A traffic drop is a fall in visits compared with a reference period. Diagnosing it means working out which number moved first: impressions, clicks or average position. The causes of a fall in impressions and those of a fall in clicks almost never overlap, and looking for them in the wrong order wastes weeks.
On 26 September 2026 I looked at the Google UK results for "website traffic drop". At the top sits an AI Overview listing five causes: zero-click searches, algorithm updates, technical issues, social media changes and seasonality. Below it come two LinkedIn posts, a Reddit thread, a guide from Plausible and articles from agencies such as Aira ("11 reasons"), TU Marketing, Exposure Ninja and Digital Wonderlab (DataForSEO, Google UK, September 2026). The first page is built around lists of causes. It does not lead with the order in which to check them, or with the exact dates of Google's latest updates, the ones to line the day of the drop up against. And several of the figures on the page, starting with the AI Overview's own "over 68% of Google searches now end without a click", come without a link to the original measurement.
Here you will find those three things: the order of the checks, a verifiable calendar of updates and the Search Console data of a real site.
Look at which number moved, not at the list of causes
The starting point is the Performance report in Google Search Console, with all four metrics switched on: clicks, impressions, CTR and average position. Three things to do straight away, before any hypothesis.
- Extend the date range to 16 months. It is the maximum Search Console keeps, and it tells you whether what you are looking at is a collapse or the end of a spike.
- Compare with the same period last year, not last month.
- Split by search type (Web, Image, Video, News) and by device. A drop that exists only on desktop tells a different story from a drop that is everywhere.
The Google Search Central documentation on debugging drops in search traffic is clear on a point that gets little attention: "impressions and clicks are ultimately the measure of success for your site". Average position on its own decides nothing, and in 2026 it can improve while clicks fall.
This is the table I use to assign a drop to its family of causes before opening any other tool.
| Impressions | Clicks | Average position | Family of causes | First check |
|---|---|---|---|---|
| Stable or growing | Falling | Stable or improving | The results page is keeping the clicks: AI Overview, featured snippet, local pack | CTR by query and by device, year-on-year comparison |
| Falling | Falling | Stable | Demand has fallen: seasonality or interest moving elsewhere | Google Trends on the top ten queries |
| Falling | Falling | Worsening across many pages | Algorithm update or competitors gaining positions | Line up the date of the drop with the update calendar |
| Collapse within one or two days | Collapse | Erratic or missing | Technical or indexing problem | robots.txt, noindex tag, server status, Page indexing report |
| Drop on one group of URLs only | Same | Same | Site change, migration, cannibalisation | Compare pages one by one, not the total |
| Flat, then zero | Zero | None | Broken tracking or data anomaly | Search Console against Analytics, Data anomalies report |
Impressions up, clicks flat: now the most common case we see
This is the row that throws meetings into confusion, because every work indicator improves and the traffic does not. These are the Search Console figures of an Italian outdoor equipment online shop we work with, traffic from Google Italy, same quarter compared year on year.
| Device | Impressions 2025 | Impressions 2026 | Clicks 2025 | Clicks 2026 | CTR 2025 | CTR 2026 |
|---|---|---|---|---|---|---|
| Desktop | 22,627 | 47,771 | 877 | 866 | 3.88% | 1.81% |
| Mobile | 57,986 | 90,714 | 1,753 | 2,128 | 3.02% | 2.35% |
| Total | 81,461 | 139,633 | 2,661 | 3,055 | 3.27% | 2.19% |
On desktop, impressions more than doubled, average position improved by more than four places and clicks fell from 877 to 866. Eleven fewer clicks after a year of work that went well. Desktop CTR halved.
The reason is how much space the generated answer takes above the results. In Italy, where this site competes, a SEOZoom study published on 2 December 2025 on around 30 million Italian keywords found an AI Overview above the fold in 54.15% of searches, against 11% in July 2025. In the same study, of the roughly 168 million domains visible on Google Italy, 175,000 are cited: 0.1%. And 40% of the cited links come from URLs that are not in the organic top 10. We have reconstructed how that selection works in AI Overviews ranking factors.
On the size of the fall in clicks there are two serious measurements, both from the US, and they should be read as such. Ahrefs compared 300,000 informational keywords, half with an AI Overview and half without, between March 2024 and March 2025: for the first organic position, desktop CTR falls by 34.5%. The Pew Research Center tracked the real behaviour of 900 US adults across 68,879 Google searches in March 2025: with an AI summary on the page, people clicked a traditional result in 8% of visits, without one in 15%. Links inside the summary got 1%. And the session ended there in 26% of cases with a summary, against 16% without.
These are US figures on a US sample. Equivalent public CTR data for most other markets does not exist yet, and that is why the Search Console table above is worth more than any imported percentage: your own site is already the measurement. For the full picture of what changes on the results page, see Google AI Overviews. On the paid side the effect looks different, and we measured it in Google Ads and AI Overviews.
If impressions fall, the problem may not be yours
Impressions falling with a steady average position means people are searching less. It happens because of seasonality, a product dropping out of the interest cycle, or a news story that has pulled attention elsewhere.
The check takes five minutes: take the ten queries with the most impressions, put them into Google Trends over five years and see whether the national curve looks like yours. If they fall together, the site has nothing to fix. If the national curve is flat and yours is not, you have lost visibility and the cause lies elsewhere.
There is a third possibility that often goes unnoticed: demand has not fallen, it has been rephrased. When the way people word a request changes, the old queries lose volume while others gain it, and the site stays anchored to the old ones. You see it by comparing new queries from the last three months with those from a year earlier, and it is the moment to redo your keyword research starting from the data you already have, checking that the search intent of your pages is still the right one.
If average position falls across the whole site, check the calendar
A drop that starts on a specific day and affects pages on different topics should be checked against the dates of Google's updates. If the collapse begins inside a rollout window, the algorithm hypothesis goes to the top of the list. If it begins two weeks after the end, you are almost certainly looking at something else.
| Update | Start | End | Duration |
|---|---|---|---|
| March 2025 core update | 13 March 2025 | 27 March 2025 | 14 days |
| June 2025 core update | 30 June 2025 | 17 July 2025 | 17 days |
| December 2025 core update | 12 December 2025 | 29 December 2025 | 17 days |
| March 2026 core update | 27 March 2026 | 8 April 2026 | 12 days |
| May 2026 core update | 21 May 2026 | 2 June 2026 | 12 days |
| August 2026 spam update | 18 August 2026 | 21 August 2026 | 2.5 days |
The last one on the list concerns anyone reading this with an August drop in hand. The August 2026 spam update started on 18 August and finished on the 21st: two and a half days, declared by Google to apply to all languages and all countries. A drop that begins on 19 or 20 August and has not moved since has a clear suspect.
There is also a measure of how much the results move during a core update, from the Italian market. In SEOZoom's analysis of the December 2025 core update, volatility in Italy reached 7.17% in news and retail and stopped at 4.89% in health. The useful reading is not the list of winners and losers: it is that a core update redistributes visibility among sites that are already present, it does not punish across the board. If your sector had low volatility and you dropped anyway, the algorithm explanation holds up less well.
When the hypothesis does hold, the work that follows is not technical. It concerns ranking factors, the demonstrable quality of the pages and E-E-A-T signals. If over the last year you have published mass-produced content, it is worth rereading what the data says about AI content and rankings before looking for blame elsewhere.
A collapse in a single day is almost always technical
An algorithmic drop has the shape of a long step, spread over days. A technical drop is a wall. If the graph plunges between a Tuesday and a Wednesday, this is the order of checks.
- The robots.txt file. A
Disallow: /that has made its way into production from a staging environment is the most common single cause and the quickest to rule out. - The noindex tag. On WordPress the "Discourage search engines from indexing this site" box does the same damage with one click.
- Server status in the Crawl stats report: a spike in 5xx responses or timeouts blocks crawling before it touches rankings.
- The Page indexing report, comparing the number of indexed pages before and after the date of the collapse.
- Redirects and the certificate, if there was a migration or a change of domain in the preceding weeks.
One clarification that saves days: site speed rarely causes a sudden collapse. Slower loading times erode results gradually, and the subject is covered properly in the guide on what Google really measures with page speed. If the traffic disappeared within twenty-four hours, the cause is in the first three points of the list.
The drop that never happened
Some of the drops we are asked to analyse are not a drop in the site but a drop in the measurement. Four cases come up again and again: the Analytics tag removed during a theme update, a Tag Manager container published without the configuration tag, a new consent banner that stops collecting data before acceptance, a filter or view changed in GA4.
The test that exposes all of them in a minute: open Search Console and Analytics on the same date range. If organic clicks in Search Console are stable and organic sessions in Analytics have collapsed, the site is fine and the problem is in the tracking. Google also keeps a Data anomalies page for recording errors on its side, and it should be checked before opening any investigation.
The two platforms will never match anyway, because they count different things: Search Console counts clicks from Google Search, Analytics counts sessions for which consent was given. A steady gap is normal; a sudden widening of it is not. We explain what each one measures in the guide to Google Analytics. If you are not yet sure which numbers to track every month, the list is in SEO KPIs.
Manual actions and security issues can be ruled out in two minutes
They are the most feared cause and the rarest. Search Console has two dedicated reports, Manual actions and Security issues, and in both cases Google notifies the verified property by email. If they are empty, the hypothesis is closed and you do not go back to it.
When there is a manual action, the message says which policy was breached. The two we see most often on the sites we audit are unnatural links bought years earlier and text packed with repetitions on category pages, which is keyword stuffing even when a plugin produces it. A different case, which does not trigger penalties but dilutes rankings, is duplicate content across URL variants: there is nothing to remove there, only a choice of which version to keep.
How long recovery takes
It depends on the family of causes, and the timescales are very different.
- Technical problem fixed: days. The time Google needs to recrawl, which on a small site is a matter of a week.
- Tracking error: immediate, because there was nothing to recover.
- Falling demand: it cannot be recovered, only replaced by covering different queries.
- Core update: months. Recent rollouts lasted 12 to 17 days, but the recovery does not come within that window: it comes when the work on content is reassessed, often at the next update.
An order of magnitude for recovering lost rankings, with our own numbers. Macropix, a Milan-based maker of custom LED screens, went from position 88 to position 2 in Italy for the keyword "ledwall", 4,400 searches a month, reaching a 25.55% share of voice ahead of Amazon. The full story is in the case study on SEO for manufacturers, and it was not quick. On how long it takes in general, we have gathered the data in how long SEO takes to show results.
If instead the traffic never arrived in the first place, that is a different case, and you can recognise it with the checks that tell you whether SEO is working. If the diagnosis stalls because the tools or the time are missing, it is the kind of analysis we do in our SEO consultancy work, and the results measured for clients are among our projects.
Frequently asked questions about traffic drops
Under seven days it is better to wait: weekly fluctuations and bank holidays produce drops that recover on their own. The signal to take seriously is a drop lasting more than two consecutive weeks compared with the same period last year, not with the previous month.
Compare the exact day the graph starts to fall with the rollout windows announced by Google. The May 2026 core update ran from 21 May to 2 June, the March 2026 one from 27 March to 8 April, and the August 2026 spam update from 18 to 21 August. If the drop starts outside these windows and affects only some pages, the algorithm hypothesis is the least likely.
In Italy, a SEOZoom study from December 2025 on around 30 million keywords found an AI Overview above the fold in 54.15% of searches. The rigorously measured fall in clicks, however, comes from US studies: Ahrefs estimates a 34.5% lower CTR for position one on desktop. The way to check it on your own site is to see whether impressions hold while CTR falls, device by device.
They count different things. Search Console records clicks from Google Search, Analytics records sessions for which it obtained consent and the tag worked. A steady gap is normal. If the gap suddenly widens, the problem is in the measurement, not in the traffic.
Yes, but rarely within the rollout window, which in recent updates lasted between 12 and 17 days. Google reassesses pages at later updates, so recovery usually comes months later and only if the content has genuinely changed in the meantime.
One thing the industry rarely says: the right time to prepare for a drop is now, when there is none. Search Console keeps 16 months and then deletes, so whoever exports the queries and pages of the last 16 months today will have, when it is needed, a baseline that nobody can rebuild later. It is ten minutes of work that separate a diagnosis from a guess.
