What Google Analytics cannot show you—and what to add

This is not an article against Google Analytics. The tool does what it was built to do well: it shows scale, direction, and location. Problems begin when a team treats it as the sole source of truth and decides to rebuild a website based on a number that does not answer the question being asked.

Your data is incomplete, and that is normal

The first thing to accept is that analytics does not see every visit. Some people reject consent, some use blockers, and some browse in modes that restrict tracking. You cannot fix this, nor should you try—you need to account for it when interpreting the data. A separate kind of gap is visits that analytics does see but without a source, such as some of the visits from AI assistants covered in our guide to AI referral traffic from ChatGPT and Gemini.

• ConsentWithout consent to analytics cookies, data is limited or modeled rather than directly counted.
• BlockersBrowser extensions and settings exclude some traffic from measurement.
• Data thresholdsA segment that is too small may be hidden to prevent individuals from being identified.
• RetentionDetailed data disappears after the configured period. Looking back over a full year may be impossible.

The practical consequence is simple: treat numbers as directional indicators and compare them with themselves over time, not with someone else’s values. Comparing your result with an ‘industry average’ from an online article means comparing two different definitions across two different datasets. Once a year, run that comparison against a fixed set of questions, such as the list in our annual website and support review.

The consent banner itself is also part of this equation—and not only from a data perspective. We explain how it affects the first seconds of a visit and how to time the launch of your tools in our article on consent banners and user behavior.

A metric does not mean what its name suggests

The second group of misunderstandings stems from names. Metrics have intuitive labels and unintuitive definitions, and those definitions change between versions of tools. An argument in a meeting is very often not about the website, but about what a given column counts. A good example is the bounce rate, which in GA4 means the share of sessions that were not engaged, not simply leaving without a click.

↔ Scroll the table sideways to see all columns.

We examine two of these metrics in separate articles: time spent on page and event selection in our article on micro-conversions. The rule they share is simple: before quoting a number in a meeting, check its definition in the tool it came from.

The same difference in definitions explains most cases where two tools show different numbers for the same period. We break these down in our article on why the data does not match.

Questions that numbers can never answer

This is the most important boundary, and better configuration cannot move it. Aggregate data describes what happened collectively. It does not contain the course of an individual visit, the substance of someone’s doubts, or the reason they gave up.

↔ Scroll the table sideways to see all columns.

This comparison reveals a simple division of labor: numbers identify the location, qualitative sources explain the situation, and a conversation verifies the motivation. We describe the full process of moving between these layers in our guide to website user behavior analysis.

How to combine both sources in one process

The biggest mistake is treating the tools as competitors. In practice, they work in sequence: one narrows the area, and the other explains it. The order matters, because reversing it leads to spending half a day watching random recordings.

1. Start with the number.Find the page or step with the largest drop. This identifies the location; it is not a diagnosis.
2. Narrow it to one segment.One device, one traffic source. Mixing them invalidates the conclusions that follow.
3. Check for a pattern.Click and scroll maps will tell you whether the behavior repeats.
4. Watch several journeys.Choose recordings from the same segment and the same step, not random ones.
5. Ask.Questions from conversations are the only source of motivation. Everything else is interpretation.
6. Return to the numbers.After making a change, check whether the scale of the problem has actually decreased.

The final step closes the loop, and it is the one most often left out of the process. Without returning to quantitative data, you cannot know whether you fixed a real problem or a single case spotted in a recording.

Before deciding that a decline is real, check whether it is simply a calendar or seasonal fluctuation. We describe how to tell the difference in our article on seasonality in traffic and inquiries.

Supplementing data and protecting privacy

Adding more sources means adding more data about people. That is not a reason to abandon them, but it is a reason to make those decisions consciously and document them before the tool goes live.

• You know what data each tool collects and can describe it in one sentence.
• Each scope is described separately in the privacy policy, rather than in one sentence about ‘analytics tools.’
• Retention periods and the people who have access are defined.
• Sensitive fields are masked before anyone watches the first recording.
• You know what happens to the data when your relationship with the provider ends.
• The team knows how to handle a request to access or delete data.

We collected the configuration questions for recordings in our article on session recording and GDPR, while the data created during a conversation is covered in our article on live chat and GDPR. These are two different scopes, and the decisions should be made separately.