← ClaudeAtlas

facebook-ads-pixel-and-attribution-auditlisted

Use for "can I trust my Meta conversion numbers", "why does Facebook claim more purchases than Shopify", "are my conversions double counted", "is my pixel firing", "how much of this is view-through", "my conversions dropped overnight", or "how much spend has no tracking" — and whenever someone doubts the numbers, even without the word "audit". Run before trusting any cost or return conclusion about the account. Meta / Facebook Ads only.
coupler-io/skills · ★ 33 · AI & Automation · score 72
Install: claude install-skill coupler-io/skills
# Facebook Ads Pixel and Attribution Audit **Tells you whether you can trust your Meta conversion numbers — before you move budget on them.** Meta's reported conversions are the least reliable number in the account and they fail quietly. A seven-day-click plus one-day-view window credits Meta for people who never clicked. Modelled conversions fill statistical gaps that look like observed events. The pixel and the Conversions API both fire on one purchase and deduplication silently fails. A campaign optimises to a custom event nobody meant to be the goal. None of that looks broken in Ads Manager, and all of it moves your reported cost per acquisition. **What you get back** - **A verdict** — *High confidence*, *Qualified*, or *Not usable* — against stated criteria, so a second run reaches the same word. - **The money you cannot account for.** Spend running with no conversion event attached, as a figure and as a share of the account. - **What is being counted, and how much of it is real.** Event inventory by volume, observed against modelled, and the share of credit that came from a view rather than a click. - **Funnel-order violations** — activations without registrations, purchases without add-to-carts — which is what view-through credit and broken deduplication look like from the outside. - **A defect list ordered by what each one costs**, each marked confirmed or suspected. - **A coverage statement.** Unrun checks are reported as unknown, never as clean. **Read-