After 16 years running PPC campaigns, you develop a feel for where attention eventually monetises. OpenAI's move to test ads in ChatGPT was inevitable. The question was never if, but how.
On the surface, the mechanics feel familiar. Keyword-linked banners appear at the bottom of AI responses, charged on ad views, clearly labelled and separated from organic content. Anyone who has lived inside Google Ads for long enough recognises the structure instantly. But the operating environment changes everything. This is not search with a different interface. It is a different relationship between intent, relevance and trust.
The Economics Driving the Shift
OpenAI is not making this move from a position of excess. The compute costs are substantial and increasingly difficult to sustain on subscriptions alone, particularly while expanding access to free and lower-tier users.
The financial pressure is real. But the path through it is not simply increasing ad density and trusting relevance scoring to clean up the mess. That approach might work in the short term, but it breaks the product. The value of ChatGPT is not traffic. It is trust. And unlike search, that trust is not optional to performance. It is structurally required.
Why Keyword Matching Is No Longer Enough
In Google Search, keywords operate as a proxy for intent. Users understand they are entering a commercial marketplace. Ads sit alongside organic results in a way that feels, if not neutral, then at least structurally honest.
In ChatGPT, language is not a proxy for intent. It is the intent. Users articulate problems, uncertainty and context in full sentences. A keyword in that environment is less a trigger and more a loose label on a much richer conversation.
This is where the model fundamentally diverges. Even if OpenAI charges on ad views, the decision logic upstream of that view is entirely different. An AI system decides whether an ad deserves to exist in the response at all. That creates a pre-auction filter Google does not have.
In search, relevance is scored after you enter the auction. In AI, relevance determines whether you are invited into the conversation in the first place. Two advertisers can bid on the same keyword, pay the same CPM, and receive very different value depending on whether their message genuinely advances the user's understanding. If the ad feels extractive or interrupts the flow, the system learns to suppress it.
You are no longer optimising to win impressions. You are optimising to earn inclusion.
What Semantic Relevance Actually Looks Like
Anyone who has managed accounts at scale has seen the gap between optimising for systems and optimising for humans. In Google Search today, relevance is still largely mechanical. Query intent inferred from keywords, Quality Score signals, historical CTR, conversion likelihood. You can win by structuring well and bidding aggressively.
Semantic relevance in an AI environment works differently. The system asks whether this genuinely helps answer the user's question. That shifts optimisation from matching terms to matching meaning.
In practice, this shows up in three ways.
Context replaces keywords. Instead of targeting "best CRM software", you are competing to be the most useful recommendation inside a broader problem space. Eligibility is not determined by a keyword list, but by how clearly your product, proof points and content map to that scenario.
Usefulness becomes measurable. Rather than CTR as the primary signal, AI platforms can observe downstream behaviour. Did the user ask a follow-up question? Did they engage with the recommendation? Did it reduce uncertainty or accelerate a decision? This is closer to decision quality than click efficiency.
Brand credibility becomes a ranking factor. In search, brand helps, but you can often outbid or out-optimise it. In AI-mediated environments, trust is foundational. Independent validation, consistent messaging and clear value articulation all influence whether an AI chooses to surface you at all.
The Trust Variable Competitors Are Already Flagging
Google DeepMind CEO Demis Hassabis has said he is "surprised" OpenAI has moved to introduce ads this early. His concern centres on how carefully monetisation needs to be handled when chatbots are meant to function as helpful digital assistants.
Having worked through multiple platform trust cycles, the pattern is familiar. Platforms rarely lose trust because of one decision. They lose it through cumulative misalignment between user value and revenue incentives.
What makes this moment different is that trust is not merely a brand attribute. It is a functional dependency. AI systems perform worse when users hedge, withhold context or second-guess motives. That creates a direct economic penalty for eroding trust.
Research already shows that while a majority of daily generative AI users report high trust in both the technology and its outputs, that trust is fragile. Small shifts in perceived intent can change usage behaviour quickly.
Where the Real Commercial Value Sits
The ServiceNow partnership signals something important. OpenAI does not see its long-term commercial future in media real estate. It sees it in workflow insertion.
Banner impressions monetise attention. Actions monetise outcomes. OpenAI is embedding itself into enterprise decision flows: incident management, IT service requests, employee onboarding and procurement logic. In these environments, the AI is not influencing what you think. It is influencing what actually happens next.
From a media strategist's perspective, three implications stand out.
OpenAI is pricing against friction, not reach. In enterprise contexts, saving minutes or preventing errors has a clear dollar value anchored to productivity, not exposure.
Commercial value shifts from persuasion to enablement. Traditional ads ask whether they can influence a choice. Actions ask whether they can help execute a decision better. The brands that win are not the loudest. They are the most easily embedded.
Actions collapse the funnel. When a system can both recommend and act, the commercial moment moves from click to completion.
What Performance Teams Need to Do Differently
This shift is already visible in client performance. For brands with strong clarity, incremental spend still converts efficiently. For others, performance plateaus early. The difference is rarely bidding strategy. It is whether the brand has already done the cognitive work before the click.
We see this on Meta and YouTube as well. CPA improves not through audience tinkering, but through narrative reframing. Moving from feature-led messaging to problem-led, proof-backed stories. The algorithm is effectively asking whether people find this useful.
If a client asked today whether they should prepare to advertise on ChatGPT, the answer would be no. Not yet. But they should absolutely prepare for what ChatGPT-like environments reward.
Three practical lenses matter.
Be clear on your decision role. Identify the moment where it would genuinely make sense for an AI to mention you. AI favours brands that help people think, not just transact.
Treat your brand as an input, not an output. AI learns from the entire information environment. Consistent positioning across site, ads, PR, reviews and third-party content matters more than any single campaign.
Invest where signal quality compounds. High-quality search content, credible proof points, authoritative thought leadership and measurement that connects brand to outcomes all feed AI judgement.
The Shift That Is Happening Regardless
If OpenAI gets this wrong and ads feel bolted on, it will damage ChatGPT. But it will not reverse the direction advertising is heading. It will simply change who gets there first.
User behaviour is moving towards conversational, assistive interfaces. Platforms are improving their ability to evaluate quality, not just engagement. Performance media is increasingly constrained by trust and saturation.
Whether or not ChatGPT succeeds as an ad platform, the bar for relevance, usefulness and credibility is rising everywhere. The easy arbitrage years are over. Performance is no longer just about finding demand. It is about being worthy of recommendation.
If this shift feels uncomfortable, that is usually a signal worth paying attention to. And if you want to pressure-test what this means for your own accounts, the conversation is already happening with the ADMATICians who work through these trade-offs every day.