Six months ago, OpenAI's ad announcement looked like a press release. Now it's a line item in media plans, and the early data is instructive for anyone trying to decide whether this channel belongs in a B2B mix.
The short version: ChatGPT ads work, but not the way Google or Meta ads work. The math is different, the audience is different, and the failure modes are different. If you're allocating test budget, you need to understand what you're actually buying before you write the check.
The Audience You're Actually Reaching
Here's the structural issue most coverage glosses over: ads only appear to Free and Go tier users. Plus, Pro, Team, and Enterprise subscribers see nothing. That's not a minor detail. Roughly 90% of ChatGPT's user base sits on the free tier, which sounds like scale until you ask who those users are.
They're students, freelancers, individual contributors, SMB operators, and consumers doing personal research. They're not enterprise procurement officers running vendor evaluations on a paid account. One agency's internal poll found that 74% of SaaS marketers are already on paid plans, meaning most of your ICP has opted out before you spend a dollar.
This doesn't make the channel worthless. It makes it a different channel than the one the headlines describe. If your product sells to individuals or SMBs, the audience math works. If you're targeting enterprise buying committees, you're fishing in the wrong pond.
What the Numbers Actually Show
OpenAI's ads pilot crossed $100 million in annualized revenue within six weeks of launch, with over 600 advertisers in the program. That's real demand. The platform is now live in the US, UK, Canada, Australia, New Zealand, Mexico, Brazil, Japan, and South Korea, with self-serve access and no minimum spend since May 5.
The cost structure has shifted dramatically since February. CPMs that launched at $60 now clear as low as $25 in some categories, and CPCs run $3 to $5 for most verticals. The $200,000 minimum spend that gated the pilot is gone.
One account running scaled campaigns over 15 days put in roughly $60K and returned about $89K in revenue, a 1.49x blended ROAS at $1.72 per click. That's a working signal, not a mature-channel result. The same data showed daily ROAS swinging from 0.2x to 2.9x, which tells you the algorithm is still learning and the variance is high.
Conversion rates across industries range from 0.2% to 5.8%, with B2B SaaS sitting around 1.2%. That's lower than organic ChatGPT traffic and lower than Google Ads for most categories. The pattern mirrors what we've seen across every paid channel: visitors who discover you organically convert at higher rates than those who arrive through paid promotion.
How the Targeting Works
Forget keyword auctions. ChatGPT ads use "context hints," plain-language descriptions of the conversations where your product might be relevant. The platform interprets intent from the full conversational context, not a single query.
If someone asks ChatGPT "Why are my Google Ads generating poor-quality leads?" and you've set context hints around Google Ads optimization, your ad might appear. The matching is probabilistic, not deterministic. You're not bidding on exact-match terms; you're describing the kinds of conversations where your product solves a problem.
This creates a different optimization loop. On Google, you refine keywords. On Meta, you refine audiences. On ChatGPT, you refine the language you use to describe your relevance. Early adopters report that specificity matters: "B2B SaaS companies evaluating CRM migration" outperforms "CRM software" because the platform can match more precisely to conversational context.
The creative constraints are tight. Headline maximum is 50 characters, description maximum is 100 characters, and you get one square image. Ads appear in a labeled "Sponsored" card below the AI's response, visually separated from the answer. OpenAI has committed to keeping ads independent of the response itself, which means you're not paying to influence what ChatGPT says, only to appear alongside it.
The B2B Problem
The structural audience issue compounds for B2B marketers. Your ICP is likely on a paid plan. The users seeing ads are disproportionately individuals, not buying committees. And the conversational context that triggers your ad may not map cleanly to purchase intent.
One analysis identified four reasons B2B marketers shouldn't rush to spend: the audience skews consumer, the targeting lacks firmographic controls, the reporting is immature, and the conversion infrastructure is weeks old rather than years old.
That said, the channel isn't irrelevant for B2B. It's relevant for specific use cases: products that sell to individuals within companies, SMB-focused offerings, and categories where the free-tier user is actually the decision-maker. If you're selling a $50/month tool to freelancers, the audience math works. If you're selling six-figure enterprise contracts, it doesn't.
What a Pilot Should Look Like
If you're testing, start with $5K to $10K over two to three weeks. That's enough to generate signal without burning budget on a channel that may not fit your funnel.
Set context hints that describe your highest-intent conversations, not your broadest category. Track post-click behavior through the Conversions API, which launched in May. CPA bidding went into early access in June, but the conversion-tracking infrastructure is roughly 90 days old. Google has been refining Smart Bidding since 2016. The feature is here; the track record is not.
Define your kill criteria before you start. If CTR drops below 0.3% or CPC exceeds $8 for your vertical, pause and reassess. If you're seeing clicks but no downstream conversion, the audience may not be your ICP regardless of how relevant the conversational context appears.
The Honest Read
ChatGPT ads are a real channel with real inventory and real early-mover economics. CPMs are low because competition is thin. That won't last. The brands testing now are building learning curves that will matter when the auction fills up.
But the channel is not a replacement for Google or Meta. It's a different surface with a different audience and a different intent signal. The users seeing your ads are in research mode, not purchase mode. They're asking questions, not clicking "buy now." That changes what you can expect from the channel and how you should measure it.
For B2B marketers, the honest assessment is this: test if your ICP includes free-tier users, skip if it doesn't. The channel will mature, the targeting will improve, and the audience composition may shift as OpenAI expands ad placements. But right now, the math only works for specific use cases.
Model it before you spend. That's the only way to know if the channel belongs in your mix.
