Key events and attribution in GA4: why your numbers never match Google Ads

Three systems, three numbers

Google Ads says 42 conversions last month. GA4 says 31 key events from Paid Search. The agency says Google Ads is right; the accountant says neither matches the 27 quote requests in the inbox. Who is lying? Nobody. Three systems measure three different things, with different rules. This lesson shows you the rules, so you know which number to use for which decision.

What a key event is and how you mark one

A key event is an ordinary event you have told GA4 means a business result. You do not create it, you mark it: Admin → Events → the "Mark as key event" toggle, or directly in Admin → Key events. The event must have been sent at least once to appear in the list; otherwise you add it manually by name.

Practical rules:

• Mark few: form submitted, phone click, order, booking. Not scroll, not page_view, not "stayed 30 seconds". If everything is a key event, nothing is.
• Event names are case- and space-sensitive. generate_lead and Generate Lead are two different events.
• Counting method: by default GA4 counts a key event "once per event". For a form someone submits three times by mistake you can choose "once per session". Decide before you compare numbers.

Parameters: what the event carries with it

A generate_lead event without parameters tells you "someone sent something". With parameters — form_name, service, value, currency — it tells you "someone requested a quote for an implant, estimated value €600, from the implants page". That is the difference between counting and understanding.

If you send value and currency on key events, GA4 and Google Ads can optimise for value, not just count — and a form for a €2,000 job is no longer worth the same as one for a consultation.

Attribution: who gets the credit

A customer searches you on Google, clicks the ad, leaves. Three days later they come back from organic Google and submit the form. Who gets the key event?

• Last click: Organic Search. The ad gets nothing.
• Data-driven: GA4 splits the credit between the two touches, in proportions calculated from the behaviour of all your users. It might come out 0.6 for Paid Search and 0.4 for Organic.

Data-driven is the default model in GA4 and the one feeding the standard reports. You can change it in Admin → Attribution settings, and in the Advertising section you can compare models side by side. For small accounts, the difference between models is often smaller than monthly noise; for accounts with many channels, it can change the budget decision.

This is also where you set the lookback windows — how far back GA4 looks for a touch that deserves credit. For acquisition key events (first_visit, first_open) the options are 7 or 30 days; for all others, 30, 60 or 90 days, default 90.

Why GA4 and Google Ads do not give the same number

Point by point — and none of it is a bug:

1. Attribution moment. Google Ads places the conversion on the date of the click. GA4 on the date of the event. A click on the 28th with a form on the 2nd of the next month lands in different months in the two reports.
2. Conversion window. In Google Ads you set it per conversion action (default 30 days for clicks, adjustable up to 90). In GA4, the lookback window is global for the property. Different windows, different numbers.
3. What each one sees. Google Ads also sees conversions on other devices of the same signed-in Google user, and view-through conversions on YouTube or Display. GA4 sees what happened in that browser.
4. Attribution model. Google Ads splits credit only among its own campaigns; GA4 splits it among all channels, including organic and email. Google Ads will almost always show more than the Paid Search column in GA4.
5. Consent modelling. When visitors refuse cookies, both systems fill the gaps through modelling, but with different algorithms and thresholds.

How to compare paid with organic honestly

• Use the same period, the same metric (key events, not sessions) and the same model. Do not compare organic sessions with paid conversions.
• Look at cost per key event on the paid channel and key events per 100 sessions on both. Organic looks free, but it carries the cost of content and time.
• Check Unassigned and Direct before any conclusion: if 25% of key events have no source, any comparison is guesswork.
• Accept that the numbers will never match to the unit. Differences of 10–20% are normal. Differences of 50% or opposite trends mean broken tracking or a double tag.