Attention In Minds & Models

For years, attention has been a human problem: did anyone notice the ad, remember it, choose the brand? That question still matters, but it’s no longer the only one. People are now handing parts of the buying journey to AI, and those models decide what to recommend based on what attention leaves behind - the searches, reviews, conversations and shares. Dull advertising now costs you twice. It wastes the media spend, and it leaves AI with nothing to find. Check out the latest article from our IAB New Zealand Measurement Working Group about attention measurement, and how to avoid the cost of dull. 

Attention has been the industry's most discussed effectiveness topic for several years, and it has historically been treated as a human problem: can we get people to notice, remember and choose our brand?

That remains critical, but the framing is now incomplete. Consumers are beginning to delegate parts of the discovery, comparison and decision journey to AI models, and the behaviours attention produces, the searches, reviews, conversations and shares, are the same material those systems draw on when they decide which brands to recommend.

Advertising that goes unnoticed does not simply waste the impression. It leaves nothing behind for a model to find.

The commercial risk is therefore twofold. Poor human attention creates a Cost of Dull, where media spend fails to build memory, consideration or preference. Poor machine visibility creates Signal Debt, where fragmented, missing or weak evidence prevents AI systems from confidently surfacing the brand.

The next advantage in marketing will come from managing both. Brands that are salient to people and discoverable to AI compound attention, trust and recommendation over time. Brands that ignore either system risk paying twice: once through wasted media spend, and again through exclusion from AI-mediated consideration.

This piece sets out where attention measurement now stands, what the evidence says poor attention costs, and how both systems should shape planning in a market the size of New Zealand.

Measuring Human Attention

Until recently, marketers relied on proxies - impressions told us an ad had been served, viewability told us it had the opportunity to be seen, dwell time suggested an audience may have remained in view. However, none of these could tell us whether somebody was paying attention. The IAB & MRC’s Attention Measurement Guidelines, published in November 2025, delivered a framework that provides the clearest foundation yet for measuring attention consistently. (IAB and MRC Attention Measurement Guidelines, 2025)

Developed through a collaboration involving more than 200 practitioners and supported by industry bodies across multiple markets; its most useful contribution for planners is the separation of attention into three components: content, placement and creative. That allows underperformance to be traced to the environment, the execution, or the work itself.

The framework recognises several measurement approaches, each with distinct strengths and limitations.

(IAB and MRC Attention Measurement Guidelines)

These methodologies are not interchangeable. Different providers use different data sources and scales, so the objective is not to compare scores between vendors but to establish a consistent basis that can be tracked over time.

One principle underpins the rest: attention is an input, not an outcome. It is neither a currency nor a business result. It is a measure of exposure quality and of the likelihood that an exposure contributes to memory, consideration and future behaviour.

What poor attention costs

The Cost of Dull Media: Down Under Edition, produced by Amplified, eatbigfish and Peter Field through OMG Australia, measured Attention Volume, the proportion of available attention that advertising actually captured, and grouped environments into three tiers. (The Cost of Dull Media - Down Under, 2026)

Advertising in the lowest-attention environments generated substantially weaker brand outcomes, with some measures showing more than ten times less uplift. In the dullest environments half the audience has gone before the third second; in the strongest, almost everyone stays long enough for a message to register.

The study estimates that advertisers lose approximately 59 cents of every dollar invested in low-attention environments, rising to almost 90 cents for challenger brands. Established brands can absorb some inefficiency on the strength of accumulated memory structures. Challengers, who depend on building mental availability, cannot.

Two findings are worth holding onto. Format matters more than most planners assume; brands of very different sizes produced remarkably similar attention profiles within the same formats, which means creative cannot rescue an environment that generates no attention in the first place. Poor attention reduces effectiveness rather than merely reducing efficiency. In some of the lowest-attention environments the study observed negative brand uplift.

The study also describes a failure mode it calls efficiency blindness. Campaigns keep producing clicks, leads and short-term results while underinvesting in the attention that sustains future demand. Existing demand is harvested, the dashboards stay green, and the erosion of mental availability only becomes visible much later.

The link to model exposure

Dull environments do more than waste money - they produce no evidence of human engagement.

Consider what a high-attention exposure generates beyond the exposure itself. Some proportion of the people who notice an ad go on to search for the brand, read about it, mention it, review it, discuss it or share it. Those behaviours leave traces across the open web. A low-attention exposure produces none of that. The impression is served, nobody engages, and nothing is left behind.

Those traces are the raw material AI systems use. Web content, reviews, editorial coverage, comparison articles, FAQs and community discussion are what a model draws on when somebody asks it to evaluate options. Advertising that is never noticed contributes nothing to that body of evidence.

This is where the cost compounds. The 59 cents is the immediate loss - the second loss is every search, review and conversation a noticed exposure would have generated, each of which would have made the brand easier for a model to find and recommend later. We would put this forward as the working group's position rather than a measured finding. The causal chain is well supported at each step, but the compounding effect has not been quantified.

What models do with those signals

Between awareness and purchase sits a stage of evaluation: comparing options, reading reviews, checking specifications, looking for proof. What we are seeing is a collapsed funnel, in which that evaluation is increasingly delegated to a model.

This changes what needs measuring - the relevant questions become how often and how accurately a brand appears in AI-generated answers, whether models link to its assets, and how closely its content matches the intent behind a query. The emerging term for this is Share of Model, the successor to Share of Voice and Share of Search.

Hidden code and keyword stuffing do not work here, a brand cannot metadata-tag its way into a model's good graces if people fundamentally dislike it, because the signals models rely on are largely signals of trust. Brands generating consistent, discoverable, independently validated signals accumulate advantage. Brands with fragmented or contradictory signals create uncertainty that people and machines both struggle to resolve.

The loop closes on itself. Human attention produces clicks, dwell time and reviews. Those behaviours build an organic digital footprint. That footprint raises the likelihood of citation and awareness shapes the questions people ask in the first place. A brand that has earned saliency is more likely to be named in the query itself, which tends to produce more accurate and better-aligned answers about it. At IAB NZ's MeasureUp 2026, Google and Tracksuit argued that brand and trust still matter when an AI agent recommends options, and highlighted branded query growth as a useful proxy for sales.

Signal Debt

Most organisations measure the cost of missing a conversion. Few measure the cost of missing a recommendation. When a brand is absent from an AI shortlist there is no impression, no click and no attribution trail. The customer may never know the brand was an option, and the organisation may never know it was excluded. The Optimisers' NZ AI Answers Project, shared at IAB NZ's Discovery: AI & Search Summit, found that the top AI answer did not reflect real market share and varied by engine. A Share of Model baseline should therefore cover more than one AI engine.

The losses take several forms; competitors become the default answer, the model finds the brand but lacks the evidence to recommend it confidently, strong claims fail to surface because nothing outside the brand's own site reinforces them, and competitors keep accumulating citations while the brand's footprint stagnates. That last point cuts hardest for incumbents holding ground against fast-moving challengers.

Unlike paid media inefficiency, these losses compound. Each month of inaction adds to the evidence AI systems hold for somebody else. This is the Signal Debt described at the outset: missing, fragmented or unvalidated signals that stop AI systems finding, understanding or trusting a brand. Efficiency blindness and Signal Debt are the same failure in two systems, a loss that does not show up in the reporting until it is expensive to reverse.

What to do

Move past proxy measures. Viewability and watch time establish opportunity to see. Human-verified attention measures should increasingly sit alongside them.

Audit investment through an attention lens. The Australian findings suggest a significant share of spend still flows into environments generating very little attention. The question worth asking is where budget is going that attention is not.

Treat signal work as infrastructure. Give every important claim, fact or answer a single clear owner, since variations of the same answer across multiple places reduce confidence for readers and models alike. Write key numbers and differentiators as standalone, quotable statements rather than burying them in prose. This work rarely sits with one team, spanning communications, PR, search, web and product.

Invest in independent validation. What a brand says about itself still matters, but only a small share of what models draw on comes from owned assets. Models cross-reference the wider web to verify claims, so reviews, expert commentary, editorial coverage and community discussion carry disproportionate weight, and can be influenced through PR and creator collaboration.

Establish a Share of Model baseline. Several tools now measure this, with varying coverage, accuracy and price. The category is young and moving quickly, so choosing one and starting to track beats waiting for a settled standard.

The New Zealand position

Ours is a market of small audiences, finite budgets and little room for inefficiency, which raises the Cost of Dull relative to larger markets. New Zealand is not short of high-attention environments. Premium video, BVOD, cinema, content-led partnerships, quality local publisher inventory and a growing set of high-impact digital formats are all available here. The constraint is scale, in that a smaller ecosystem limits how far budgets can be re-weighted into premium environments before reach suffers. The task is identifying where higher-attention environments have the greatest impact, not abandoning lower-cost channels.

The same size works in our favour on the signal side. Word of mouth, reviews and local reputation carry disproportionate weight in the material models draw on here, and a comparatively modest volume of local commentary can represent a meaningful share of what exists about a brand in this market. With fewer degrees of separation between channels and audiences, authentic local engagement produces a more concentrated footprint than the same effort would achieve in Australia or the US.

In Summary

Attention was treated as a human problem for years, and the evidence that poor attention carries a measurable cost is now solid. What has changed is that the same signals building awareness, trust and preference among people also shape what machines can find, understand and recommend.

Human attention creates digital signals. Digital signals shape AI recommendations. AI recommendations influence discovery to decision. The work of getting noticed now has a second payoff, and the work of being ignored has a
second cost.

Thank you to the members of the IAB New Zealand Measurement Working Group who co-created
this article: 
James Cartwright, Sales Director ANZ & NZ Country Manager, Azira
Tim Richardson, Client Partner NZ, InfoSum
Nicola Yates, Senior Marketing Manager & Chapter Lead Media & Planning, Woolworths New Zealand
Q Naim, Chair, Data, Privacy & Measurement Council. Director, Team Circle