since 2026-09-03 this site has published 27 posts, one a day, most of them case studies like this one. I inspected all 27.

27 posts published 2026-09-03 to 10-05, inspected 2026-10-06
Submitted and indexed                 1   (crawled 2026-09-25)
Discovered - currently not indexed   18   (no crawl recorded)
URL is unknown to Google              8   (5 of them in the sitemap Google read on 10-01)
the pillar they all link up to            indexed, last crawled 2026-07-22

None of those 27 pages has ever had an impression, so none of them is part of either event above. They are a separate finding: a month of daily publishing that Google has discovered and not fetched. The indexing cliff I wrote up on the other property in Google stopped indexing my new pages is now on this one. And the explanation I gave in do internal links help SEO applies to me: every one of the 27 is linked from the other 26, none of which is crawled, and from a hub page Google last fetched on 2026-07-22. Links from pages Google does not visit pass nothing.

What this does not tell me

• Why event two happened. The sequence excluded a reporting change and a serving fault and confirmed the loss on the results page. It did not name a cause, and this post does not.
• Three days are missing. Data ends 2026-10-03, and 10-01 to 10-03 can still fill in. 09-28 to 09-30 (4, 2, 0) are final.
• What the deep tail was. 212 impressions at positions 42 to 68 in nine countries with zero clicks is consistent with people paging deep and with automated traffic. Search Console cannot separate them and I do not guess.
• One live read, one country, one day. Four queries on US mobile on 2026-10-06. Other countries and desktop could differ.
• Bing has no dates. Bing Webmaster reports 158 impressions and 37 pages in its index for this site, but the connector I use returns no window, so Bing could not be a second control.

What we have not done

We have not changed where the new posts are linked from; the pillar they all report to was last crawled on 2026-07-22, and our own tenth case study says a link from an uncrawled page passes nothing. We have not requested indexing for any of them. We have not reviewed our own publishing routine against Google's spam policies since the update; that review is open. The next 28-day window closes on 2026-10-26, and this post gets a dated follow-up when it does, whether the line recovers or not.

How to run this on your own site

1. Pull the daily series, not the compare. Performance, Search results, last 3 months, Date view. Find the day position moved and the day impressions moved. Same day: one event. Different days: two, and read each on its own.
2. Split the position day by country. Countries tab, the window before the day against the window after. Rows that drop from position 40 and beyond to nothing, with zero clicks, are the tail ending. Impressions per day holding is the confirmation.
3. Split the impressions day by page. Pages tab, two weeks before against the days after. If the pages that earned your clicks lost 80% or more of their impressions, the loss is real whatever position says.
4. Run the four external checks.The Data Anomalies page for a logged glitch; a second property over the same days; the Search status dashboard for a dated update; URL Inspection on your biggest losers for "Submitted and indexed" and a recent fetch.
5. Search the head query yourself. If you are in the top ten, the report is noise. If you are absent where the report says position 8, the report is averaging the days before the loss, and the loss is real.

If you want the serving side of a page checked while you are at it, the free triage report runs eight checks from the URL alone. The query table that could not help here is the subject of the queries-do-not-add-up post, and the rest of these measured reads are in the AI-assistant SEO series.

FAQ

Why did my impressions drop while my average position improved?

Because average position is averaged over the searches that produced an impression. If the impressions you lost sat at deep positions, the ones that remain are the well-ranked ones and the average moves up. That is harmless when impressions per day hold and the lost rows were at positions 40 and beyond. It is not harmless when the impressions at positions 5 to 10 go too. On traffictriage.com both happened a month apart, and the 28-day compare showed one blended number.

How do I tell a Search Console reporting change from a real loss?

Four checks in order: the Data Anomalies page for a logged glitch, a second property you own over the same days, the Search status dashboard for a ranking update, and the live results page for a query where your reported position is still good. A reporting change moves every property at once and leaves you on the results page. A real loss moves one property and takes you off it. On 2026-10-06 our second property was flat and we were absent from four top tens.

Is the September 2025 reporting change still the reason for impression drops?

It explains a one-time step in September 2025, when Google stopped counting the deep-position impressions that tools generated. It does not explain a drop a year later on one property while another property on the same account holds steady. Check the dates and the control before reusing that explanation.

Should I act on an impressions drop if clicks did not change?

Not before you know which impressions went. Pull the daily series by country and by page. If the lost rows were at deep positions with zero clicks, there is nothing to fix. If the pages that earned your clicks lost their impressions and you are no longer on the results page for their queries, you have a ranking problem that clicks will follow within weeks.

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.

Sign up freeSee how it works