A GA4 conversion setup that does not lie to you

GA4 will happily report a number for anything. Whether that number describes reality is a separate question, and one worth answering before you move budget because of it.

Step 1: fire events on actions, not page views

The most common broken setup marks the thank-you page view as the conversion. It seems reasonable and it over-counts constantly — someone refreshes, someone bookmarks it, someone hits back and forward, a bot crawls it.

Fire on the actual submit, after your handler confirms success:

gtag('event', 'generate_lead', {
  currency: 'USD',
  value: 50,                      // what a lead is worth to you
  lead_source: 'landing_page',
  form_location: 'hero'
});

Use the reserved names where they exist — generate_lead, purchase, sign_up. GA4 and Google Ads both understand them, so they slot into reporting and bidding without extra mapping.

Step 2: mark key events, and only the real ones

In GA4, Admin → Events → mark as key event. Be strict here. If you mark scroll depth and outbound clicks as key events, your conversion rate becomes meaningless and — worse — Google Ads will import and optimise toward them.

Worth knowing: a "key event" in GA4 is what used to be called a conversion. The rename caught a lot of documentation out, so older guides pointing at a "Conversions" menu are describing the same thing.

Step 3: exclude your own referrals

This one silently destroys attribution. If a visitor goes to Stripe, PayPal or a booking system and comes back, GA4 counts the return as a new session from a referral. Your ad gets no credit; Stripe gets it all.

Admin → Data Streams → Configure tag settings → List unwanted referrals. Add every payment, booking and authentication domain you use.

Step 4: configure cross-domain if you span domains

If your landing page and your booking flow are on different domains, configure cross-domain measurement or one visitor becomes two users with two sessions and no connection between them.

Same screen: Configure your domains. List every domain in the journey.

Step 5: consent mode, before you need it

Under GDPR — and increasingly under the DPDP Act and CCPA — marketing tags must not fire before consent. Consent Mode v2 handles this properly: tags load, but withhold identifiers until consent arrives, and Google models the gap.

// BEFORE the GA4 tag loads
gtag('consent', 'default', {
  ad_storage: 'denied',
  ad_user_data: 'denied',
  ad_personalization: 'denied',
  analytics_storage: 'denied',
  wait_for_update: 500
});

// then, when the visitor accepts
gtag('consent', 'update', {
  ad_storage: 'granted',
  ad_user_data: 'granted',
  ad_personalization: 'granted',
  analytics_storage: 'granted'
});

The default block must come first. A consent banner that fires after the tag has already set cookies is not compliance, it is decoration.

Step 6: attribution — know what you are reading

GA4 defaults to data-driven attribution, which spreads credit across touchpoints. Google Ads reports its own conversions with its own model and lookback window.

They will not match. That is expected, not a bug. Pick one as your source of truth for budget decisions — I use Google Ads for in-platform bidding and GA4 for cross-channel comparison — and never average the two.

The verification checklist

1. DebugView — submit your own form and watch the event arrive with the right parameters.
2. Realtime — confirm the key event increments by exactly one.
3. Load the thank-you page directly. If a conversion fires, you are still tracking a page view.
4. Check the referral list after 48 hours for your own payment domains.
5. Compare GA4 leads with actual inbox leads for a week. If they differ by more than about 10%, something above is still wrong.

That last check is the one that matters. Everything else is process; matching GA4's number against the leads that genuinely arrived is the only proof.