What is view-through attribution (VTA)?

Why VTA exists

Most advertising that works is never clicked. A user sees a brand, absorbs it, and acts later through search or directly. Click-only measurement records none of that and concludes the exposure did nothing, which is why upper-funnel media has always struggled to justify itself in digital reporting.

VTA is the attempt to capture it: if a user was served an impression and converted soon after, credit the impression.

Why it over-claims

The mechanism does not distinguish influence from coincidence. A user who was going to convert anyway, and who happened to be served an impression that day, produces exactly the same record as one who was genuinely persuaded.

Because heavily-targeted campaigns concentrate impressions on users already close to converting, VTA credits are densest exactly where the counterfactual is weakest. Retargeting with a long view window can appear to drive enormous volume while a holdout shows very little causal lift.

There is also a visibility problem beneath it: an impression counted as served is not necessarily an impression a human saw, which is what viewability standards exist to constrain.

Using VTA defensibly

• Keep the window short. 24 hours is the common ceiling. Multi-day view windows credit ambient conversion.
• Report it separately. Never blend view-through and click-through into one conversion count.
• Require viewability. Crediting an impression that never entered the viewport is indefensible.
• Calibrate against a holdout. Once you know the true lift, you can size what proportion of VTA credit is real, and apply that knowledge consistently.

The AI parallel

AI media has the same structural property as view-through, in a sharper form: a recommendation inside a conversation frequently produces action with no click. The exposure was real and influential, and the click-based record is empty.

The difference is that VTA's failure is over-claiming while the AI case is under-counting. But the lesson transfers exactly — an exposure without a click can be genuinely causal, and the only instrument that separates influence from coincidence in either case is a holdout, not a longer window.