The advertising industry has spent the last six-plus months debating whether conversational AI would become a real media channel, and on May 5th, OpenAI answered that question by launching a self-serve Ads Manager with CPC bidding, conversion APIs, and pixel-based measurement. This is no longer a managed IO experiment with six-figure minimums and a $60 CPM that only holding companies could stomach. It's a performance channel with the same buying infrastructure marketers expect from any scaled platform. The conversation has moved from "should we pay attention to AI ads" to "how do we structure a test."

The biggest objection from the start was cost, and it was a fair one. $60 CPMs, no self-serve access, limited measurement. But at Direct Agents, we had the chance to run live campaigns across both B2B and B2C brands during the closed beta, and the story underneath the CPM headline was far more compelling. CTRs landed in the 1 to 2% range, comparable to what we see in non-brand search. Blended CPCs came in roughly 55% below our paid search benchmarks. And real conversions started coming through. Users aren't just tolerating these ads, they're engaging with them mid-conversation, mid-decision, in one of the highest-intent digital environments that exists today. That engagement signal is the data point the industry has been waiting for, and we now have it.

The infrastructure is catching up to the opportunity quickly. CPMs have already compressed from $60 at launch to as low as $25. CPC buying removes the impression-based risk that kept performance marketers hesitant. Conversion APIs and pixel tracking give us the downstream visibility to actually optimize toward outcomes, not just impressions. CPA bidding is on the roadmap. Third-party measurement partnerships are in motion. Major holding companies are already partnering on the channel. The economics and the tooling are moving in one direction, and it's favorable for advertisers willing to test now.

What a structured test actually looks like

Start with your highest-intent mid-funnel audience. The conversational context rewards ads that meet users at an active decision point, not a passive browse. B2B and considered-purchase B2C are the clearest fits right now. If you're running a non-brand search, that's your baseline and your first comparison point.

Get the measurement configured before you spend a dollar. Conversion APIs and pixel tracking are available today, and setting them up correctly at the start is the difference between a test that generates a real signal and one that just burns budget. CPA optimization is coming. The brands with clean conversion data already flowing will be first in line to use it.

Write creative for the context. Copy built for a banner or a search ad reads as out of place in a chat interface. The ad appears mid-conversation, mid-decision. Creative that fits that moment converts. Creative that interrupts it doesn't. This is the biggest execution gap we're seeing from teams jumping in without adjusting their asset approach.

Hold it to the same standard as any other performance channel. CTRs in the 1 to 2% range and CPCs running 55% below paid search are real numbers from real campaigns. If your first test isn't trending toward those benchmarks within the first few weeks, that's a signal worth diagnosing before you scale, not after.

We've been running these campaigns for clients across both B2B and B2C since the closed beta, and we've seen enough to have a clear point of view. If you're allocating budget to mid-funnel performance, this channel deserves a structured test today, not next quarter. The brands that establish benchmarks now will be making allocation decisions with a year of learning behind them by the time this auction gets crowded.