OpenAI pitches intent-over-identity ad philosophy at Advertising Week

OpenAI's Samantha Jacobson, VP of Monetization Partnerships, stood at Advertising Week and told the room that identity targeting is over. The future, she said, is understanding what a user is trying to do in a conversation, not who they are. Demographics and third-party attributes are "really creepy." The clean pitch comes with aggressive numbers: a $1 billion annualized ad run rate by late August, a $2.5 billion target for this year, and $100 billion penciled in for 2030.

The tension is immediate. Getting from $1 billion annualized to $2.5 billion in a single year means roughly tripling the current pace inside the back half of the year. Brand-safe "enrich the experience" formats don't move that needle. Performance advertising does, and performance advertising is exactly what Jacobson just called creepy on stage. One of those two commitments bends in 2026.

The signal problem compounds it. Intent targeting only works if the signal trains the bidding. ChatGPT sessions are sporadic and jump across topics. Search queries are not. The signal is thinner than the pitch suggests.

For operators, the practical problem is more immediate. A media buyer opening their DSP (demand-side platform, the software used to buy ads programmatically) finds nothing to traffic against: no cookie, no mobile ad ID, no third-party audience segment. No reach-and-frequency controls, no standard brand-safety taxonomy, no attribution model that maps to anything currently in use. The test budget gets into the room on the strength of the $1 billion number. Campaign operations hits the wall fast.

The companies with the most to lose are the mid-tier identity-graph vendors, the ones whose business is matching people across sites and devices. LiveRamp-adjacent infrastructure. If conversational intent shows measurable return on ad spend at scale, that category gets labeled legacy and repriced. Google and Meta have brand-direct relationships OpenAI cannot touch for years, so they are insulated for now.

The advertiser is being asked to hand budget to a black box with no auditable signal and no agreed measurement currency. That is a defensible test-budget decision. It is not a defensible 2027 allocation until someone shows return on ad spend that holds up next to a search control.

Our call: OpenAI misses its $2.5 billion 2026 ad revenue target. The ramp is too steep for brand formats alone, and the performance formats that could hit it contradict the trust pitch. In year one, with nervous CMOs and no measurement standard, the trust pitch wins and the number slips. We'll know by Q1 2027.

The longer question is more consequential. If intent targeting ever posts return that genuinely beats search, the identity-graph infrastructure gets repriced overnight. That is the two-year call, and it decides whether conversational advertising becomes a real budget category or stays a test line. This year just tests whether the growth math and the privacy story can both be true simultaneously. They cannot.

OpenAI stood up at Advertising Week and told the room that identity targeting is over. Samantha Jacobson, OpenAI's VP of Monetization Partnerships, said the future is understanding what you are trying to do in a conversation, not who you are. She called the old way, demographics and third-party attributes, "really creepy." Underneath the mission talk is a $1 billion annualized ad run rate reached by late August, a $2.5 billion target for this year, and a $100 billion number pinned on 2030.

What's actually being decided: whether advertisers move real experimentation budget to a surface that has no cookie, no device ID, no segment to buy, and whether "intent beats identity" becomes the way AI-native advertising gets sold. This is easy to undo for any single advertiser. A test budget is a test budget. It is hard to undo for the identity-graph vendors if the narrative sticks, because you can't un-ring the bell once "legacy" gets attached to your category. No hard deadline, but Q4 planning and 2027 upfront conversations set the clock.

The Skeptic A $1 billion run rate in six months is real money, and I'll give OpenAI that. But the $2.5 billion full-year target means roughly tripling the current pace in a single quarter. You don't get there with "enrich the experience" formats. You get there with performance ads that look a lot like the creepy stuff Jacobson just disowned on stage. "Conversations stay private" is a policy PDF, not an architecture. Meta made that promise too, more than once. And intent signals need repetition to train bidding. ChatGPT sessions are sporadic and jump around. Search queries don't. The signal is thinner than the pitch. For a non-specialist: they're promising clean targeting and aggressive growth at the same time, and those two usually fight.

The Market Analyst The $100 billion by 2030 figure is the kind of number that gets pasted into a bull-case model and never checked again. Ignore it. The claim that moves money is narrower: if intent targeting shows measurable return on ad spend at scale, the identity-graph infrastructure gets repriced as legacy. That hits the mid-tier vendors first, the ones built on matching people across sites and devices, LiveRamp-adjacent plumbing. Google and Meta don't care yet. They have brand-direct relationships OpenAI can't touch for years. For a non-specialist: the companies that sell "here is who this person is" data are the ones with the most to lose if "here is what this person wants right now" wins.

The Operator Tuesday morning, a media buyer opens their DSP seat and there is nothing to buy. No cookie, no mobile ad ID, no third-party segment to drop in. Intent inventory from a chatbot needs new creative briefs, new measurement proxies, new optimization levers. The $1 billion number gets agencies in the room. Campaign ops hits the wall fast: no reach and frequency controls, no standard brand-safety taxonomy, and attribution that maps to none of their existing models. Trafficking breaks first. For a non-specialist: the machinery agencies use to run campaigns assumes tools OpenAI doesn't offer, so the first thing that fails is just getting the ad live.

The Customer / End User Two customers here. The advertiser is being asked to trust a black box: hand me budget, I'll match it to intent, but you can't see the conversation and I won't sell you the data. That's a hard sell to a CMO who answers to a CFO on attribution. The ChatGPT user is the other customer, and the quiet question is whether ads in a tool you pay to use, or use to think, feels like help or feels like intrusion. Search ads survived because the ad often was the answer. In a conversation, the line between enriching and interrupting is thinner. For a non-specialist: nobody has proven people tolerate ads inside a chat the way they tolerate them above search results.

The CFO The real cost isn't the test budget. It's the opportunity cost of building a parallel measurement stack for one channel that may not scale, and the risk of pulling spend off channels where you can actually prove return. OpenAI is asking advertisers to pay for a signal they can't audit, on a surface with no agreed currency. That's fine for a line-item experiment. It's not fine as a planned 2027 allocation until somebody shows return on ad spend that holds up next to search. For a non-specialist: the money a company spends testing this is small, but the money it can't measure is the problem.

Tensions

The Market Analyst and the Skeptic split on whether the signal is even good. The Analyst says if intent shows return at scale, the identity vendors are in trouble. The Skeptic says ChatGPT sessions are too sporadic to train the bidding that return requires. Both can't be right, and the whole thesis hinges on that one fact.

The Strategist's "clean slate beats legacy" story collides with the Operator's Tuesday morning. A better signal is worth nothing if buyers can't traffic against it, measure it, or cap frequency on it. Narrative pulls experimentation budget. Plumbing decides whether that budget renews.

And there's the honesty gap the Skeptic keeps circling. "No identity, conversations private" is the brand-safe pitch. "$2.5 billion this year" needs performance formats that strain that pitch. One of those two commitments bends in 2026.

What it hinges on

Does conversational intent deliver measurable return on ad spend at scale, next to search? That's the one fact under everything. If it does, the identity-graph category gets the "legacy" label and the repricing follows. If it doesn't, OpenAI has a nice brand-safety story and a run rate built on novelty budgets that don't renew.

The council leans skeptical on the near-term revenue math and genuinely open on the long-term signal thesis. Before committing real 2027 allocation, an advertiser should demand a return-on-ad-spend read against a search control, and watch whether OpenAI quietly ships performance formats, because that's the moment the privacy pitch meets the growth target.

Prediction: OpenAI will miss its reported $2.5 billion ad revenue target for 2026, as measured against the ~$1 billion annualized run rate it disclosed in late August 2025, when results are known by the end of Q1 2027.

Confidence: Medium. The required ramp is steep and the formats that would hit it contradict the pitch.

Why: OpenAI hit a $1 billion annualized run rate by late August 2025, and the $2.5 billion full-year goal requires roughly tripling that pace inside the back half, a jump you don't make with "enrich the experience" brand formats alone. The only lever fast enough is performance advertising, the direct-response, click-and-convert kind, and that is exactly the targeting OpenAI stood on stage and called creepy. So OpenAI is caught between its growth number and its trust pitch, and in year one with nervous CMOs and no agreed measurement currency, the trust pitch wins and the number slips. The opposite outcome, hitting $2.5 billion cleanly, would mean advertisers moved serious budget onto an unmeasurable surface in a single quarter, which the attribution gap the Operator and CFO both flag makes unlikely.

Revisit by 2027-04-12: We're right if credible reporting or OpenAI disclosure puts 2026 ad revenue below $2.5 billion. We're wrong if it lands at or above $2.5 billion.

The longer game is more interesting than this number. If intent targeting ever posts return that beats search, the companies selling "who this person is" get relabeled overnight. That's the two-year call. This one just tests whether the growth math and the privacy story can both be true in year one. They can't.