Why Is Google Analytics Not Tracking Data?
Before the fault finding, the calibration. This tool records a portion of what happens on your website rather than all of it. It always has. A great many of these questions are somebody discovering that for the first time.
It Is Never Complete
Analytics does not record every visit to your website. It never did and it is not supposed to. Expecting it to match reality exactly is the source of most of the questions on this page. What you have is a large, useful, systematically incomplete sample.
Why some visits are missing. Several ordinary reasons.
Consent choices, browser settings, extensions that block measurement and connections that fail before anything is sent. All routine and none of them faults.
How much is missing. Nobody can tell you.
It depends on your audience, their devices and their habits. We publish no figure for it. Any supplier quoting you one has invented it.
What that means practically. Treat it as indicative.
Your figures are directionally reliable and not exact. Decisions should survive the numbers being approximately right rather than precisely so.
What is still worth investigating. Change.
A stable gap between the reporting and reality is normal. A sudden change in that gap is a fault. The rest of this page is ordered by how likely each cause is.
Why the distinction matters commercially. Different urgency.
An empty property is a fault losing you data every day it continues. A property recording less than you hoped is usually working correctly. Rushing to fix it produces changes nobody needed.
The question to ask first. Nothing at all, against less than expected.
Those are different problems with different causes. Block two establishes which one you have.
Checking It Works
One test settles whether anything is arriving at all. It takes two minutes. Everything after this block depends on knowing the answer, so it is worth doing before reading further.
The test. Visit your own site.
Open your website, then look at the live reporting shortly afterwards. If your visit appears, data is arriving and the property is connected to the site.
What a positive result means. Narrower problem.
The installation works, so anything wrong is about what is being recorded rather than whether anything is. Skip to block four or five.
What a negative result means. Block three.
Nothing is reaching the property, which is a clear fault with a short list of causes.
The complication to allow for. Your own exclusion.
If internal traffic is excluded, your own visit may correctly not appear. Test from a mobile connection outside your network before concluding anything.
The second test worth running. An enquiry.
Send a real enquiry through your own form and check it was recorded. Visits arriving while enquiries are not is a common and more expensive fault, covered in our conversions guide.
When to run both. After any site change.
A redesign, a new form or a platform update. Ten minutes then saves months of quiet loss.
Nothing At All Arriving
An empty property has four usual causes and one of them accounts for most cases. If your reporting shows zero rather than something disappointing, this is a short list to work through.
Removed during a rebuild. The commonest by far.
A new website goes live and nobody carried the measurement across. The reporting stops on the launch date, which is the giveaway. Months pass before anybody notices.
Never installed. More common than you would think.
A property created during onboarding and the identifier never added to the site. Everything looks configured and nothing was ever connected.
Pointing at the wrong property. The confusing one.
Data arriving somewhere, just not where you are looking. Businesses with several properties from previous suppliers hit this regularly.
Blocked by the platform. Occasional.
A setting on the website preventing measurement from running. Or a consent arrangement configured to block everything by default.
How to date the stoppage. Find the last day.
Look for the final date with data and compare it against what happened to the website that week. In almost every case the two match. That is the answer without any further investigation.
Where the fix lives. The installation guide.
Adding analytics to your website covers the routes for each platform. the setup guide covers the stream pointing at the wrong address.
What you have lost. That entire period.
Which block eight covers. It is the reason a rebuild deserves a tracking check within days rather than months.
Some Data But Less Than Expected
Data is arriving and the figures look lower than the business believes they should be. Most of the time this is not a fault at all. It is the incompleteness from block one, meeting an expectation that was never realistic.
Consent choices. A genuine reduction.
Where a visitor declines measurement, their visit may be recorded in limited form or not at all. That is the arrangement working as intended.
Blocking. Increasingly ordinary.
Browser settings and extensions that prevent measurement running. Some browsers do a degree of this by default now, without the visitor choosing anything.
What you should not do. Try to defeat it.
Configurations exist that claim to recover blocked visits. We do not describe them and we do not advise on what your consent arrangement should cover, which is a question for somebody qualified.
How to tell normal from faulty. The trend.
A figure consistently lower than you expected is calibration. A figure that dropped noticeably on a particular date is a fault. The date tells you what to investigate.
The date to check first. Any site change.
A redesign, a plugin update, a consent banner change or a platform migration. All can reduce what is recorded without anybody intending it.
What we never publish. A shortfall figure.
The direction is knowable and the size varies enormously by audience, so a percentage would be a guess dressed as a fact.
More Data Than Expected
Figures higher than they should be is the problem nobody investigates, because a good month is rarely questioned. It is almost always duplication. It inflates everything at once.
The usual cause. Installed twice.
The measurement code present twice on the same page, so every visit is counted twice. It happens when a plugin, a theme setting and a manual addition are all doing the same job.
How to recognise it. The shape.
Visits roughly doubling with no corresponding change in enquiries. Engagement figures that look implausibly poor, because each visit is being split.
The other tell. A step change.
The inflation starts on a specific date, usually the day somebody added the second installation. That date is the whole diagnosis.
The version that is harder to spot. One section only.
Where the duplication affects part of a site rather than all of it, usually a template used by some pages. Those pages look far more popular than they are, which distorts decisions about what content works.
Where it does most harm. Shops.
Duplicated purchases inflate recorded revenue as well as visits, which our ecommerce guide covers.
What fixing it does to your history. Creates a step.
Figures will drop sharply on the day you fix it. That is the correction rather than a decline. Write it down so nobody panics in three months.
Where to check. The installation guide.
Adding analytics to your website covers how sites end up measuring everything twice.
Data Sampling
Sometimes the tool calculates a figure from part of your data and scales it up rather than counting all of it. That is sampling. It is disclosed in the interface. It is also the usual reason two views of the same period disagree.
When it happens. Long ranges, complex questions.
Large date periods and analysis with several conditions attached. Ordinary monthly reporting on a small site rarely triggers it.
Why it exists. Speed against precision.
Counting everything for a complicated question across a year is expensive, so an estimate is returned quickly instead. That is a reasonable trade rather than a shortcoming.
What it affects. Precision, not direction.
The shape of a comparison stays reliable. The exact figures are approximate, so quoting them to the last visitor overstates what you know.
How to avoid it. Ask less at once.
A shorter period or a simpler question frequently returns a full count. Worth trying before accepting an estimate.
Where it surfaces most. Explorations and dashboards.
Free-form analysis and connected reporting both meet it regularly, which our explorations guide notes.
What to do if it reaches a report. Label it.
Say the figure is an estimate. Presenting a sampled number as a count is how confident wrong statements enter a business.
Thresholding
Some figures are withheld when the group they describe is very small. The purpose is preventing individuals being identified from a report, so a segment covering a handful of people can appear empty when it is not.
What it looks like. A gap, not a zero.
Rows missing. A total that does not add up from its parts. A report that appears blank despite traffic existing. Nothing announces that anything was withheld.
When it catches people. Fine cuts.
A small town, a narrow age group, one day of a quiet week. The more precisely you slice, the more likely you are to meet it.
Why it is not a fault. It is a protection.
A report describing four people in one postcode could identify them. Withholding it is deliberate and reasonable. It is not something to configure away.
How to work around it. Widen the question.
A longer period or a broader group usually brings the figures back. That is the correct response rather than a workaround.
Why it matters for local businesses. Small numbers everywhere.
A trade serving three towns is working with exactly the group sizes where this appears, which is worth knowing before concluding a town produced nothing.
What we will not cover. Obligations.
This is described as a thing that affects your figures. Nothing here advises on data protection requirements.
Fixing Missing Data
Data that was not collected cannot be recovered. There is no archive to request, no backup to restore and nothing any supplier can retrieve. Every fix on this page applies from the moment you make it and not one day earlier.
Why that is absolute. Nothing was recorded.
The information does not exist in a degraded form somewhere. It was never captured, so there is nothing to reconstruct from.
What that costs. The comparison.
A business that fixes a fault after eight months has eight months missing from every year on year comparison it will make for the next year.
Why setup mistakes are expensive. Same reason.
Not defining what counts as an enquiry is not a small oversight to correct later. It is a year of outcome data you will never have.
What you can do instead. Annotate.
Write down what was wrong and for how long, then keep it with the reporting. A gap somebody can explain is far more useful than one nobody understands.
What can partly stand in. Your own records.
Invoices, enquiry emails and diary entries covering the missing period. Cruder than analytics and genuinely useful for establishing whether a quiet stretch was real or simply unrecorded.
What to do the same day. Fix it, then export.
Correct the fault and begin exporting monthly, so the next gap costs you nothing.
Where migrations sit. A separate subject.
Continuity across a rebuild is covered in our site migrations material. It is the situation that causes most of these losses.
Do Not Compare Across A Change
Any period spanning a change to how measurement works is not comparable with the period before it. That sounds obvious and it is broken constantly, usually by somebody producing a year on year chart across a migration.
What counts as a change. More than you think.
An installation altered, a duplicate removed, a consent banner replaced, a site rebuilt or a move between versions of the product. All of them shift what gets recorded.
What the chart will show. A step.
A movement at the change date that looks like performance and is measurement. Somebody will explain it with a theory. The theory will be wrong.
What to do instead. Break the series.
Treat the change as the start of a new comparable period. Awkward for a year and correct throughout.
The habit that makes this possible. Dating everything.
A short note of every change to the site or the measurement, kept with the reporting. Ten seconds each time and it answers questions nobody could otherwise resolve.
Who this protects. Everybody.
Including whoever inherits the account and finds an unexplained step in a chart three years later.
Where the rest sits. The hub.
Everything on this tool is on the Google Analytics guide. Our measuring performance material covers reporting across a change.
Your reporting
stopped on the
day the new
site went live.
Nobody carried the measurement across, nothing warned anybody and the property has been empty since. Months of data are gone permanently, because analytics records forward only and there is no archive to request. The date it stopped is usually the whole diagnosis.
What we check when the figures look wrong:
A stable gap between the reporting and reality is normal. A gap that changed on a particular date is a fault worth finding.
Every guide.
One tool.
Getting it installed and configured properly, reading the reports without being misled, measuring outcomes rather than activity, the advanced reporting and what to do when the numbers look wrong.
