Before any campaign goes live, client or our own, we run a tracking audit. Ad spend is only as smart as the conversion data behind it, so we check the data first.
This time the site was ours. We were preparing a small Meta lead-gen campaign for convertivemedia.com, and the audit caught four issues. None of them threw an error. Everything looked fine in the browser and in Ads Manager.
That's typical. The tracking problems that cost the most money are rarely the ones that break loudly. They double-count, mislabel or quietly drop people, and Meta's algorithm optimizes toward whatever it's given.
Why AI-built sites need this check
Our site is built with Lovable, an AI website builder. It's a fast way to ship pages, and we'd recommend it. But it highlights a pattern we're seeing more often.
Ask an AI builder to "add the Meta pixel" or "add a contact form" and it will. It will also make a series of reasonable-looking decisions on its own: which events fire, when they fire, how spam is filtered and which visitors get tracked. Three of the four issues below came from decisions like that. They weren't wrong in isolation. Nobody had reviewed them against how the ads would actually use the data.
The same applies to sites built on page builders, Shopify themes with years of app installs, or anything a former freelancer set up.
1. Automatic Events mislabeled our calls to action
What we found: Every click on our "Book a growth review" button was logged as a SubscribedButtonClick. Meta's Automatic Events feature watches a site and guesses what its buttons do, and it guessed wrong. Since August 3, 2026, Automatic Events is on by default for every new pixel.
Why it matters: Meta says it dedupes overlapping events, so this won't necessarily inflate your purchases. But it clutters reporting, and it's easy to build a custom conversion or audience on a mislabeled event without realizing it.
How to check: In Events Manager, open your pixel's Test events tab and click through your site. Anything marked "Automatically logged" came from Meta's detection. If you track your events deliberately, turn Automatic Events off in the pixel's Settings tab.
2. One form fired two conversions
What we found: Our contact form sent a Lead and a Schedule event on every submission. There's no booking calendar on the site, so the Schedule event had nothing behind it.
Why it matters: Double-counted conversions make results look better than they are and teach the algorithm the wrong lesson about who converts. On Shopify, the most common version is a store with the Facebook & Instagram app installed plus pixel code pasted into the theme. Every purchase can be counted twice, which inflates reported return on ad spend.
How to check: Complete one real action, a purchase or a form submit, and watch Test events. It should fire exactly once.
3. The spam filter was likely blocking real people
What we found: On some browsers, clicking "Request my growth review" did nothing. No error and no thank-you page. The form uses a honeypot, a hidden field that only bots should fill in. Browser autofill and password managers are known to fill hidden fields too, which makes the form treat a real person as a bot and stop silently. That's the most likely cause here.
Why it matters: The pixel never fires because the form never finishes. You lose the lead and the conversion, and nothing in your reporting tells you it happened.
How to check: Submit your own forms the way customers do, with autofill on, on desktop and mobile. If you don't reach a confirmation, some of your visitors don't either.
4. A consent check defaulted to "don't track"
What we found: For privacy compliance, the site checked each visitor's location before loading the pixel. When that lookup took longer than two seconds, the code blocked the pixel by default.
Why it matters: Some US visitors on slower connections were never tracked. Smaller audiences, fewer attributed conversions and campaigns that look weaker than they are.
How to check: If you use a consent banner or geo tool, find out what it does when it can't decide. For US traffic, the fallback should usually be to track.
The takeaway
Meta's algorithm can only optimize toward the signals it receives. Phantom conversions send it looking for more phantom buyers. Missing conversions make working campaigns look broken, so they get cut.
The audit took an afternoon, and each fix took minutes. That's the case for checking before you spend, not after.
If you'd like us to run the same audit on your account, that's what our conversion tracking service is for.