Diagnostics · 12 min read
Organic traffic dropped: find the cause before you change anything
A working triage sequence. Confirm the drop is real, classify it by shape, segment it until it stops being a chart and becomes a specific loss, then work the branch that matches. Most declines are diagnosable in a day if you resist fixing things first.
Written by Zubair Afzal, FounderUpdated
First rule
Do not fix anything yet
The most expensive hour in a traffic decline is the first one, because that is when someone decides what happened and everyone else starts working from it. Once a theory is in the room, evidence gets collected to support it rather than to test it, and the actual cause goes unexamined for a fortnight.
Resist the two reflexes. The first is to roll back the most recent release, which destroys the timeline you need. The second is to blame a Google update, which is unfalsifiable, requires no action, and is wrong more often than it is right.
The sequence below is deliberately ordered. Each step either eliminates a class of cause or narrows the search, and the cheap eliminations come first. Worked properly, most declines are traced to a specific cause within a day.
Triage
The six-step diagnostic sequence
Work these in order. Do not skip ahead because a later step matches your theory — the point of the order is to eliminate the cheap explanations before you invest in an expensive one.
Confirm the drop is real
Before anything else, rule out measurement. Did a tag get removed from a template in a release? Did a consent banner change, or a default consent state? Was a filter, a bot exclusion or an internal traffic rule added? Did the reporting property, timezone or channel definition change? Cross-check the analytics decline against Search Console clicks, which are collected independently. If Search Console is flat and analytics has fallen, you have a tracking problem, not a search problem.
You get: A yes or no on whether visitors actually fell
Classify the shape
Plot daily clicks for at least six months. You are sorting the decline into one of three shapes: a cliff that falls within a day or two and stays down, a step that drops to a new level and holds, or a bleed that erodes a few per cent a week. The shape narrows the cause more efficiently than any tool, because a cliff implies an event with a date and a bleed implies an accumulation.
You get: A dated start point for the decline
Segment until it stops being a chart
Split the loss by brand versus non-brand query, by page template, by country, by device and by search appearance. Almost every decline concentrates somewhere. A drop confined to one template points at a code change; a drop confined to non-brand queries on commercial pages points at competition or relevance; a drop across everything at once points at indexing, tracking or a broad ranking change.
You get: The specific segment carrying the loss
Align the start date against known events
Lay the dated start point against your deploy log, CMS release history, redirect changes, tag manager versions, hosting or CDN changes, and the confirmed update dates on the Google Search Status Dashboard. Correlation is not proof, but a decline that begins on the morning of a release and is confined to the templates that release touched is not a coincidence either.
You get: A shortlist of candidate events
Check indexing and serving on the affected segment
For the pages carrying the loss: are they still indexed, and are they being served the same HTML to a crawler as to a browser? Inspect a live URL, check for a meta robots or canonical introduced by the release, review the robots.txt as it was on the day, look for a staging canonical or an X-Robots-Tag header, and confirm redirects still resolve in one hop. Check the manual actions and security reports while you are there.
You get: Confirmation of whether the pages can still be served
Form one hypothesis and test it
Write the cause as a sentence that could be wrong: "the June release added a canonical from every product variant to the category page, and product pages have been dropping out since." Then find the evidence that would disprove it. If you cannot state the cause in one falsifiable sentence, you do not have a diagnosis yet, and you are about to spend money on a guess.
You get: A written, testable cause
The same sequence, as branches
Three questions, each of which halves the search
Every answer eliminates a whole class of cause rather than one candidate. Worked in this order, most declines are narrowed to a named segment inside a morning.
Have Search Console clicks fallen as well?
Yes: Visitors genuinely fell. Move on to the shape of the decline.
No: Analytics fell alone — a tracking or consent change, not a search loss.
Did it fall inside a day or two and stay down?
Yes: A cliff. Look for a dated event: a release, a redirect, a block, a manual action.
No: A bleed. Look for accumulation: better-funded rivals, decay, coverage slipping away.
Is the loss confined to one template or query type?
Yes: Work out what that segment shares. The cause is almost always inside it.
No: Suspect something site-wide: indexing, serving, measurement, or a broad ranking change.
Traffic charts do not have causes. Segments do. Until a decline has been narrowed to a page template, a query type, a country or a device, every explanation on offer is a guess wearing a graph.
Classification
What the shape of the decline tells you
Two different investigations. Running the cliff playbook on a bleed wastes a week, and running the bleed playbook on a cliff misses an event you could have reversed.
| Dimension | Sudden cliff | Slow bleed |
|---|---|---|
| What it looks like | Traffic falls within a day or two and holds at the new level | Traffic erodes by a few per cent a week over two or three months |
| Usual causes | A deploy, a robots or canonical change, a migration with broken redirects, a manual action, a tracking break, a broad update landing hard | Competitors improving, content going stale, links decaying, buyer intent shifting, results-page features absorbing the clicks |
| First thing to check | The deploy log, and the indexing report filtered to the affected template | Query-level impressions and average position, split brand and non-brand over twelve months |
| Typical time to diagnose | Hours — the cause is an event with a date on it | Days, because several small effects have to be separated from each other |
| Most common misdiagnosis | Blaming an update that landed a week before the actual start date | Blaming an update because nobody kept a change log and the decline has no visible start |
| What recovery looks like | Often fast once the cause is reversed, unless the site was restructured on top of it | Gradual, and only if the underlying reason is addressed rather than the symptom |
Frequently missed
Six causes that get overlooked every time
- The traffic did not go anywhere. The measurement did.
- A tag dropped from one template during a redesign, a consent banner defaulting differently after a policy change, a new bot filter removing traffic that was previously counted, or a switch between reporting properties. Search Console click data is collected independently of your site, so a divergence between the two is the fastest way to catch this.
- Impressions are flat and clicks have fallen.
- You are still being shown; people are choosing something else, or the answer is now visible without a click. That points at the results page rather than at ranking: a new feature above you, a competitor with a better snippet, a title rewritten by the search engine, or a query that has simply stopped needing a visit. It is a snippet and intent problem, not an authority one.
- A redirect chain lost its middle link.
- Sites accumulate redirects across rebuilds. A tidy-up removes an old rule that looked obsolete, and a chain that used to resolve now ends in a 404 or loops. The affected URLs are usually the oldest and best-linked ones on the site, which is why the traffic loss is out of proportion to the number of URLs involved.
- A release reintroduced a staging setting.
- A meta robots noindex, an X-Robots-Tag header, a canonical pointing at a staging hostname, or a robots.txt disallow copied from a pre-production environment. These are common, they are invisible in a browser, and they take a site down in search within days while everything looks perfectly normal to the people who work there.
- It is seasonality, and you only have one year of data.
- Comparing to last month is how a normal seasonal trough gets diagnosed as a crisis. Compare the same twenty-eight days against the same twenty-eight days a year earlier, and if you do not have that history, check whether search interest in your category moves the same way.
- One page was doing most of the work.
- Many sites have a single article or landing page carrying a large share of organic sessions. When it slips a couple of positions, or the query behind it changes shape, the site-wide chart falls hard and nothing else has actually changed. Segment by landing page before treating it as a site-level problem.
Set-up
What to have open while you diagnose
Nothing here is exotic. The difference between a one-day diagnosis and a three-week argument is usually whether these eight things were available in the first hour.
- Search Console performance report in comparison mode, twenty-eight days against the same twenty-eight days last year, filtered to exclude brand queries
- The Search Console page indexing report, filtered to the affected template or directory
- Your analytics landing-page report over the identical window, using the same date logic
- The deploy log and CMS release history, including tag manager versions
- The robots.txt and XML sitemap as they existed before the decline began
- The Google Search Status Dashboard, for confirmed ranking update dates
- Server log files for crawler activity, if you can get them from hosting
- A rank tracker configured to the location and device your customers actually use
Questions
What people ask in the first week of a decline
How big a drop is normal fluctuation?
Day-to-day movement of ten to fifteen per cent is unremarkable on most sites, and weekends, holidays and paydays produce swings that look alarming in isolation. Compare rolling twenty-eight day windows against the same period a year earlier before treating anything as a decline.
What matters more than size is persistence and shape. A fall that holds for two full weeks across several page templates is a real event. A single bad Tuesday is not.
Was it a Google update?
Check the dates before assuming it. Google publishes confirmed ranking updates on its Search Status Dashboard, and the rule is simple: if your decline does not start inside the announced rollout window, the update is not your explanation.
Updates also make a characteristic pattern — broad movement across many queries and page types at once. A drop confined to one template or one country is almost never an update.
Should we roll back the release?
Only if you have evidence linking it to the decline, and only if the rollback is cheap. Rolling back blindly destroys the timeline you need for diagnosis and often introduces a second change to untangle later.
The faster move is to diff the release against the previous one for anything touching robots rules, canonical tags, meta robots, internal links, redirects or server-side rendering, and to test the affected template in a live inspection tool.
How long does recovery take?
It depends entirely on the cause, and honest answers here are ranges rather than dates. A reversible technical error can recover within a recrawl cycle once corrected. A decline caused by losing relevance to competitors is not a recovery project at all — it is the ordinary work of becoming a better result, which takes months.
Nobody can promise a return to a previous number, because the results page you fell out of is not the results page you would be returning to.
Traffic fell but enquiries did not. Does it matter?
Sometimes not at all. Losing informational visitors who were never going to buy is commercially neutral, and it happens routinely when an answer block satisfies a question on the results page.
It matters if the lost traffic was a source of future demand — people researching now and buying in six months — or if it is an early sign of a decline that has not yet reached your commercial pages. Segment by page type before deciding it is harmless.
Get a second pair of eyes on the decline
If you have worked the sequence and the cause is still not obvious, that usually means the loss sits across two things at once. Bring the charts to a call and we will tell you what we would test next.
If we don't deliver the work we agreed to deliver for reasons within our control, you don't pay for the undelivered work. Read our guarantee
Last updated