The ads you skip still shape what you buy. Last-click attribution cannot see them.

In 1968 Robert Zajonc showed people Chinese characters, nonsense words and photographs, then asked them to rate what they saw (Journal of Personality and Social Psychology). They preferred the items they had seen before, even when they could not say they had seen them. Robert Bornstein's 1989 meta-analysis of more than 200 studies from 1968 to 1987 found the effect held across stimulus types, exposure lengths and groups of people (Psychological Bulletin). Zajonc's 2001 review went further: exposure below conscious awareness can produce a stronger preference than exposure people notice.

The mechanism is processing fluency. A familiar thing is easier for the brain to process, and that ease feels like liking. You do not think that you have seen the logo before. You think you like the brand. Because the feeling is misattributed, conscious judgment does little to override it.

Familiarity is not the whole story. Meaning, context and emotion matter in how brands are chosen, and Les Binet and Peter Field's analysis of the IPA Databank found that emotional, broad-reach brand campaigns deliver the largest business effects over time. The two work together: a customer cannot form an emotional bond with a brand they have never been exposed to.

Mental availability, not persuasion

Byron Sharp and the Ehrenberg-Bass Institute define mental availability as the probability that a brand comes to mind in a buying situation. In How Brands Grow (2010), Sharp argues that brands grow mainly by reaching more buyers and staying easy to recall, not by deepening loyalty. On that view most advertising is not there to persuade. It builds and refreshes the memories that make a brand easy to notice when someone is standing in an aisle or typing into a search box. Liking the ad is optional.

This is not an argument for bad ads. The mere exposure effect works best on stimuli that start neutral or slightly positive. If an ad provokes real hostility on first contact, repetition strengthens the dislike.

Attention modulates the effect too. A 2018 study in Frontiers in Psychology found that the shift in attitude was stronger when viewers attended to the image, but it persisted under peripheral processing (study). Raoul Bell and Axel Buchner (Journal of Interactive Marketing, 2018) found that mildly disruptive advertising improved preferences, and that the effect turned negative once the disruption became real annoyance.

So the range that does the most work is narrow: ads that are a little loud, a little awkward, intrusive enough to be processed, and not offensive enough to be rejected. The split-screen sponsor ad in the middle of a live match that everyone complains about is often in that range. People dislike it, and the brand sticks.

Standard digital attribution counts conversions inside a 7-day or 28-day attribution window and gives the credit to the last touch. The real journey is longer. A TV spot in week one, a skipped YouTube pre-roll and a podcast mention in week two, a social scroll in week three, then a retargeting ad and a branded search that ends in an order. The window sees the last two steps and nothing before them.

Brainlabs ran 17 Meta conversion-lift studies for 12 advertisers and found that paid social drove a 19% incremental increase in search visits; 71% of that extra search traffic was organic (Brainlabs). Those visits, and the orders that followed, were credited to search, not to the Meta ads that caused them. Meta's 2023 Measurement 360 report puts the undervaluation of upper-funnel ads under last-click attribution at up to 40%. A 2022 MediaScience study found that a TV ad shown before a digital ad gave 125% higher unaided brand recall and 18% higher purchase intent than two digital ads alone.

The loop that starves the top of the funnel

The result is a structural feedback loop. Channels that create demand look expensive and unproductive. Channels that capture it look efficient. Budget moves from creation to capture, reach falls, share of voice falls, and some weeks later revenue follows. The dashboard then blames the performance campaigns, which were only riding momentum the brand stopped building. Teams that cut awareness spend and then watch branded search volume drop have seen this loop run.

Performance channels depend on brand channels. Retargeting and branded search mostly collect demand created somewhere else, often by the channels attribution cannot see. Cut the upper funnel because it does not convert inside the window, and the lower funnel slowly stops working too. Scott Brinker and Frans Riemersma's 2026 State of Marketing Attribution report describes the needed shift as moving from attribution as credit allocation to attribution as direction: not which channel gets the credit, but what to do next for each customer. Three changes follow.

1. Measure lift and value, not events in a window

Count what a channel added with an incrementality test: a holdout group that does not see the spend, compared with one that does, in your own orders. Then judge customers by predicted lifetime value over 90 days and 12 months. If your measurement window is shorter than your customer's decision cycle, you cannot see the touchpoints that matter most.

2. Capture the signals the pixel loses

A browser pixel misses visits whenever ad blockers, browser limits or declined consent get in the way. Then the model is not wrong about which channel works; it does not know the customer was there. Server-side tracking records events where they happen and sends them on, so the data you judge channels by is complete.

3. Give every channel the same goal

Ad tools optimise for the cheapest next conversion and email tools for the next open. Neither optimises for what the customer is worth over time. One objective across site, messages and ads, such as predicted customer value, stops each channel claiming orders it did not cause. The decision intelligence guide explains how that works in practice.