Your Google Analytics is lying to you. Your Meta Ads Manager is lying to you. Your LinkedIn Campaign Manager is lying to you. Not maliciously — but systematically, and in ways that are getting worse.
The iOS 14 problem (that you probably already know)
When Apple introduced App Tracking Transparency in 2021, it broke the cross-app tracking that Meta's attribution relied on. Meta's reported ROAS dropped an average of 30-40% for most advertisers — not because performance got worse, but because attribution got blinder.
Most teams adapted by using Meta's Conversion API, which does server-side event matching and partially compensates for the tracking gap. But it's still a probabilistic model, not a deterministic one.
The AI search problem (that most teams haven't caught up to)
Here's what happens when someone discovers your product through ChatGPT: they ask a question, get a recommendation that includes your brand, and then — if they're interested — they either search your brand name on Google or type your URL directly. In both cases, the referral source is recorded as "direct" or "organic search."
ChatGPT gets zero credit. Your AI search visibility efforts get zero credit. Your content team, which spent three months optimizing for GEO, gets zero credit.
Plexrio Attribution OS tracks a new event type: AI_REFERRAL. When a visitor comes from a URL that matches known AI assistant patterns — chat.openai.com, claude.ai, perplexity.ai — we capture it as an explicit AI referral event. It's not perfect, but it's dramatically better than attributing everything to direct traffic.
The platform overreporting problem (that nobody talks about)
Every ad platform has an incentive to claim credit for as many conversions as possible. Meta's attribution window defaults to 7-day click, 1-day view. Google's defaults to 30-day. LinkedIn's defaults to 28-day.
If a customer sees a Meta ad on Monday, a Google ad on Wednesday, and converts on Friday — all three platforms claim the conversion. Your reported ROAS across all channels is 3x your actual blended ROAS.
The fix is incrementality testing — measuring the actual lift from each channel by running holdout experiments. Plexrio Attribution OS configures and runs these automatically, so you know what's real.