AI platforms are under increasing pressure to find a sustainable revenue model. Around 90-95% of ChatGPT users do not pay for the product. Google is urgently searching for a successor to its search advertising revenue. Perplexity and other emerging players need a dependable commercial engine to keep funding their growth.
Every AI platform is now starting to look towards advertising as the answer. It is a tried and trusted value exchange that has powered the internet for more than twenty five years: free services in return for attention. For decades it has created a stable equilibrium between user experience and commercial pressure. Now every AI platform is attempting to recreate that playbook inside an environment that works nothing like the open web.
Old Logic, New Medium
The problem is that AI platforms are trying to apply the same old thinking to a completely new medium. They are looking to wedge ads into conversational interfaces in the same way we once wedged ads into webpages and search results. The formats look familiar. The logic looks familiar. The compromises look familiar. And if they continue down this path, the same relevancy issues that have plagued digital advertising for two decades will reappear inside AI platforms.
At their core, these relevancy issues have always stemmed from the same pattern. Ads are shown because they can be shown, not because they should be shown. Despite the advertising industry’s best efforts, we rely on loose proxies for intent, imperfect targeting data, and bidding systems that reward the highest payer rather than the most suitable option. This means consumers are routinely interrupted with messages that have nothing to do with what they want, where they are, or what they are trying to achieve in that moment.
Put simply, AI platforms are in danger of treating advertising as a UI problem rather than a reasoning problem. It’s a strategic failure in the making. They are sitting on something the advertising industry has never had before: an intelligence that genuinely understands what a user is trying to do. Not a keyword. Not a demographic bucket. Not a retargeting signal scraped from another tab. A real-time, contextual interpretation of intent. Yet instead of using this intelligence to fix advertising, they are just trying to put display ads inside an assistant. This is incredibly one dimensional thinking.
Let The Model Think About The Ad
The opportunity AI platforms are missing is to let their models think about the advertising. Allow the model to evaluate it in the same way it evaluates any other piece of information. By introducing paid advertising opportunities into the reasoning loop, AI models can judge whether it is relevant to the task, the context, and the user’s stated or implied intent. This idea works because of what large language models already do - they evaluate information, filter options, and assemble an answer.
Paid advertising opportunities would not look like ads in the traditional sense. They would be structured inputs, including product information, visual assets, messaging, and a URL. This turns advertising into a format the model can reason with so it can decide if it’s relevant, doing everything the advertising industry has spent twenty years trying to do manually.
The flow should be simple: advertisers provide structured inputs. The model evaluates them. Irrelevant options drop out. Relevant ones surface with full disclosure.
To be clear, paid advertising opportunities would exist for one purpose only: to be evaluated by the model for relevance. The model decides whether the advertising content informs the answer. If it doesn’t, the user never sees it. If it does, it appears in the output with full disclosure and a clear citation. This mirrors how AI platforms cite sources in search-style answers today - the mechanism already exists, they can simply be extended to paid content.
This approach requires a simple, transparent rule set. Any paid influence must be cited. If a sponsored option contributes to the output, the user sees that immediately. If a sponsored option is considered but rejected, the platform can log that privately so advertisers can understand why and improve future ads.
Most importantly, the integrity of the AI platform must be non negotiable. Putting ads on the surface invites scepticism. Putting ads into the model’s reasoning, with full disclosure, invites relevance. One is a dark pattern. The other is a design pattern.
A Healthier System for Everyone
Advertising inside the model flips the dynamics of advertising completely. Brands would not win just because they outbid competitors. They would win because their messaging genuinely matched the user’s intent as interpreted by the model. If instead AI platforms move advertising into the reasoning layer, they unlock something the industry has never had. True contextual relevance at the moment of intent. Ads that are additive. Ads that are filtered by intelligence rather than interrupted by design. For the first time, advertising would be additive rather than interruptive and has the additional benefit of not adding unwanted clutter to the user experience that could break flow, annoy consumers, and erode trust. This is a healthier system for everyone involved.
This also creates a new commercial opportunity. A cost per thought when the model evaluates a paid advertising opportunity. A cost per impression when that option is surfaced. A cost per click if the user takes action. It’s a clean hierarchy that mirrors the actual funnel of the AI model’s reasoning that will give advertisers flexibility, allow them to control budgets around different stages of decision making, and align incentives around quality instead of intrusion.
If AI platforms get this right, they can build a new kind of advertising system. One built on intelligence, context and intent. One that rewards relevance as a first principle. One that gives users better answers and gives brands better opportunities.
The future of advertising will not be won by whoever finds the cleverest place to stick an ad in a chatbot. It will be won by whoever lets their model put the most relevant content in front of consumers.
“The future is already here, it’s just not evenly distributed.“
William Gibson
Great piece Sean, thank you. In some ways your post and the last two paragraphs describe beautifully the influence, loyalty and value (which translated into significant profitability) high quality consumer magazines once enjoyed. Across fashion, music, specialist automotive, homes and gardens, destinations and many more; advertising wasn’t an interruption, it was inspiration, where editorial and advertising often merged accepted with open arms by loyal audiences. I wonder based on your thinking the AI players might have no choice but evolve their own multi-dimensional versions of these proven past paper giants? Of course the AI players would have to establish the environment where powerful editors and their teams flourish. Back to the future?
As an example, a user asks "Where can I buy a red jumper?" The model refers to its ad base. Ah, an ad for a red jumper. Perfect. Here's where you can buy a red jumper. I suppose the question is, if there is no ad for a red jumper, will the model scan the web for red jumpers, and potentially suggest the same brand. If so, why advertise? If the model only shows your ad when it's relevant to the conversation, would it not pull up the same product without the need for ads?