How to Interpret a Traffic Drop
Most reported traffic drops are not drops. They are shorter months, seasonal patterns, a tracking change or two periods that were never comparable. Establishing that first saves an enormous amount of work. It is the step almost everybody skips.
Most Drops Are Not Drops
The first job is establishing whether anything actually happened. Not what caused it. Whether there is anything to explain at all. Jumping to causes before answering that is how businesses break sites that were working perfectly well.
Why the order matters so much. Everything downstream depends on it.
A cause found for a drop that never occurred is a false lesson somebody will act on for years. Getting this wrong is worse than getting no answer.
The four false alarms. In order of frequency.
An unfair comparison, a measurement change, a season arriving and ordinary variation. Between them those explain most of what gets reported as a fall.
How long the check takes. Half an hour.
All four can be ruled out in one sitting, without any specialist knowledge. That half hour regularly prevents weeks of unnecessary work.
What a genuine drop looks like. Three things at once.
Sustained rather than momentary. Concentrated somewhere identifiable rather than spread evenly. And on searches that actually produce work.
What we will not give you. A percentage.
No figure at which a fall becomes real. Any number would be invented and it would be quoted back as a rule in situations it never fitted.
How this page is arranged. Elimination, then location.
Rule out the false alarms, find where a real fall is concentrated, then hand the cause to whoever owns it.
What Normal Variation Looks Like
Search traffic moves week to week for reasons nobody can name. That is not a problem to be solved. It is the ordinary condition of the data. Knowing what it looks like prevents most of the alarm this subject generates.
Why it happens. Countless small causes.
Weather, news, what competitors published, how people happened to phrase things that week. None is identifiable afterwards and together they move every figure constantly.
What that means for small movements. They mean nothing.
A modest fall between one week and the next carries no information at all. Reacting to it is reacting to noise.
What to look at instead. The shape.
How the figure behaves across a longer window. A dip inside a rising quarter is not a decline. Only the wider view shows that.
The pattern that fools people. A flat month.
Growth arrives in steps rather than a smooth line, so flat periods are normal even when everything is working. Our trends guide covers that properly.
How to tell noise from signal. Persistence.
Noise recovers on its own within a period or two. Something that keeps falling across several periods is a different matter entirely.
The discipline this produces. Wait before acting.
A single bad week is one week. Waiting to see whether it persists costs nothing and prevents most unnecessary intervention.
Check The Comparison First
Before anything else, look at what is being compared against what. An enormous proportion of reported drops are two periods that were never comparable. The percentage attached to them is describing the calendar rather than the business.
Different month lengths. The obvious one.
A short month against a long one loses several days of traffic before anything else happens. That alone produces a fall nobody caused.
Different trading days. Less obvious.
Two months of the same length can contain different numbers of weekends. For a business whose enquiries arrive on weekdays, that shifts the figure noticeably.
Holidays. Frequently missed.
A bank holiday, a school break or the period around Christmas changes behaviour for everybody. Comparing across one without noting it produces a confident wrong conclusion.
An unusually strong previous period. The subtle one.
If last month was exceptional, this month looks like a collapse while being entirely normal. Comparing against a peak is comparing against an outlier.
What to compare instead. The same period last year.
Same length, same holidays, same position in the seasonal cycle. That single change removes most of the problems above at once.
The reporting habit that prevents it. Label the basis.
Every percentage should state what it is comparing. A figure with no stated basis is the commonest way a report misleads without anybody intending it.
Then Check The Measurement
A measurement fault looks exactly like a traffic loss. Identical shape, identical timing, identical alarm. It is also far more common than people expect, particularly on any site that has been worked on recently.
The classic cause. A site update.
Tracking removed or not reinstated during a rebuild, a redesign or a template change. Visitors keep arriving and the recording stops.
What that looks like. A cliff.
A sudden fall on a specific date rather than a gradual decline. Search behaviour rarely changes overnight, so a sharp edge points at measurement first.
The second cause. A filter.
Something applied to exclude certain visits, frequently for good reason, then forgotten. The figure falls and nothing about the site changed.
The third. Consent changes.
An alteration to how visitors are asked about tracking changes how many agree, which changes how many are recorded. Legitimate, though it reduces the totals.
How to check quickly. Two sources.
Compare your site analytics against your search reporting. If one fell and the other did not, the problem is measurement rather than visitors.
Why this must be ruled out early. It is free to fix.
Discovering it after three months of content work means three months aimed at a problem that never existed.
Who checks it. Your developer.
The mechanics belong to whoever maintains the site. Asking whether anything changed on the date the fall started usually settles it.
Then Check Seasonality
Nearly every trade has a shape to its year. A fall arriving when it always arrives is your season rather than a problem. This is the check most likely to end an investigation, plus the one people are most reluctant to accept.
Why it gets missed. Seasonality sounds like retail.
Most trades assume it does not apply to them. In practice it applies to almost every business in one form or another, frequently more sharply than in shops.
The check itself. Same period, previous year.
If the fall matches what happened at the same point last year, you have your answer. That comparison takes minutes.
What to do without a year of data. Use what you know.
A business that has traded for years knows its own quiet periods even if the website does not. Your own knowledge of the trade is legitimate evidence.
The pattern that surprises people. Searching comes first.
People search in advance of buying, so the search peak precedes the trading peak. A fall in searches can arrive while the phone is still busy.
Where this is covered fully. Its own guide.
How seasonality affects the data covers the shapes, the lag and the dates that move each year.
The warning attached. Do not overuse it.
Seasonality explains a great deal and cannot explain everything. A supplier reaching for it every time a number falls is avoiding a question.
Now Look At Where It Fell
If a fall survives all three checks, something real happened. The next question is where, because a drop concentrated in one place and a drop spread evenly across everything point at completely different causes.
Concentrated in one section. Something specific.
One service, one part of the site or one group of pages. That points at something done to those pages. Or at somebody outperforming you on that subject.
Concentrated on one page. Narrower still.
Almost always something about that page or the search it served. The easiest kind to diagnose and frequently the easiest to reverse.
Spread evenly across everything. Site wide.
A technical fault affecting the whole site. Or something that changed in search generally. Nothing about one page will explain it.
Concentrated by device. Worth checking.
A fall on telephones alone points at how the site behaves on a small screen, which is a different investigation entirely.
Concentrated by place. For local businesses.
Losing one town while holding others suggests something local rather than something about the site.
What to use to find it. The search data.
Break the fall down by page and by search. The search metrics guide covers reading those numbers together.
The date, again. Write it down.
When the fall began is the single most useful fact in the whole investigation, because it lets you ask what else happened that week.
Branded Or Non-Branded
Split the fall into searches containing your business name and searches that do not. Those two behave completely differently and they have completely different causes, so this single split frequently resolves the whole thing in minutes.
If your own name fell. Not a search problem.
Fewer people searching for you means fewer people know about you or fewer are thinking about you. That is awareness rather than anything a search engine did.
What causes that. Usually offline.
Marketing that stopped, a quiet season for word of mouth, a sponsorship ending or simply a slower period in the trade. None of it is fixable through the website.
If everything else fell. A search problem.
Fewer people who did not know you are finding you. That is the half this work is bought to affect and it warrants proper investigation.
If only your name held up. The clearest signal.
Reach fell while recognition held. That pattern points squarely at something in search rather than at the business.
Why the combined figure hides it. They cancel.
A rise in one and a fall in the other can produce a flat total, which is how a genuine problem stays invisible for months.
The habit worth adopting. Always split.
Not only during an investigation. Reporting the two separately every month means the pattern is visible before it becomes a question.
Then Hand Off
Once you know a fall is real and where it sits, the cause belongs to whichever subject owns it. This guide covers whether something happened. Working out what, from here, is somebody else's territory and we would rather point you there than half explain it.
If everything moved together. An update.
Several unrelated sections falling at once, with competitors moving too, points at a change in search rather than in your site. Our algorithm updates guides cover that.
If it followed a rebuild. A migration.
A fall beginning at the point a site was rebuilt, moved or restructured is a migration question. Our migration material covers what goes wrong and in what order to check it.
If there is a notice waiting. Formal action.
Rare, quick to rule out and it changes everything if present. Our risky practices material covers what that means and what follows.
If one page lost ground. Content and competition.
Read what now outranks you and ask what it covers that yours does not. That is usually the whole answer.
If nothing fits. Block nine.
Sometimes nothing about your site changed and the answer is somebody else, which is the next section.
What we will not tell you. That it comes back.
Recovery depends entirely on the cause and we promise none of it. Some positions do not return. Knowing that early is more useful than reassurance.
When It Is Somebody Else Rather Than You
A site can lose traffic without anything happening to it. Somebody else improved, something new appeared above you or a larger organisation entered the space. Nothing on your site changed and your traffic fell all the same.
A competitor improved. The commonest.
Somebody published a better answer than yours on a subject you owned. Being overtaken feels identical to being demoted and it has a completely different remedy.
A large site arrived. Less common, harder.
A national brand or a big publisher deciding to cover your subject. That may not be winnable back. Recognising it early saves wasted effort.
The results page changed shape. Nothing to do with anybody.
New features appearing above the ordinary results push everything down without changing a single position. Fewer visits, same placement, nothing on your page to fix.
The answer got shown directly. Increasingly common.
Where a search is satisfied on the results page itself, people stop clicking. Your position holds and your visits fall.
How to check all four. Search it.
Look at the results page for the affected search as it is today. Two minutes of looking explains most of these immediately.
What follows if it is one of these. A different question.
Not what broke. Whether the search is still worth competing for. What would need to be true to win it back.
Do Not Start Changing Things
Changing the site during an unexplained drop removes your ability to diagnose it. This is the same rule our migration material applies to a rebuild. It is the single most valuable instruction on this page.
Why it destroys the diagnosis. Too many variables.
Once several things have changed, no subsequent movement can be attributed to anything. You have exchanged a problem for a mystery.
What panic usually produces. A rewrite.
Somebody rewrites the affected page. If it recovers, the rewrite takes credit for something that may have recovered anyway. The wrong lesson is learned permanently.
The one exception. A confirmed fault.
If you find something genuinely broken, fix it. That is not a speculative change, it is a repair with a known cause.
What to do instead of acting. Record.
Write down the date, what fell, where it was concentrated and what you have ruled out. That record is worth more than any hurried change.
The rule for when you do act. One thing at a time.
One change, dated, then wait. Several at once means never knowing which worked, which matters more than speed.
What prevents the next one. A change log.
A dated note of every alteration to the site, kept from today. Without it no future drop can ever be explained.
Where everything sits. The hub.
All nineteen guides are on the measuring performance guide.
A measurement
fault looks
exactly like a
traffic loss.
Identical shape, identical timing, identical alarm. Far more common than people expect on any site that has been worked on recently. Tracking removed during a rebuild, a filter applied and forgotten, a consent change altering what gets recorded. Visitors keep arriving and the recording stops. Compare your site analytics against your search reporting and if one fell while the other held, you have your answer.
What we rule out before touching anything:
Changing the site during an unexplained drop removes your ability to diagnose it. You exchange a problem for a mystery.
Every guide.
One question.
What each number actually means, how to read the data without being misled, what belongs in a report and what does not, then how to connect any of it to the enquiries the business is actually paying for.
