On Monday morning there is a step in the chart where last week there was a line. The reflex is the same everywhere: change something, now. Rewrite the titles, pad the copy, install a plugin that promises improvement, rebuild half the site if that is what it takes. And with that the diagnosis is over before it started, because from here on nobody can separate the drop from the reaction to it.
This piece walks the diagnosis in five stages, from "is the drop even real" to "a competitor simply got better". The rule that goes with it: you touch nothing until one stage has produced an answer. The centrepiece is a table that reads clicks, impressions and position against each other, because those three numbers together almost always tell you which kind of problem you have.
Stage zero: is the drop even real
A surprising share of all ranking drops is a reading error. Four things have to be ruled out before you look any further.
Search Console lags behind. The last two to three days are always incomplete, and because the chart draws them anyway, the line always falls off at the right edge. Anyone checking the last data point every morning sees a crash every morning. Cover the last three days mentally and what remains is the actual curve.
A rank tracker is not evidence. It checks one position, in one location, on one device. Google serves different results depending on place, device, language and history. Weakest of all is searching for yourself in your own browser: you have been clicking your own site for years, and Google knows it.
The comparison window has to match on weekdays. Compare 28 days against the 28 before, not month against month. Otherwise a month with five Mondays beats one with four, and in business-to-business that is a double-digit percentage on its own. For seasonal topics add the year-over-year view, because a heating page always does worse in July than in October.
Look in Search Console, not in analytics. Analytics only counts whoever accepted the cookie banner. A new banner or a changed default looks exactly like a traffic drop there, and is not one. How large that gap is on your site is covered in consent mode and your GA4 data.
The three numbers that already hold the diagnosis
The performance report shows clicks, impressions, CTR and average position. Switch all four boxes on and read them together. The combination tells you which stage you need next.
| Pattern | likely cause | continue at |
|---|---|---|
| Position steady, impressions and clicks down | demand fell, your ranking did not | seasonality, year-over-year |
| Position down on a few queries, the rest steady | a competitor, or a changed results page | stage five |
| Position and impressions down across all queries, sharp step on one date | update or technical defect | stages two and three |
| Clicks down, impressions and position steady | the results page changed, not your placement | AI Overviews, snippet |
| Everything down, starting exactly on the day of a deploy | it was you | stage two |
The first row is the most common case and the most misread. An accountant loses a third of their clicks in July because nobody searches for tax return help in July. Position 4.1 before, position 4.0 after. Nothing is broken there, it is just summer. Whoever starts rewriting pages in response spends the autumn changing exactly the pages that were working.
The fourth row is the newer variant, rare three years ago. Your position holds, impressions hold, clicks fall anyway. Then something has slotted in above you that already delivers the answer: an AI Overview, a featured snippet, a map pack, one more ad row. The impression still counts, which is why it looks like a CTR collapse. What you can do about it is in Google AI Overviews and in impressions but no clicks.
One more distinction that saves a lot of panic: check whether the drop hits all queries or only some. Sort the report by click difference and see whether two rows sit at the top or two hundred. Two rows are a competitive matter. Two hundred rows on a single day are an update or a defect.
Stage two: was it you
This stage comes before the technology and before Google, because it is the most common real cause and can be checked in ten minutes. The suspects, by frequency:
- a deploy or a relaunch, including a small one
- a theme or plugin update that emits meta tags differently
- a
noindexthat travelled along from the staging environment - changed canonical tags that suddenly all point at the home page
- a new redirect chain with two hops in a row
- a rebuilt navigation that dropped internal links to important pages
- deleted or merged pages with no redirect
- a new cookie banner that loads the content later than before
- a bot shield or a new firewall rule at the host
If nobody wrote down what went live when, the deploy history, the CMS revision log or the repository will. For the page itself use the URL inspection tool in Search Console: it shows what Google last saw, which canonical applies and whether the page is indexable. One look at a single affected page exposes half of these causes.
If the drop coincides with a relaunch, the cause is almost always the redirect structure. The order of work for that is in relaunch without losing traffic.
Stage three: technical causes
Only now the technology, and in this order:
Indexing. Are the affected pages still in the index at all? The page indexing report shows shifts between the status classes, and a page that has fallen out of the index does not rank. What those messages mean is in page not indexed.
Server errors and response times. The crawl stats report shows what Googlebot got when it fetched. A cluster of 5xx responses, or response times suddenly measured in seconds instead of milliseconds, throttles crawling and everything downstream of it.
Blocks nobody ordered. An expired certificate, a firewall rule, a bot shield at the CDN that turns Googlebot away too. The last one has become more common since many hosts ship protection against AI crawlers as a default.
Load time. Honestly placed: Core Web Vitals are a tiebreaker, not a cliff. A page does not fall from position 3 to position 20 over 0.3 seconds. But if load time jumped from two seconds to nine because an image module landed in between, that is more than a tiebreaker. What actually moves the numbers is in improving Core Web Vitals.
Stage four: a Google update
Recognition marks: the step sits on one date, hits many queries at once, and it does not only hit you but your competitors too. Google announces the larger updates publicly and documents when the rollout starts and ends. A rollout often stretches over weeks, which is why things wobble in both directions during that time.
The thing to rule out first, because it causes the most fear and applies the least often: a manual action. If there is one, it sits in the security and manual actions report, in plain words. If nothing is there, you have no penalty. Then an algorithm has assessed your site differently, and that is a difference like the one between a fine and a worse place in the queue.
The clear position on this: the worst thing you can do after an update is to rebuild everything at once. Afterwards you never know what worked, and at the next update you stand there without knowledge again. Recovery takes weeks to months and often arrives with the next update, because the assessment gets recalculated. What makes sense in the meantime is the unspectacular work: take the ten pages with the largest loss and answer, for each, the uncomfortable question of what a reader gets there that they do not get on five other pages. If the answer comes out thin, you have found your task. The way through it is in updating existing content.
Stage five: someone got better
The least dramatic cause and, on individual queries, the most common one. No defect, no update, just a competitor who put a better page on exactly this topic.
The check takes five minutes: search the affected query in a window where you are not signed in, look at the two results now sitting above you, and answer in one sentence what they do better. Usually the answer is specific and uncomfortable: prices are stated, and you hide yours. A question is answered that you only mention. There are photos of real projects instead of stock imagery. Rarely is the answer "they have more backlinks".
Why a change log halves the diagnosis
Everything in stage two becomes trivial if there is a record of what went live when. Two lines per change are enough: the date and what happened. Whoever keeps that diagnoses in ten minutes what otherwise costs a week of guessing.
The second part is noticing the drop at all. Most people find it four weeks later, when the cause can no longer be reconstructed. Traffalyzer takes over that mechanical part: it compares the periods continuously and flags the step when it happens, instead of when you next open the report. You still have to walk the five stages yourself.
Where to start
- Cover the last three days in the chart and compare 28 days against the 28 before.
- Switch on all four metrics in the performance report and find your pattern in the table above.
- Sort by click difference: two affected rows, or two hundred?
- Ask about deploys, updates and plugin changes inside the window of the step.
- Run one affected page through URL inspection: indexable, right canonical, recently crawled.
- Only if one through five turn up nothing: check whether an update was rolling out.
- Change one thing, write down the date, and wait at least two weeks.
Frequently asked questions
Why did my Google ranking suddenly drop?
In this order of frequency: a change of your own at the start of the window, fallen demand, a technical defect, a Google update, a better competitor. The fastest route to the answer is not Google but your own deploy history. And before that, the question of whether the drop is real: the last days in Search Console are always incomplete and therefore always look like a crash.
How do I find out whether a Google update hit me?
Three marks have to coincide: the step sits on one date, it hits many different queries at once, and it also hits other sites in your field. Google publishes the larger updates with a start and end date for the rollout. If your drop only affects two queries, it was not an update, whatever the forums are saying that week.
How long does it take for rankings to recover?
For a change of your own that you revert: days to a few weeks, once Google has recrawled the pages. Same for a technical defect. After an update it takes weeks to months, and recovery often coincides with the next update. Google states no deadline for this, and any agency guaranteeing one is telling you something it cannot know.
Why do my rankings fluctuate daily?
Because there is no single position. Google serves different results by location, device, language and history, and continuously tests variants on top of that. An average position of 6.2 does not mean you sit at position 6; it means the positions you actually appeared at average out at 6.2. Swings of one or two positions are noise, not movement.
My traffic fell but my position did not. What happened?
Then the results page changed, not your placement. Something slotted in above you that already gives the answer: an AI Overview, a featured snippet, a local map pack, one more ad. Second possibility: Google rewrote your title or snippet and the new version sells worse. Third possibility, if you are looking at analytics rather than Search Console: your cookie banner changed.
How do I know whether Google penalised me?
By an entry in the manual actions report. If nothing is there, you have none. The word penalty gets used for algorithmic reassessments too, and that misleads: a manual action is a human decision with a stated reason and a way back through a reconsideration request. An algorithmic reassessment has neither.
We had a relaunch and the traffic went with it. What now?
Almost always the redirects. Check a handful of the old URLs that had the most clicks before: do they land on the matching new page through a single permanent redirect, or on the home page, or on a chain of three hops, or on a 404? The order of repair is in relaunch without losing traffic.
Should I have the copy rewritten after a drop?
Not as the first measure, and never all of it at once. As long as you do not know whether it was seasonality, a defect or an update, new copy is an expensive bet on a cause you have not identified. If it does turn out that the assessment of your content fell, work on ten pages and wait for the result rather than on a hundred.
How often should I be looking at all?
Weekly is enough for monitoring; daily invites overreaction. Two exceptions: after a deploy or relaunch you watch closely for three days, and when an alert about a technical defect arrives, immediately. Everything else is noise, and noise is easier to read at a weekly rhythm than a daily one.