Conversion rate in Google Analytics: what the tool counts, where to read it, how to segment it

You are reading a conversion rate in Google Analytics and it does not say what you expected: it differs from the back office, it has moved without anything changing on the site, or two screens of the tool contradict each other. This page is for the marketing or e-commerce manager who has to present that figure: what GA4 counts, where it displays it, how to break it down without distorting it, and what you are entitled to conclude from it.

What Google Analytics counts when it displays a conversion rate

GA4 is built on events, and what used to be called “conversions” now corresponds to key events that you mark as important (purchase, lead generation, demo request, and so on). The point is to select a small number of genuinely business-relevant events so that your Google Analytics conversion rate stays readable and actionable.

You mark the action tied to revenue, depending on your business:

• E-commerce site: the conversion usually comes down to a sale.
• B2B or services site: it is a qualified contact request, a quote request or a demo request.
• Intermediate actions: signing up to a newsletter, joining a community, downloading a white paper are tracked separately, as micro-conversions.

Everything else stays an ordinary event, available to consult but excluded from the rate. That is the definition used to judge a good conversion rate: micro-conversions are read alongside, never inside the rate.

The declaration is made in the property admin. Three rules are worth more than a sequence of clicks:

• Where: on the screen listing the events already collected, each one carries a setting that promotes it to a key event.
• What: an event is only marked if it changes a commercial decision; a newsletter sign-up and a quote request do not have the same status.
• How many: count your key events and set yourself a ceiling; three to five is enough on most sites.

Beyond that, the rate rises mechanically without anything having improved.

The denominator, for its part, produces the most misunderstandings in meetings. GA4 displays a session key event rate (the sessions that fired at least one key event, divided by sessions) and a user key event rate, built the same way on users. The session rate is the one used as the reference for benchmarking yourself; the user rate is read alongside.

The numerator is therefore not the number of events. Out of 100 sessions, if 8 sessions lead to a purchase and 10 orders are placed, the converted session rate is 8%, while orders set against sessions come to 10 per 100. They are not the same indicator, and the gap widens as soon as one visitor orders twice: check the metric displayed and the period before commenting on a gap.

Where the rate is read, report by report

Three scopes coexist in the tool, and one and the same key event produces three values with different meanings there:

Each family of reports calculates on its own population: that is the first explanation for gaps internal to the tool.

Three places cover the essentials:

• Engagement reports give the rate at page level: “does this page convert?”, attaching the key event to the page where it fired or to the entry page, depending on the column used.
• Acquisition reports give the rate at source level: “does this channel convert?”, with the session or the user as the unit, never the page.
• Explorations are not one more report: they are the space where you cross dimensions and key events that the prebuilt reports do not assemble for you.

Two screens of the same tool can therefore display two rates without either being wrong. The causes, in order of frequency, are as follows:

• the denominator is not the same: sessions here, users there;
• the event scope is not the same: one screen totals every key event, another isolates a single one;
• the date window or the filter applied differs.

These three points can be checked in a few minutes, before any business explanation. Hence a reporting rule: one reference screen per question, written down somewhere, so that the rate presented in committee can be reopened with the same settings by someone else. Without that, the figure ends up being rejected wholesale.

For the reader who does not yet have the tool in hand (property setup, data model, a full overview of the reports), what Google Analytics lets you track sets out the general frame these screens sit in.

Building the four breakdowns inside the tool

An overall rate is an average, and a badly built breakdown does worse: it produces a precise, wrong figure that nobody questions because it came out of the tool. The four useful breakdowns (channel, campaign, entry page, device) require the dimension used to cut to be filled in the same way across the whole period being compared.

Hence a discipline that comes before any segmentation: nomenclature.

• decide once how each traffic source is named;
• write that convention down and share it with those who do the labelling;
• attach each channel to a source/medium consistent with it.

A campaign labelled differently from one month to the next cannot be compared with itself.

By acquisition channel and by campaign

The channel breakdown rests on the Default channel group (Organic Search, Paid Search, Email, Social and so on), which the tool fills on its own for organic and direct, and from your labels for the rest. The steps can be checked one by one:

• open Traffic acquisition;
• take the session default channel group as the dimension;
• add the key event rate column;
• select in its menu the key event you want instead of leaving them all, otherwise you are reading the sum of your markings.

Check that your UTMs are consistent for SEA and email: that is where breakdowns break, silently. Badly labelled traffic migrates into “direct” or “unassigned” and distorts two lines at once.

Then read several columns, never just one:

Confusing volume and rate leads to cutting the channel that converts best because it weighs little.

After every release, reopen the breakdown over the last seven days: an abrupt swing between two channels nearly always signals broken labelling, not a change in behaviour.

By entry page and by device

For reading by page, the Landing page report carries the same key event rate column, attached this time to the session's entry page. Do not confuse it with the page where the event fired, which is another column: one says where the visitor came in, the other where the action took place, and on a journey across several pages they are not the same.

Beyond that, an Exploration crosses landing page, source/medium and key event (for example lead generation) to pinpoint the pages to optimise. The terms typed into the internal search engine complete that reading: they reveal keywords you have not yet worked on; the function itself is configured with the product team.

The device breakdown (computer, tablet, mobile) signals, it does not conclude: a gap on the same page is first a matter of usage context. When it really is the page, what it points to goes back to the product, UX and development teams:

• responsive layout;
• form legibility;
• loading time.

Your share is to quantify the shortfall and date the observation, so that the request can be arbitrated.

Building a funnel and reading the drop-offs

The conversion funnel is the journey at the end of which the visitor becomes a buyer. Google Analytics isolates each of its steps to spot the weak ones and the ones that perform well.

Two exploration tools handle this, and they do not answer the same question. The funnel (funnel exploration) is declarative: you set the expected steps (landing, product page, form, confirmation) and the tool gives you the pass-through from each one to the next. The path (path exploration) is descriptive: it shows what visitors actually did, detours included. The first tests a hypothesis, the second helps you form one.

Then analyse drop-offs step by step. A pass-through gap between two steps allows three conclusions, and only three:

• where the journey breaks off;
• for what proportion of those entering;
• since when.

It does not allow you to say why. A drop at the form can come from the form, from the price shown at the previous step, from a traffic mix bringing in less advanced visitors, or from an event that has stopped firing. First check: is the step still measuring what it was measuring last month?

Once the measurement is confirmed and the step really is at fault, the matter changes owner. Checkout journey, form fields and payment methods are fixed with the product and development teams, on their calendar. The funnel gives them the exact step, the proportion lost and the period: a documented request comes before an argued one.

When two tools do not say the same thing

The GA4 figure matches neither the CRM figure, nor the back office figure, nor the one from an advertising platform. That is normal. The problem starts when nobody knows which one is authoritative: the meeting arbitrates between two exports instead of arbitrating a decision.

Two moves are enough, in this order: understand where the gap comes from (technical), then decide which source answers which question (governance), which settles the subject for good.

Why the figures diverge

Gaps are common. Their usual causes are known:

• attribution differences (last click or multi-touch);
• time windows;
• consent management;
• analytics blocking (adblockers);
• conversion deduplication on the CRM side.

It still helps not to confuse the two settings filed under “attribution”. The attribution model decides which touchpoint gets the credit for a conversion; the lookback window decides how far back the tool goes to look for it. They are set separately, and acquisition reports read on the session channel do not go through them: they attach the conversion to the session.

Each cause moves the figure in a predictable direction, which lets you recognise it without an investigation:

Deciding on one source of truth per indicator

The most reliable approach is to define one source of truth per type of KPI (the back office for revenue and sales, GA4 for behaviour) and to align the definitions: what a conversion is, and at what moment it is counted. Written down, the rule holds: one line per indicator, its source and the person who produces it.

One check saves you redoing this work every quarter: after every release, and before commenting on the slightest movement, verify the tagging, consent, the acquisition mix and the pages changed. The causes of a sudden drop and the reading rhythm that lets you spot it are set out with the conversion rate KPIs.

What Google Analytics does not yet measure: traffic from generative engines

In 2026, a growing share of traffic goes through AI-generated answers: AI Overviews in Google, assistants such as ChatGPT or Perplexity. Knowing where GA4 files them stops you underestimating your real conversions.

• Clicks from an AI Overview: they come through as google / organic, mixed in with the rest of organic traffic. GA4 does not distinguish them from a standard click.
• “Dark” traffic: these are visits that arrive with no referrer, from apps, messaging services or some assistants. They show up as “direct” or “unassigned”, and the channel breakdown inherits the bias. The referrers of generative engines that do pass one can be grouped into a breakdown of their own; the rest is monitored through the volume of “direct”.
• Pre-click micro-conversions: a significant share of searches end without a click, the answer being read directly in the AI Overview. Being cited in the answer then becomes a visibility KPI, which GA4 does not see since it only counts visits; the scale of the phenomenon can be consulted in the SEO statistics.
• Completing GA4 with GEO tracking: to measure the real impact of AI, cross GA4 data with GEO tracking (generative share of voice, citation frequency, positioning within AI answers) and adjust your content strategy accordingly.

Until the attachment is reliable, these visits are presented as a tracking line, never as a quantified contribution to revenue.

Continue reading

Depending on what your breakdowns have brought to light, three reads take this one further:

• You want to compare channels with one another: if the question becomes which one deserves which budget, the comparison changes in nature, with the digital marketing conversion rate by channel.
• You are isolating the organic channel on its own: once the breakdown is in place, reading it continues with the SEO conversion rate.
• You have identified the page or the step that is dropping off: if the gap holds over several periods, the subject becomes page prioritisation and effect measurement, covered by optimizing the conversion rate.

To cross GA4 data with the tracking of your presence in generative answers without rebuilding the export by hand, the performance reporting module combines, at the time of writing, GA4 data and GEO tracking, including traffic from AI engines.

Frequently asked questions

How do you analyse the conversion rate?

You analyse a conversion rate by first checking what the tool counts: the key events declared and the denominator of the rate. Next comes the breakdown by acquisition channel, which separates organic from paid, then the breakdown by entry page, which locates the gap. Device comes last: it refines a diagnosis more than it establishes one.

Where do you find the conversion rate?

In GA4, the conversion rate is found in the engagement reports for pages and in the acquisition reports for channels. Custom explorations are used to cross it with other dimensions, such as source, device or entry page. Check the unit displayed every time, because two screens can give two correct values on two different populations.

What is a good conversion rate?

A good conversion rate is one that makes your acquisition cost compatible with the margin per conversion and that improves against your own history, with a constant definition. It is then read by segment: channel, device, page type. Published benchmarks give an order of magnitude, with an identical denominator, but never set a target.

How do you calculate the conversion rate in Google Analytics (GA4)?

GA4 calculates the session key event rate by dividing the sessions that fired at least one key event by all sessions; the user rate follows the same logic. The numerator counts sessions or people, never event occurrences. Only events marked as key events enter the calculation.

Why does my GA4 conversion rate suddenly fall or rise?

A GA4 rate that moves suddenly most often signals a change in measurement, not in behaviour. Frequent causes are a key event added or removed, a new consent setting, an acquisition mix bringing in more visitors at the start of their journey, or pages that were changed. The deployment history and the breakdown by channel and by page settle it quickly.

Which micro-conversions should you track before a final conversion?

In B2B, the useful micro-conversions are a click on a call to action, a white paper download, a view of the pricing page, a form start or a meeting booked. In GA4, they are configured as events and tracked separately. They shed light on the journey but do not enter the conversion rate, which they would inflate artificially.

How do you read the conversion rate with GA4 explorations?

GA4 explorations read the conversion rate step by step, with the funnel to test an expected journey and the path to observe the actual one. The gap between two steps says where the visitor drops off, in what proportion and since when. It does not say why, and the first check is always that the step is firing correctly.

Why is my conversion rate different between GA4 and my CRM or e-commerce tool?

GA4 and a CRM or back office diverge because they do not count the same thing or in the same way: attribution, time window, consent, blockers and deduplication move the figure. The gap is normal. It stops being a problem once each indicator has a written source of truth: the back office for sales, GA4 for behaviour.