Skip to content
Evergreen diagnosis · Published

Search Console impressions drop: Google bug or real decline?

Between num=100, an almost year-long logging error and the disappearance of FAQ rich results, most of the impression drops you are seeing today are not a real decline. This article brings the decision tree, the dated table of breaks, and the protocol for comparing periods without fooling yourself.

By Alejandro Verjel · Founder of Rankiamos

Search Console impressions drop: first rule out measurement

If your clicks held steady and only impressions fell, it is almost always a change in how Google counts them — num=100, a dated logging error — or the loss of deep impressions that never generated clicks anyway, not a real decline. Confirm it with the three-question tree and the dated table of breaks below.

A Search Console impressions drop is one of the alerts that causes the most panic and deserves the least, especially between May 2025 and today. Over that stretch Google racked up several measurement breaks — some a deliberate product change, others an acknowledged bug — that alter what you see in the Performance report without your site having lost a single real visit. The problem is that almost no guide puts them all together: they talk about num=100 as if it were the only cause, or only mention the logging error, or only average position. This article brings all seven documented breaks into one table with the date, what each one contaminates, and whether it touches clicks, and before that it gives you a three-question tree so you know which branch you are on in under five minutes.

Worth saying up front: not every impressions drop is innocent. When the real position of your queries also dropped, you have a real decline, and the right diagnosis runs through a different path, which we cover toward the end of this article. The way to avoid wasting time is always the same order: date first, then clicks, then position. That is exactly the tree that follows.

The three-question decision tree: date, clicks and position

Three questions, in this order, and do not skip any. Each one sends you down a different branch, and only one of the three ends at "you have a real problem".

  1. Question 1 · Does the drop start on a date from the breaks table?

    Look at the exact day the decline starts in the Performance chart and match it against the table in the next section (allow a day or two either way, for Search Console's usual reporting lag). If it lines up, mark that window with your own annotation and only compare periods that fall on the same side of the break. If the date does not match any of them, move to question 2.

  2. Question 2 · Did clicks fall in the same proportion as impressions?

    Turn on all four metrics and compare the percentage drop in clicks against the drop in impressions. If clicks stayed stable (or fell much less) while impressions collapsed, export the affected queries and check whether they came from deep positions — rank 11 and below, num=100 territory — or from an appearance like FAQ that stopped showing. If so, that is a measurement change or the loss of impressions that never drove clicks anyway, not a real decline: that is a different case from clicks dropping while position holds steady, which has its own diagnosis. If clicks did fall by a similar proportion, move to question 3.

  3. Question 3 · Did position drop for those same queries?

    Check the average position of the queries where both clicks and impressions fell together. If position held steady, the problem is CTR or a change in the SERP (a new block, a new competitor, an AI summary now sitting where you used to). If position also dropped, this is not a measurement problem: it is a real drop in Google rankings, and that is where the right triage begins.

A hypothesis we do not yet have our own data to confirm, but worth keeping in mind: on young domains, where nearly all historical impressions come from deep positions (11 and below) simply because the site is just starting to rank, the num=100 effect shows up proportionally larger than on an established domain where most queries already sit in the top 10. If you run a newer site and the September 2025 drop looked more dramatic than what larger domains you follow reported, that is a reasonable explanation.

For question 2, exporting queries goes faster if you know exactly what to look for: on the Queries tab, sort by lost impressions (not clicks) and check the average position column for the period before the drop. An average position of 15 or worse across most of the affected queries is the typical num=100 fingerprint; an average position of 3 to 8 among the queries that dropped points more toward a real SERP change or the FAQ appearance.

The dated breaks that skew your comparisons (2025-2026)

This table is the backbone of the whole article. The primary source is Google's official "Data anomalies in Search Console" help page, checked on September 10, 2026; the num=100 row comes from industry reports, not that page, because it was not a logging error but a product change.

Breaks that skew your Search Console comparisons, 2025-2026
Date What happened What it contaminates Affects clicks? Source
May 13, 2025 – Apr 27, 2026 Logging error: Search Console did not count impressions accurately (notice dated Apr 3, 2026) Performance impressions, and with them CTR and average position No, per Google Data Anomalies
Sep 10-14, 2025 (approx.) Google stops honoring the num=100 parameter Impressions from deep positions, unique queries and average position (apparent improvement) Not directly Industry reports and the LOCOMOTIVE study — not on the Data Anomalies page, since it was a Google change, not a bug
Feb 28 and Mar 1, 2026 Two days missing from bulk BigQuery export, unrecoverable Bulk data export N/A Data Anomalies
Apr 16-27, 2026 Impressions and clicks for job-posting appearances were not reported Performance, job-posting appearance Yes, in that appearance Data Anomalies
Since May 7, 2026 FAQ rich results stop showing in Google Impressions from the FAQ appearance Real SERP change, not a logging error Data Anomalies
May 7-8, May 21, Jun 24 and Aug 13, 2026 Logging errors in Discover Discover clicks and impressions Yes, in Discover Data Anomalies
Aug 13-17, 2026 Logging error in the Search AI-generative report; data restored Impressions in that report N/A Data Anomalies (read in our AI Overviews citations post)

One row left out on purpose: some SERP watchers reported a brief drop in Performance impressions and clicks on August 12-13, 2026, which John Mueller confirmed and which was fixed days later. It is not on the official Data Anomalies page, so if you mention it in a report, cite the source that flagged it and the date — do not treat it with the same certainty as the seven rows above.

The practical takeaway from this table is not to memorize seven dates: it is to understand that the more breaks fall inside the range you are comparing, the less trustworthy that impressions comparison is. The one thing that does not depend on how Google decides to count that month is a position history measured under the same conditions every time, independent of Search Console's product changes.

Search Console and num=100: why impressions nobody was seeing disappeared

The num=100 parameter asked Google to return a hundred organic results on a single page instead of the usual ten. Almost no human user relied on it — nobody scrolls to result 90 of a search — but SEO tools used it heavily: rank trackers, third-party scrapers and browser extensions that needed to read a keyword's whole ranking depth in one request. Between September 10 and 14, 2025, according to industry reports, Google stopped honoring the parameter: anyone requesting it today gets the same first page of results as a regular searcher. Google num 100 went from a technical curiosity to something dozens of SEO articles were writing about within days.

The impact on Search Console was immediate and massive, because every automated query made with num=100 active had been logging a real impression on whatever property showed up, even at position 87. An analysis of 319 properties by Tyler Gargula, of LOCOMOTIVE agency, found that 87.7% of the sites reviewed showed an impressions drop after the change, and 77.6% lost unique queries visible in the Performance report — queries that used to show a single impression at position 60 and now simply do not register at all. The study also flags something uncomfortable for the whole rank-tracking industry: a meaningful share of those impressions were never seen by a human searcher. They came from the measurement tools themselves, Rankiamos included, which queries the SERP through DataForSEO. We are not saying Rankiamos never contributed an impression to your Search Console; we are saying there is no public way to measure how much, and assuming it was zero would be just as unrigorous as assuming it was everything.

A side effect of num=100 almost nobody connects to its real cause is the sudden "improvement" in average position many sites saw in September 2025: once thousands of deep-position impressions dropped out of the calculation, the average climbs even though no query moved a single spot. It is arithmetic, not an SEO win, and we walk through it carefully — including the two paradoxes of averaging positions that different — in the guide to Search Console average position versus real position.

The Search Console impressions error (May 2025 – April 2026): what Google says, and what it does not

This is the longest break of the seven, and the one with the most room for confusion, since it spans almost a full year. The official notice, published on the Data Anomalies page on April 3, 2026, explains that a logging error kept Search Console from accurately counting impressions between May 13, 2025 and April 27, 2026. Google's wording is explicit on one point and deliberately silent on another: it confirms that "clicks were not affected", but it does not say whether impressions in that window ended up above or below what they should have been. This article holds to that same restraint: you will not find a claim here about whether your impressions from that year were inflated or under-counted, because Google did not make that claim either.

What is certain, and follows directly from how the metrics are calculated, is that CTR and average position for that window are also contaminated, since both derive from logged impressions. If you compare June 2025's CTR against June 2026's to "confirm" your content improved, you are comparing a number touched by the error against one that is not: the comparison has no footing. Clicks, by contrast, are comparable within that same window, because they are the only metric Google confirms the error left alone.

One number worth recalculating every time this article is read, since it shifts with the calendar: Search Console only retains 16 months of visible history, so a shrinking share of that window falls inside the error as months pass. As of this publish date (Oct 23, 2026), of the roughly 487 days the date picker shows, about 308 fall inside the error window — close to 63%. Put differently: nearly two-thirds of what you see today in your 16-month impressions chart covers a period Google itself flags as affected.

The safest way to work with this window is neither to ignore it nor to write off the whole year and a half of data: mark it with an annotation on the Performance chart (Search Console has allowed these since November 2025), limit impressions comparisons to periods that fall entirely on one side — before May 13, 2025 or after April 27, 2026 — and lean on clicks as your reference metric in the meantime.

FAQ, Discover and AI-generative: breaks that touch only part of your data

Not every break in the table is a logging error. The disappearance of FAQ rich results, in effect since May 7, 2026, is a real change to how the result looks in Google, not a measurement failure: if your content used to show that expandable question-and-answer block and it stopped, the impressions tied to that appearance fall because the block no longer exists on screen, not because Search Console stopped counting them correctly. To isolate this effect from a measurement break, filter the Performance report by "Search appearance" and compare only the FAQ category against the rest of your results: if the drop concentrates there, you have confirmed the cause.

Discover has its own error history, separate from the Performance report: point-in-time logging failures on May 7 and 8, May 21, June 24 and August 13, 2026. If you run a news site or content that leans on Discover traffic, these dates matter far more than the regular Performance report ones; if your traffic is mostly traditional search, treat them as a footnote.

The Search AI-generative report — which shows whether your pages were cited in Google's AI summaries — had its own logging error between August 13 and 17, 2026, with the data restored afterward. We do not develop that here; it deserves its own analysis, so if you want to know how to read that report and what questions it does and does not answer, that lives in our separate guide to tracking your site's AI Overview citations.

Comparing periods in Search Console without hitting contaminated windows

Comparing periods in Search Console with all these breaks active takes a protocol, not just good intentions. This is the one we use ourselves and the one we recommend to anyone who has to report results to a client or a boss using data from this past year and a half.

  1. Step 1 · Always start with clicks

    Clicks are the one metric none of the breaks in the table are confirmed to contaminate. If the goal is to show organic traffic grew (or fell), the conversation often starts and ends there, no need to touch impressions.

  2. Step 2 · Mark each break with your own annotation

    Since November 2025, Search Console lets you annotate the Performance chart with your own notes. Use the seven dates in the table above to record each one once; it saves you from re-explaining the same break every time someone looks at the chart.

  3. Step 3 · Use weekly or monthly view, not daily

    Since December 2025, Search Console offers weekly and monthly aggregation alongside daily. Daily view amplifies any site's normal noise; weekly or monthly smooths those swings and makes it much easier to spot a real break versus a one-day blip.

  4. Step 4 · Compare only windows of equal length on the same side of each break

    If you are comparing impressions, both periods need the same duration and need to fall entirely before or entirely after every relevant break. A period running from April 2026 to June 2026 crosses the end of the logging error midway through: it is not usable, not for impressions and not for average position.

  5. Step 5 · Segment by query, page, country and device

    A site-wide break can hide the fact that only one section was affected (for example, if you lost the FAQ block only on certain pages). Segmenting tells you whether the problem is general or localized, and that changes what you report and to whom.

  6. Step 6 · Only compare impressions inside one clean window

    The final rule: never compare impressions from a period that touches a break against one that does not. If you do not have two clean windows on the same side to compare, it is better to say "there is not enough clean history yet" than to force a comparison that means nothing.

This protocol is also, at bottom, the reason a rank history you own matters so much in this context: while Search Console rebuilds how it counts, a position series you measure yourself, under the same conditions every time, does not depend on any of these seven breaks.

Comparing year over year in Search Console: when it works and when it does not

Comparing year over year in Search Console is one of the most used features of the Performance report, and today it is also one of the easiest to misuse. For clicks, the year-over-year comparison still holds without conditions, since Google confirms the logging error did not touch them. For impressions, CTR and average position, the comparison only holds if neither of the two periods you are crossing touches the May 13, 2025 to April 27, 2026 window.

In practice, that means the first fully clean year-over-year impressions comparison becomes available starting late April 2027: only then can you put, say, May 2027 against May 2026 knowing both months sit outside the error. If your site used to show FAQ results that disappeared on May 7, 2026, the first clean year-over-year comparison for that specific appearance lands a month later too — May 2027 against May 2026 — since you need both sides of the year to reflect the same product reality.

And there is a structural limit rarely mentioned alongside all this: Search Console only retains 16 months of data, so the real window for a year-over-year comparison (which needs at least 12 months on either side) was always narrower than most people assume, even setting aside this year and a half of breaks. The details of that retention — and how to save your own history before Search Console trims it — live in the guide to Search Console's 16-month data limit.

When the drop is real: the signals that do not depend on how Google counts

Everything above is meant to rule out noise, not to make you look away from a genuine decline. The most reliable sign that you have a real problem, not a measurement break, is simple to state: clicks and position dropped together, on the same queries, on dates that do not appear in this article's breaks table. If that describes your case, here are three possible causes, each with its own treatment we will not repeat here since it already has its own guide.

  • General triage of a ranking drop: if you are not sure where to start, the full process for figuring out why your Google ranking dropped is the right starting point.
  • A Google core update: if the drop lines up with a confirmed algorithm update, and your competitors moved during the same weeks, a core update is a likely cause. We do not develop that here; it is a different diagnosis with its own guide.
  • A domain or platform migration: if you changed domains, switched CMS, or ran a large URL restructuring around the date of the drop, that is almost always the more likely explanation before you go looking for a cause inside Google. We only mention it here; the full treatment of migrating without losing rankings is a separate article.

What the three have in common is that none of them get solved by staring harder at the Performance report: they get solved by looking at what changed on your site, in the algorithm, or in the SERP during those specific weeks, with the real position of your queries as the main evidence.

Your control series for the next break

There will be another measurement break. The question is whether you will have something to compare it against.

If the seven breaks in this article teach anything, it is that Search Console is going to keep changing how it counts, and it will probably do it again without the warning you would like. The one thing that does not change with each Google product update is a position series you measure yourself, under fixed conditions, starting the day you add each keyword. And here is the line that holds up the whole method: if you were not tracking before September 2025, your series is not a control for that break; it is a control from the day you start, and for the next one.

Rankiamos tracks your Google.com position under conditions that do not change month to month — same country or city, same device, same frequency — and comes with Search Console connected from the Free plan so you can read both reports side by side: what Google reports on impressions and clicks, next to what you measure directly, independent of how Search Console decides to count that month. The Pro plan's daily check is what lets you date the next break precisely — not the seven that already happened, those are already in the table above.

How Rankiamos handles it

Your own series, fixed conditions

Same city or country, same device and the same frequency every time: weekly on Free, daily on Pro. It does not depend on how Google counts impressions that month, because it does not measure impressions: it measures where your URL shows up in the SERP, down to position 30.

Search Console connected from Free

Up to 250 queries from the last 28 days, with clicks, impressions, CTR and average position per query, including a "Top queries" view with the 50 highest-impression ones. Seeing it next to your measured position is what lets you notice at a glance if one fell and the other did not.

Daily checks and manual refresh on Pro

Re-measuring a keyword you already tracked by hand is a Pro feature: every 6 hours, with 30 units a day. On Free, measurement is weekly and automatic, and the first measurement of each keyword goes out instantly when you add it. The Pro plan's daily automatic check is what lets you date the next break to the day, not the week.

Competitor history

Each keyword's top 10 (on both plans) with 45 days of history helps you settle question 3 of the decision tree: whether the whole SERP moved (new competitors, new blocks) or the change was yours alone.

What Rankiamos does NOT do about these breaks

Rankiamos history is not retroactive: it starts the day you add the keyword, so it is no control for breaks that already happened — num=100, the May 2025 logging error — if you were not already tracking before them. It works from today and for the next one. Rankiamos also queries the SERP via DataForSEO, so we will not tell you our own measurements never generated one of your Search Console impressions: there is no public way to prove it either way. On the Free plan the check is weekly and on a single device, so a break gets dated with up to a week of margin, not to the day. And Rankiamos does not replace Google's official Data Anomalies page, nor does it detect Google's own logging errors: for that, the official page is still the source to check first.

The full breakdown of what each plan measures is on the Rankiamos features page, and each plan's quotas and price on pricing.

Try Rankiamos free

Free plan: 2 projects, 6 keywords, weekly check · iPhone · iPad · Apple Watch

Frequently asked questions

Why did my impressions drop in Search Console if clicks stayed the same?
Chances are you have not lost any real traffic. There are at least three documented breaks that explain an impressions drop with no click drop: the num=100 change in September 2025, which pulled deep-position impressions (rank 11 and below) out of the count; the impression logging error between May 13, 2025 and April 27, 2026, which Google says did not touch clicks; and the disappearance of FAQ rich results starting May 7, 2026. To confirm it in five minutes: check that clicks stayed flat over the same date range, then export the affected queries and see whether the lost impressions came from deep positions (11-30) or from that FAQ appearance. If clicks also fell, this is not a measurement issue, and the right move is to check the position of those same queries before drawing any conclusions.
What was the num=100 change and why did it affect Search Console?
num=100 was a URL parameter that asked Google to return a hundred results on a single page instead of the usual ten. SEO tools, rank trackers and scrapers used it to pull an entire keyword's ranking depth in one request. Between September 10 and 14, 2025, according to industry reports, Google stopped honoring it: anyone requesting it today just gets the first page like any regular user. The effect on Search Console was immediate, because every one of those automated queries had been logging as a real impression on the site's property. An analysis of 319 properties by Tyler Gargula, of LOCOMOTIVE agency, found that 87.7% of the sites reviewed showed an impressions drop after the change, and 77.6% lost unique queries visible in the report. The uncomfortable part the study itself points out: some of those impressions never came from a human searcher. They came from tools — Rankiamos included, which queries the SERP through DataForSEO — requesting deep positions no user ever scrolls to see. It does not appear on Google's official Data Anomalies page because it was not a logging error: it was a Google product decision.
Did Google fix the impressions from the registration error between May 2025 and April 2026?
The official notice, published on Google's Data Anomalies page on April 3, 2026, explains that a logging error kept Search Console from accurately counting impressions between May 13, 2025 and April 27, 2026, and states plainly that "clicks were not affected". The notice does not say whether impressions in that window ended up inflated or under-counted, so this article does not claim that either — stick to what Google actually published. What is certain is that CTR and average position for that window are also compromised, since both are derived from impressions. Clicks, on the other hand, are comparable within that same window, because they are the one metric Google confirms the error did not touch.
Where do I check the official data errors for Search Console?
On the "Data anomalies in Search Console" help page (inside Google's Help Center), which lists every known logging error by date and which report it affects: Performance, Discover, the AI-generative report, bulk export. Checking it before reporting a drop to a client saves you more than one awkward conversation. One detail worth knowing: the num=100 change does not appear on that page, because it was not a logging error — it was a deliberate change in how Google delivers search results. Since November 2025, Search Console also lets you add your own annotations to the Performance chart: mark each break there the first time you spot it, and anyone who opens the property afterward — you, a teammate, a client — will see it without needing another explanation.
Why did my average position improve in September 2025 without me doing anything?
Because what goes into the average changed, not your actual ranking. When Google stopped honoring num=100, huge numbers of deep-position impressions (rank 11 and below) dropped out of the count; once those low numbers leave the calculation, the average climbs even though no query actually moved a single spot. To check it on your own account, look at the clicks and position of queries that were already in your top 10 before that date: if they did not move, the "improvement" is arithmetic, not an SEO win. The full arithmetic behind average position — and why it sometimes contradicts what you see with your own eyes in Google — is covered in the guide on Search Console average position versus real position.
Can I still compare year over year in Search Console after all these changes?
Year-over-year clicks are still fine, because Google confirms the logging error did not touch them. Impressions, CTR and average position are not, as long as one of the two periods you are comparing falls inside the May 13, 2025 to April 27, 2026 window. In practice, the first fully clean year-over-year impressions comparison becomes possible starting late April 2027 (comparing, say, May 2027 against May 2026), and from May 2027 if your site used to show FAQ results that stopped appearing on May 7, 2026. Keep in mind too that Search Console only retains 16 months of history, so the real window for a clean year-over-year comparison is narrower than most people assume.
Were rank trackers inflating my impressions? Does Rankiamos also query Google?
It is a fair question and the honest answer has nuance. After the num=100 change, the hypothesis spread — backed by data from the LOCOMOTIVE study — that a chunk of the impressions that disappeared were never seen by a real user: they were generated by scrapers and tools requesting a hundred results per query. Rankiamos also queries Google's SERP, through DataForSEO, under fixed conditions and without requesting num=100 (weekly on the Free plan, daily on Pro). There is no public way to measure how much any tool — ours included — contributed to your historical impressions, so we are not going to tell you it was zero. That is why the practical rule in this article is to look at clicks first, since they are confirmed as the clean metric, and to use your own position history as the control series instead of trying to explain impressions that already happened.

14 days of Pro, free

Try Rankiamos free for 14 days.

The Free plan needs no card. The Pro trial runs through your Apple account; cancel before day 14 and you pay nothing.

Download Rankiamos on the App Store

Requires iOS 17.6+ · iPadOS 17.6+ · watchOS 10.6+