How Does Advertising Work? Mostly Not the Way People Think

The dominant empirical account comes from work by Andrew Ehrenberg and, more recently, Byron Sharp and the Ehrenberg-Bass Institute, and it contradicts a good deal of marketing folklore. Its central findings are that most categories have very few genuinely loyal buyers, that people buy from a repertoire of acceptable brands rather than committing to one, that brand growth comes overwhelmingly from increasing the number of occasional buyers rather than from deepening loyalty, and that the purpose of advertising is therefore to build and refresh memory structures so that a brand comes to mind and is recognised at the point of purchase. On this account a advertisement does not need to convince anyone of anything; it needs to be noticed, linked unmistakably to the brand, and repeated. That explains why so much advertising contains no argument, why distinctive assets such as colours, characters, shapes and sounds matter more than messages, and why brands advertise when they already have nothing new to say.

Measurement improved substantially once large randomised experiments became possible, and the results deflated a number of claims. Studies running controlled holdout groups have repeatedly found effect sizes far smaller than the attribution models the industry was using, with a well-known analysis of a large retailer's search advertising finding that most of the sales credited to it would have occurred anyway. Effects are real, small per person and large in aggregate, which is exactly what one would expect from a mental availability mechanism. Long-term brand-building effects are considerably larger than short-term activation effects and are harder to measure, which produces a persistent institutional bias toward whatever can be tracked, a problem Les Binet and Peter Field have documented in analyses of effectiveness data showing a drift toward short-term measurable activity at the cost of long-term growth. The honest summary is that advertising works, that the average effect is modest, and that the variance between campaigns is enormous.

Several famous stories are worth correcting because they shape public belief about the industry:

• •Subliminal advertising, based on James Vicary's 1957 claim that flashing messages in a cinema raised sales of popcorn and cola, which he admitted in 1962 was fabricated; laboratory work finds some very weak priming effects under artificial conditions and nothing supporting the original claim
• •The idea that advertising creates desires from nothing, whereas the evidence suggests it mostly redistributes existing category demand between brands and has modest effects on category size
• •The belief that phones listen to conversations to target advertisements, which has been tested by researchers examining network traffic without finding evidence, and which is better explained by the genuine precision of behavioural and location targeting and by frequency illusion
• •The assumption that people can identify what influenced them, which self-report consistently fails at, since the mechanism operates through familiarity and ease of recall rather than through remembered arguments
• •The notion that a clever advertisement is an effective one, since awards and effectiveness correlate only loosely and distinctiveness matters more than creativity as the industry uses the word

The technical system is considerably stranger than the creative one. When a page loads, a request describing the available space and whatever is known about the viewer is sent to an exchange, where an automated auction runs among bidders, completes in around a hundred milliseconds, and returns the winning creative, all before the page finishes rendering. That system is built on tracking infrastructure that follows behaviour across sites, which is what privacy regulation and browser changes have been progressively dismantling, with third-party cookies removed or restricted in most browsers and attribution consequently becoming harder. It is also subject to substantial fraud, with automated traffic generating impressions nobody sees, and to a long-running transparency problem in which a meaningful share of what an advertiser spends is consumed by intermediaries, documented in a widely cited 2020 study that could not account for roughly a third of the money in the chains it examined.

Most jurisdictions restrict advertising along a consistent set of lines. Claims must be substantiated and not misleading, which is enforced by a regulator or a self-regulatory body with the power to order removal. Particular products are restricted or banned, with tobacco advertising prohibited almost everywhere, alcohol and gambling restricted by time and content, and prescription medicines advertisable directly to consumers in only two countries. Advertising to children is limited on the reasoning that below a certain age children cannot recognise persuasive intent, which is supported by developmental evidence. Disclosure of paid endorsement is required, which is the rule that influencer marketing repeatedly breaches. Political advertising is the least regulated category in several countries and the most in others, and the arrival of synthetic media has made the existing rules look inadequate almost everywhere.

The evidence supports mental availability rather than persuasion as the main mechanism: advertising builds and refreshes memory structures so that a brand is easy to bring to mind and recognise when a purchase is being made, which is why distinctive colours, characters and sounds matter more than arguments. Controlled experiments find real effects that are small per person and large in aggregate, and long-term brand building outperforms short-term activation while being harder to measure. Subliminal advertising was fabricated by its own originator in 1957.