Sponsored Intelligence and the Trillion Dollar Sentence

How advertising will unlock ubiquitous AI - and how to make that future one we want to live in

I once had a conversation with a visionary tech executive who said, and I am not making this up: “In the future there will be no advertising. Robots will print everything we need. But in the short term, I need ad revenue.”

There is no question that ChatGPT needs ads. There is also no question that ads will change the experience, because that’s what advertisers are paying for: the ability to influence consumer behavior.

OpenAI is in the unenviable position of trying to figure out a new advertising surface - AI chat - both from scratch and in the public eye. Regulators and consumer advocacy groups will be scrutinizing every move at the same time that the basic economics of AI are putting immense pressure on the company to go fast and competitors are shipping new models and new products. It’s an almost impossible challenge.

On the plus side, every advertiser in the world wants to influence consumers in ChatGPT. It’s the most powerful advertising surface in history: the trusted advisor that people ask about their most intimate and fundamental problems. As Hiten Shah put it, “ads now live inside an active reasoning context... adjacent to thought, not distraction.” This isn’t a banner next to content. It’s a sentence inside a decision.

Perhaps the best analogy here is your relationship with your doctor. You tell your doctor everything about yourself and the medical condition you have. They recommend a drug, you take it. Then you find out that the pharmaceutical industry has spent billions to convince doctors to prescribe that drug, including flying doctors to ski resorts and sending attractive young people in to hospitals. You begin to wonder whether you got the best drug or just the best marketed drug.

Fidji Simo says “ads will not influence the answers ChatGPT gives you.” She’s saying the AI will always give you the best answer, with ads shown alongside but never influencing the response. This is the medical equivalent of your doctor saying “I’ll always prescribe the best drug, but pharma reps can leave brochures in the exam room.”

I don’t think this model survives contact with either consumers or advertiser needs.

From a consumer perspective, either you trust your doctor to handle the potential conflicts of interest - in which case, why not just hand you the damn brochure - or you don’t, in which case you won’t come back¹.

On the other side, if ads don’t change behaviors, they’re worthless. Google changes your results. Amazon changes your recommendations. Apple takes billions to make Google your default search. Every successful ads business in history has worked because it changed what consumers saw, thought, or did.

The question isn’t whether ChatGPT will nudge you toward certain products, flights, or services. If they don’t, advertisers won’t pay for it. The question is whether that nudge will be disclosed, consumer-friendly, and grounded in real value.

So what should OpenAI do? They should make the most effective ad product in the history of the world by clearly prioritizing privacy first, consumers second, and advertisers third.

Privacy-first: Nothing leaves the building

If you’re going to build an ads business on the most intimate consumer relationship in history, you have one non-negotiable: consumer data cannot leave your walls. Not directly, not indirectly, not in “anonymized” form that we all know can be re-identified with a few signals.

I’m worried about OpenAI’s current language. They say “we keep your conversations private from advertisers and we never sell your data.” What they don’t say is whether they’ll share inferences, aggregates, or behavioral signals derived from those conversations. That’s a gap you could drive a surveillance truck through.

Here’s how I’d design it instead:

Advertisers get almost no influence over where their ads show, because even the most basic targeting could be reverse engineered to understand or even identify the consumer. Instead, they use a protocol like AdCP to characterize their brand, their products, their offers, their landing pages (or MCP endpoints), their privacy policy, and their safety and suitability standards.

When a user chats, pick the best offers to show them. If they want to engage with an advertiser, ask explicitly: “Can I share your contact info and what you’re looking for, or would you rather start fresh with them?” One click, clear consent, every time. This is like handing the consumer a prescription they take to their pharmacy. Their data and intent travels with them to their destination, fully in their control.

Reporting to advertisers stays opaque. They learn what worked, not who it worked on or what those people were asking about. Advertisers will complain. They’ll survive.

Don’t underestimate how quickly this becomes political, by the way. Members of Congress use ChatGPT. Probably for drafts of legislation, but also for relationship advice, medical questions—the same intimate stuff we all ask. The moment one of them realizes their data could be monetized, or weaponized, we’ll have a federal privacy bill. This might be the catalyst, and if the AI ecosystem gets privacy right, they’ll benefit from it - as will we as consumers.

Consumer-second: advocacy and offers

We have a friend who was part of a multi-level marketing program for vitamins and protein powder. She was always telling us how great the products are. We finally bought a very expensive box of powders. They were fine. The next time we saw her she was doing a different MLM scheme and has never mentioned the powders again.

That was not a great experience with our friend or the brand, which I literally don’t remember.

We recently hired a designer to help make our living room look less like “all the random furniture we’ve ever had plus a random area with Legos and books” and more like a place where adults could sip tea or whatever adults do on matching furniture. She explicitly said to us that she gets a designer discount on what she buys, but that we don’t pay for it - it comes from the store. I thought about that a lot. Does that give her an incentive to shop only at certain stores? (answer: no, since everybody provides this discount)? Does that mean she’s going to recommend more expensive stuff? (answer: clearly yes)

This was a great experience except for my bank account, and also do not buy a cream-colored rug when you have a small child and/or cats.

What we expect from our AI assistant is best efforts to help us. We are ok with our assistant getting paid, especially if it’s not coming directly from us.

As a consumer, I want this:

I see a few flight options to Boston tomorrow morning. I know you prefer United, so that's a good choice - they have a 6:45 AM for $245. However, Delta (which is a preferred partner of mine) has a 7:10 AM for $240. Would you like me to book either of these for you?

I don’t really care that the second option was Delta vs JetBlue or whatever. I mainly care that my preference for United was recognized (trust!) and that the other option is viable. Or how about:

I see a few flight options to Boston tomorrow morning. I know you prefer United, so that's a good choice - they have a 6:45 AM for $245. However, Delta (which is a preferred partner of mine) has a 7:10 AM for $240, and I can get you a free upgrade to their Economy Plus section. Would you like me to book either of these for you?

This to me is consumer first: we’re asking advertisers to work for our business, with ChatGPT as a conduit and advocate.

In sum: if you can look me in the eye (or equivalent metaphor for non-embodied entities) and tell me you stand behind your recommendation even though you have economics, I’m cool with it. That goes for the doctor analogy too: I didn’t care that Dr Adams used the heart valve ring he invented when he did my surgery, but I did appreciate that we talked about it.

Advertiser-third: The trillion dollar sentence

In this privacy-first, consumer-second model, advertisers are incentivized to create compelling offers and to actually deliver great products.

What is that flight recommendation worth to Delta? It’s a simple question of math. If they pay to be an OpenAI ads partner, they get shown in results 200,000 times more a month resulting in 50,000 bookings. Their incremental profit here is $2M a month. OpenAI can charge at least $400K for this and perhaps more, depending on how Delta sees LTV and the value of share gain. That’s a $2,000 CPM if you’re counting, but hey, AI ain’t cheap.

Perhaps travel is a unique category. But let’s see. If I have a sick kid and I’m asking ChatGPT for advice, and I get back a recommendation to give her some vitamins, am I cool with a nudge to a particular brand or retailer? Sure. Let’s say that in-chat transaction “Want me to order these for you? Walmart (a preferred partner) has a same-day delivery option and I can have them to you in 90 minutes” drives 100,000 orders a month. These are high margin transactions, sure. But how is this really different from “I need a new credit card” or “where can I get A2 milk” or “why will Anthropic not increase my monthly credit limit”?

What I’m saying is that I think that “answer independence” is naive and actually a poor consumer experience and a poor advertiser experience. Advertisers are paying you to drive sales. That means you have to change your answers. Google does. Amazon does. Apple does. I’m trying to think of an ads platform that doesn’t. If football fans put up with teams putting airline logos on their jerseys, and consumers are willing to pay money to be walking advertisements for clothing companies, I think we can live with an extra sentence or two in our chat responses - a sentence that will generate trillions of dollars of incremental revenue to advertisers and power a monumental ad business for OpenAI.

Sponsored Intelligence

AI capabilities will be everywhere and, inevitably, will be funded by advertising. The Sponsored Intelligence revolution is upon us, and it will be the defining concept in advertising for decades to come. The trillion dollar sentence - perhaps muttered seductively to you by your AI companion, perhaps embedded in a meta-review of product reviews by Rufus - will underpin an agentic economy that fuels unprecedented economic growth.

We know that LLMs are (currently) trained by reinforcement learning: when they give a good answer, they get positive reinforcement; when they give a bad answer they get negative reinforcement. Ads in the context of incentives play elegantly into this model. If you have a catalog of offers from advertisers, you can assess how this extra sentence impacts the quality and accuracy of the response AND what the expected revenue will be.

The consumer team at OpenAI has probably experimented with this and perhaps they’re not ready to do the trillion dollar sentence yet. But I think they will, and if not, the myriad other AI surfaces will lead them to it. It’s going to be incredibly exciting.

To unlock the trillion dollar sentence, Sponsored Intelligence platforms need the right incentives: privacy first, consumer trust second, and advertiser third. Platforms that nail these will make so very much money - and their users will keep coming back for more.

Implications for the advertising ecosystem

Bad news for the “open internet” folks: it ain’t gonna be open and you won’t be able to buy it programmatically. But you probably knew that already.

So how can we help OpenAI build ChatGPT into a massive ads business?

First, we need a way for advertisers to express their goals, standards, creative assets, offers, measurement data, and budgets in ways that Sponsored Intelligence platforms can leverage. This is a universal ads layer that needs to be independent of decisioning logic (since decisioning must happen fully inside the platform). One of the core design principles for the Ad Context Protocol was to provide the protocol layer for the next generation of AI-native ads platforms, and I’m excited to see many of these Sponsored Intelligence ad networks ramping up to launch with AdCP in the weeks to come.

Second, it looks like OpenAI is considering letting advertisers provide a direct chat experience that consumers can interact with. Love this concept and also have many questions. Will this become an industry standard where you can provide a standardized MCP interface to the chat platform for these interactions? Will we have third party MCP ad servers (see Firsthand)? We need a spec for Sponsored Intelligence ads, along with clear guidelines on how PII is handed off.

Third, we need to convene a bunch of people, quickly, to figure out how this is all going to work. We need to get all the stakeholders - including advertisers, agencies, AI platforms, ad tech folks, consumer advocates - into a forum to work out a framework for this $100B+ industry that builds on everything we’ve learned in the past 30 years of digital advertising. I would welcome anyone interested in this to join AgenticAdvertising.org and jump into the Sponsored Intelligence working group.

Machina in deo

If you have ever stood in a real-world bookstore looking up reviews of a book on Amazon or Goodreads (as I have) you will know exactly where this is going. We will soon be fully disintermediated from the real world by our devices. Our sponsored AI assistants will be gently and profitably nudging us, that billion dollar sentence ringing in our ears, as we hone in on the perfect book. It’s the Everywhere Store, the ultimate sponsored link, machina in deo.

Sponsored Intelligence is coming. The only question is how long it takes until the trillion dollar sentence becomes the greatest ad unit of all time.

1. Looking forward to ad-supported OneMedical subscriptions…