Meta's attribution models decide how conversion credit is distributed across ad touchpoints in your customer's journey. When someone sees or clicks your ad then converts later, the model determines whether that ad gets counted as driving the sale. Meta deprecated its 28-day click window in 2021 and now defaults to 7-day click / 1-day view, meaning conversions get attributed if the click happened within 7 days or the view within 1 day before the conversion event. You can adjust attribution windows in Meta Ads Manager under your conversion event settings. Available options include 1-day click, 7-day click, and view-through windows of 1 day. Shorter windows like 1-day click give you cleaner data less affected by organic traffic or other channels, but they'll show lower conversion volumes because you're excluding people who saw your ad on Monday and bought on Friday. Longer windows capture more of the actual influence but muddy the waters with conversions that might have happened anyway. Meta uses last-click attribution by default, meaning the final ad interaction before conversion gets full credit. This differs from Google Analytics 4's data-driven model that splits credit across multiple touchpoints. The practical tension: Meta will claim credit for conversions that GA4 attributes elsewhere, especially when someone clicks a Meta ad, then later Googles your brand and converts. Neither is lying, they're just using different counting rules. At Ottawa SEO we typically run 7-day click as the primary reporting window for clients spending under $10K/month CAD because it balances signal strength with realistic influence. For retargeting campaigns we'll sometimes check 1-day click to isolate bottom-funnel performance. The key mistake is comparing Meta's attributed conversions directly to your Shopify order count or GA4 conversions without accounting for these model differences. Track your blended ROAS across all channels and use Meta's attribution as a directional gauge, not absolute truth. iOS 14+ privacy changes have made all attribution fuzzier, so your model choice matters less than consistent measurement over time.