Amazon's OpenAI Deal and the Behavioral Bias Making AI Invisibility Lethal for Brands
Between April and August of this year, ads served per user per hour on ChatGPT in the United States grew 163%, according to Sensor Tower data cited by Adweek. OpenAI is projecting advertising revenue will cross $1 billion over the next twelve months. And Amazon just partnered with OpenAI to let brands extend their existing ad campaigns directly into ChatGPT through Amazon's demand-side platform.
That 163% figure is the one worth sitting with. It is not a growth rate for a nascent product finding early adopters. It is a signal that AI-mediated surfaces are becoming primary decision interfaces at a pace that makes the transition from desktop to mobile look measured. And Amazon's response to that signal reveals something about the strategic stakes that the ad industry's framing (this is about buying audience reach in a new channel) almost entirely misses.
The real issue is not presence. It is what absence costs.
The Availability Heuristic, Updated for 2026
Daniel Kahneman and Amos Tversky described the availability heuristic in the early 1970s: people assess the probability of an event based on how easily examples come to mind. If a brand is immediately recalled, it feels like the probable choice. If it requires effort to recall, it feels less relevant. This is why advertising at scale worked: repetition built mental availability, and mental availability converted to perceived legitimacy.
The heuristic has been the foundation of brand strategy for fifty years. Reach and frequency was a media buying philosophy, but underneath it was a cognitive claim: the more times someone encounters a brand, the more available it becomes in memory, and the more available it is in memory, the more it converts when purchase decisions arise.
What is changing now is the mechanism through which availability gets built, and what happens when it is absent.
UnAvailability Bias: When Absence Becomes Active Evidence
A recent piece in BehavioralEconomics.com introduces the concept of UnAvailability Bias: the tendency to treat absent information as proof of nonexistence. In an era of information abundance, absence carries different weight than it used to. When we expect comprehensive coverage from a system and it fails to surface something, we interpret that failure as evidence the thing does not qualify -- not as evidence the system is incomplete.
This is the cognitive shift that makes AI visibility structurally different from SEO visibility.
When a brand didn't rank on Google, users understood they were looking at one filter's output. They knew the internet contained more than what Google showed them. The mental model was partial coverage: Google is a tool, it returns results, other tools might return different ones. A brand absent from page one of a search result was invisible in that context, but not actively disqualified. The searcher held open the possibility that the brand existed and was simply not surfaced.
AI assistants operate under a different user mental model. ChatGPT, Claude, and their successors are increasingly perceived as comprehensive knowledge interfaces. When a user asks an AI assistant to recommend project management software, or evaluate procurement vendors, or suggest financial planning tools, and a brand does not appear, the user does not typically think "the AI may have gaps." They think "this brand probably doesn't meet the criteria." The AI's omission reads as a considered judgment, not a technical limitation.
That is UnAvailability Bias in action. And it has a direct structural consequence for brand strategy: the cost of AI absence is not just missed impression share. It is active disqualification at the moment of highest purchase intent.
The 163% Growth Rate Is Not About Advertising
Amazon's DSP integration with ChatGPT is being covered as an advertising story. It is, in a narrow sense: brands can now buy placements in ChatGPT through Amazon's existing ad infrastructure. But that framing obscures what Amazon is actually building.
Amazon's demand-side platform is not primarily an advertising product. It is a first-party data infrastructure that connects product discovery, purchase behavior, and fulfillment into a single signal chain. When Amazon integrates that infrastructure with ChatGPT, it is not just selling impressions. It is building the plumbing that connects AI-mediated intent to retail fulfillment, and in doing so, it is creating a new surface where the brands that are present are legible to the system and the brands that are absent are invisible to it at the point where decisions convert to transactions.
The 163% growth rate matters here because it suggests users are already treating ChatGPT as a discovery and consideration interface, not just an information tool. As that behavior solidifies, the value of being in that interface shifts from media value to something closer to shelf placement. You are not just buying an ad. You are buying visibility at the moment the system is consulted on a decision.
Nokia's CPO and the B2B Dimension of the Same Problem
Nokia's chief procurement officer Sanjay Mehta was recently featured in a McKinsey interview about procurement's evolving role in an era of persistent uncertainty. The core argument: procurement is no longer a back-office cost center. In an environment of supply chain volatility, geopolitical risk, and accelerating technology shifts, the CPO belongs at the decision-making table.
That is a separate story from Amazon and OpenAI, but it connects at the mechanism level in a way that is worth tracing.
Procurement decisions in enterprise B2B have always been filtered by a combination of vendor reputation, past relationship, and institutional knowledge. Sales cycles existed partly to get brands into the mental models of the people making those decisions. A vendor not in the consideration set at the start of a procurement cycle rarely made it into the final selection. Awareness was a prerequisite.
What AI does to that dynamic is move the initial filtering upstream into a system. When a procurement team uses an AI assistant to build a vendor shortlist (and this is happening in more organizations than most sales teams currently acknowledge), the brands that appear in that shortlist are the ones legible to the system. Not the most experienced sales team. Not the most impressive deck. Legibility to AI.
Nokia elevating procurement to the strategic table is a recognition that procurement decisions now involve information synthesis at a scale and speed that changes how vendor selection works. UnAvailability Bias operates in the same space: a vendor not surfaced by the AI tool the CPO's team uses does not just lose an early-stage opportunity. They lose it in a context where the user has been conditioned to treat the AI's output as comprehensive, not partial.
Brand Strategy Built for Frequency Cannot Win on AI Surfaces
Twenty years of brand-building wisdom, including what Branding Strategy Insider describes in its retrospective on twenty years of brand development principles, has centered on reach, repetition, and differentiation. The frameworks that emerged from that era, some of them still operationally dominant, assumed a world where the consumer was making active comparisons and needed repeated exposure to build mental availability.
AI surfaces break several assumptions in that model simultaneously.
The first is frequency. Traditional availability heuristic strategy relied on repeated exposure to build recall. AI interfaces don't reward brand frequency. They reward information density, consistency, and semantic coverage. A brand that produces less content but is consistently cited, consistently associated with specific use cases, and consistently described in ways the AI can pattern-match has better AI availability than a brand that runs high-frequency campaigns but produces shallow semantic signal.
The second is differentiation. Conventional brand differentiation advised standing out in categories. AI interfaces don't just sort by category; they sort by relevance to the specific decision being made. A brand that has built broad category presence but thin specificity around particular use cases may appear less in AI results than a competitor with narrower category presence but deeper decision-relevant content.
The third is the relationship between brand spend and AI presence. High advertising spend builds availability in attention-economy surfaces where frequency matters. AI surfaces are not attention economies. They are inference economies. Brand spend that does not translate into semantic signal, citations, and legitimate information coverage does not compound into AI availability. The two accumulation mechanisms are structurally different.
The Behavioral Dimension the Sports Betting Parallel Illustrates
The rise of sports betting debt in the U.S. is superficially unrelated to AI brand visibility, but it shares a behavioral architecture. Sports betting platforms are specifically designed around variable reward schedules -- intermittent reinforcement -- because variable rewards produce stronger habitual engagement than consistent ones. The platforms optimize for availability in the user's attention architecture by engineering the reward pattern.
This is availability-by-design, in the traditional sense: the platforms want to be the first thing that comes to mind when the user thinks about entertainment or opportunity. It works, and the debt numbers confirm it works too well.
What makes AI visibility different is that the design pattern that creates availability in attention economies does not transfer. You cannot engineer availability in an AI inference system the same way you engineer it in a social feed or a betting app. The AI is not responding to reinforcement schedules. It is responding to information quality and semantic coverage. Brands that have spent the last decade optimizing for attention economy availability have built muscles that don't train the relevant system.
This is the strategic discontinuity that Amazon's OpenAI deal forces into view. Not "we need to be where the users are." That framing has always been true and always will be. The harder question is: what kind of presence actually registers as legible and credible inside an AI inference system, and what kind of presence is invisible to it regardless of how much brand equity it represents in other contexts?
What This Changes for Brand Investment
The practical implication is not that brand strategy built on traditional availability is worthless. It is that AI-era brand strategy requires a second accumulation track that operates on different inputs and compounds differently.
The first track is familiar: reach, frequency, distinctive assets, emotional resonance. These still drive purchase behavior in contexts where humans are doing unmediated evaluation. They matter for categories where AI assists but does not filter.
The second track is newer and less codified: information legibility, semantic specificity, third-party citation, consistency of positioning across AI-indexed surfaces. This track is what determines whether a brand gets surfaced when AI systems are doing the filtering. It does not replace the first track; it runs alongside it, increasingly determining whether a brand even enters the consideration set.
The implications for brands in AI-mediated categories -- and that category is expanding faster than most brand teams are tracking -- are similar to what happened with agentic AI in the advertising ecosystem but further upstream. It is not just about ad placement. It is about whether the brand is structurally visible to the decision architecture.
Amazon's DSP integration with ChatGPT is one answer to the second-track problem: paid placement that ensures presence. But paid placement is a floor, not a ceiling. The brands that are building genuine semantic legibility are the ones that will be surfaced whether or not they are buying the placement. The ones relying solely on the Amazon/OpenAI ad pipeline are renting AI visibility rather than building it.
Nokia's CPO being at the table is a recognition that supply chain decisions cannot be made by people who only show up at the end. The same logic applies to AI brand strategy. The decisions about how a brand is represented in AI-indexed surfaces, what it is consistently associated with, and how its claims hold up to AI evaluation are not marketing execution decisions. They are positioning decisions that belong at the table early.
The Compound Nature of AI Visibility
One more thing the Amazon deal reveals: AI visibility may compound differently from SEO or share of voice, because the inference architecture learns from what it has already surfaced.
Behavioral data has a half-life in AI systems that is different from traditional purchase intent signals. A brand that earns consistent positive surfacing in AI responses builds a reinforcement loop that makes future surfacing more likely. The reverse is also true: a brand that is absent from AI responses, or that appears in contexts that suggest low credibility, is compounding the cost of that absence.
UnAvailability Bias means the first miss is already expensive. The compounding means the cost of sustained absence grows faster than linear.
Whether $1 billion in OpenAI advertising revenue over the next twelve months represents a structural shift or a transitional bubble is an open question. But the behavioral architecture underneath it, the cognitive shift in how users interpret AI-mediated absence, is not a function of the advertising market. It is a function of what users now expect from AI interfaces, and those expectations are solidifying faster than most brand strategy teams are adjusting to account for them.
The brands that figure this out now will not just be present in the systems where decisions are being made. They will be present in the architecture of how decisions get made. That is a different kind of advantage than share of voice, and it does not respond to the same interventions.
If you're thinking about how your brand investment strategy maps to AI-mediated decision surfaces, our research on decision intelligence and agentic commerce covers the measurement frameworks that actually track what matters.