Why GA4, Google Ads and your shop never show the same numbers
Three systems, three different numbers, one truth. The differences are not a fault — they follow from each system counting something else. What each measures, which gaps are fine and when it really is a problem.
- Google Ads credits a conversion to the click date, GA4 to the conversion date. That alone creates a gap that is not an error.
- Each system uses a different attribution model and window, so the totals cannot agree.
- The deciding number is the order in your own system. Ad data is input for optimisation, not accounting.
A scene that repeats in every second meeting. Google Ads reports 180 conversions, GA4 shows 143, and the shop recorded 161 orders over the same period. Someone concludes that none of it can be trusted.
The truth is that all three numbers can be correct. They are measuring three different things.
What each platform counts
| Google Ads | GA4 | Your shop | |
|---|---|---|---|
| Which day it credits | The click date | The conversion date | The order date |
| Default attribution | Data-driven, across Google channels | Data-driven, across all channels | None, or your own |
| Impressions without a click | May count them (view-through) | Does not count them | Does not count them |
| Across devices | Yes, for signed-in users | Limited | Only when the customer signs in |
| Modelled conversions | Yes, when consent is missing | Yes | No |
| Cancellations and returns | Unaware unless you send them | Unaware unless you send them | Aware |
The first row alone is enough to produce a noticeable gap. If someone clicks on 29 January and buys on 3 February, Google Ads credits January and GA4 credits February. For longer purchase cycles that matters, and it cannot be “fixed”.
Differences that are fine
- Google Ads reports more than GA4. Almost always. The reasons: view-through conversions, cross-device measurement, modelling and crediting to the click date.
- GA4 reports less than the shop. Blockers, refused consent, orders taken by phone, purchases in another browser.
- Channels in GA4 do not sum to the total. Data-driven attribution splits credit to decimal places across several channels.
The rough tolerance we work with: a difference of around ten to fifteen per cent between the ad system and the shop, for shops with a short purchase cycle, is normal. A larger gap is worth investigating.
Differences that mean a problem
- Google Ads reports double the orders. Almost certainly duplicate tracking: the tag deployed twice, or a conversion firing on every load of the thank-you page.
- GA4 reports more orders than the shop. Either incomplete payments are counted, or the event fires repeatedly.
- Values differ by exactly the VAT rate. Decide whether you send values with or without tax and keep it consistent everywhere.
- Numbers fell by tens of per cent overnight. A deployment error, a change to the consent banner, or a tag broken by a new site release. See Consent Mode.
- A large share of traffic is “direct”. Lost parameters on redirects, missing campaign tagging, or visits from AI assistants. See traffic from AI assistants.
Attribution models, briefly
- Last click. Everything to the final channel. Simple, but systematically overstates brand and remarketing at the end of the journey.
- First click. The opposite. Overstates discovery.
- Linear, time decay, position-based. Rules, not measurement. They split credit by a table rather than by influence.
- Data-driven. The default today. It compares journeys with and without a conversion and derives the contribution of each touch from that. The best available attribution — but still a division of credit, not a measurement of cause.
No attribution model answers what would have happened without the advertising. Only an experiment answers that.
Which is why, at larger budgets, incrementality measurement belongs alongside attribution.
What to do in practice
- Pick one source of truth for the business. It is your shop or CRM. That is what you invoice on and what decides whether marketing earns.
- Use ad data for optimisation. For decisions inside campaigns it is fine, because you care about relative comparison rather than absolute counts.
- Compare regularly, not once. Monthly, the ratio of conversions in Google Ads to orders in the shop. What matters is the trend in that ratio, not its value.
- Unify the definitions. Same period, same currency, same VAT treatment, same handling of cancellations. Most “discrepancies” disappear right here.
- Send cancellations and returns back. Otherwise the algorithm optimises for orders that come back.
- Do not compare months with different numbers of weekends and holidays. It sounds trivial, but in March versus April it does more than you would think.
What to take away
Agreement between systems is not a goal and will never happen. The goal is to know why they differ and to have one number that decides about money — the order in your own system. Advertising platforms are optimisation tools, not accounting, and once a company accepts that, half the arguments in meetings disappear.
Common questions
Why does Google Ads show more conversions than GA4?
Because each counts something different. Google Ads credits a conversion to the date of the click while GA4 credits the date of the conversion itself, which alone creates a gap for longer purchase cycles. On top of that Google Ads may count conversions after an impression without a click, links devices better for signed-in users, and models conversions where consent is missing. A gap in favour of Google Ads is therefore normal, not a fault.
What size of difference between systems is still acceptable?
For shops with a short purchase cycle we treat a gap of roughly ten to fifteen per cent between the ad system and real orders as ordinary. A larger deviation is worth investigating. Clearly suspicious is an ad system reporting a multiple of your orders, which usually means duplicate tracking, or analytics reporting more orders than the shop, which usually means incomplete payments being counted. More important than the absolute size of the gap is whether it stays stable over time.
Which attribution model should I use?
Data-driven, because instead of a fixed table it compares journeys with and without a conversion and derives the contribution of each touch from that. It is the best available attribution, but it is still only a division of credit — it does not answer what would have happened without the advertising. At larger budgets, incrementality measurement by experiment therefore belongs alongside it, especially for brand campaigns and remarketing where attribution systematically overstates.