OpenAI's self-serve ad platform went live to all US businesses on , and most B2B marketing teams are still sitting on the sidelines. That hesitation is understandable: the platform is young, the targeting model is unfamiliar, and the measurement stack is thin. But the math on early-mover advantage is clear. According to Opascope's 15-day benchmark study, scaled campaigns returned a 1.49x blended ROAS at roughly $1.72 per click, well below the $2.96 average CPC on Google Search this year. The window won't stay open.
This guide walks through seven steps to launch your first ChatGPT Ads campaign, with the assumptions, constraints, and pilot design your CFO will actually sign off on.
Step 1: Confirm Eligibility and Audience Fit
Before you touch the Ads Manager, answer one question: does your ICP actually see ads on ChatGPT?
OpenAI's documentation is explicit: ads only serve to logged-in adults on the Free and Go ($8/month) tiers. Users on Plus, Pro, Team, Enterprise, and Edu plans see nothing. For B2B SaaS, that's a structural constraint. Omni Lab'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.
If your buyers are primarily enterprise decision-makers who expense their AI tools, this channel may not reach them at scale. If you're targeting SMB operators, freelancers, or early-career professionals who use the free tier for research, the fit is stronger. Model the audience overlap before you model the budget.
OpenAI also restricts categories including financial services, legal advice, gambling, and alcohol. Martech.org notes that some policies are self-contradictory (legal ads are running despite policy language suggesting otherwise), so check the current changelog if you're in a gray zone.
Step 2: Set a Test Budget That Generates Signal
The $200,000 minimum from the pilot phase is gone. Commerce Pundit's guide confirms you can now start with any budget, and daily minimums are as low as $25.
For a B2B test, I'd recommend a 14-day pilot at $100–$150/day, giving you $1,400–$2,100 in total spend. That's enough to generate statistically meaningful click volume at the platform's current CPCs ($2–$5 range, with Martech.org reporting that bids under $3 often fail to clear delivery thresholds). At $3 CPC, you're looking at 450–700 clicks over two weeks, which is enough to see directional conversion data if your landing page is instrumented.
Frame this as a learning investment, not a demand gen channel. The goal is to answer three questions: What CPCs do we actually pay? What context hints drive relevant traffic? Does the traffic convert at all?
Step 3: Write Context Hints, Not Keywords
This is where ChatGPT Ads diverge most sharply from Google. There are no keywords, no match types, no Quality Score. Instead, you write "context hints" at the ad group level: plain-language descriptions of the conversations where your product is relevant.
The Context Hints Guide frames it well: "Imagine you could whisper a one-paragraph briefing to ChatGPT before every relevant conversation, telling it when your product is worth surfacing." That briefing describes the prompt, the buyer, and the intent, not a literal phrase to match.
For a B2B example, instead of bidding on "best CRM for healthcare," you'd write something like: "Users evaluating CRM platforms for healthcare organizations, particularly those asking about HIPAA compliance, EHR integrations, or patient outreach workflows. Decision-makers comparing vendors or asking for feature comparisons."
OpenAI's help center is clear that hints are not exact-match and do not guarantee delivery in specific conversations. The model interprets intent from the full conversational context. This is a fundamentally different targeting paradigm, and it rewards specificity about buyer situations over keyword volume.
Step 4: Build the Ad Unit
The creative requirements are lean. Each ad includes:
- Headline: 3–50 characters
- Description: up to 100 characters
- Image asset
- Favicon/logo
- Landing page URL with UTM parameters
Just Global's B2B playbook notes that the ad unit sits clearly separated from the organic response and is labeled as sponsored. Users know what's paid. That separation is a feature: you're not trying to blend in, you're trying to be relevant at the moment of decision.
For B2B, lead with the outcome, not the product. "Cut CAC payback by 40%" beats "AI-powered marketing platform." The user is already in research mode; give them a reason to click that maps to their problem, not your feature list.
Step 5: Instrument Measurement Before You Launch
The reporting stack is basic: impressions, clicks, spend, CTR, average CPC, average CPM, and conversions. OpenAI's documentation confirms that conversion measurement is available through their Conversions API and pixel, but the data is thin compared to mature platforms.
Set up UTM parameters on every landing page URL. Use a dedicated landing page or at minimum a dedicated UTM campaign so you can isolate ChatGPT traffic in your analytics. If you're running a B2B pilot, make sure your CRM can attribute leads back to the source. The platform won't do this for you.
Jolly Good Web's update from July 2026 notes that conversion-optimized campaigns using CPA bidding are now available, along with "automatic advanced matching" for hashed user data. If you're running a longer test, consider switching to CPA bidding once you have baseline conversion data.
Step 6: Launch with Geographic and Tier Constraints
Just Global's playbook lists current markets as the United States, Canada, Australia, New Zealand, Japan, South Korea, and the United Kingdom. Geo exclusions are now available, so you can carve out regions where you don't have sales coverage.
For B2B, start with a single market (likely the US) and a single ad group with one set of context hints. Resist the urge to launch five ad groups on day one. You need enough volume per ad group to see signal, and spreading budget too thin will leave you with noise.
Step 7: Read the Data, Then Decide
After 14 days, you should have answers to the three questions from Step 2. If CPCs are in the $2–$4 range and conversion rates are comparable to your other paid channels, you have a case for scaling. If CPCs are above $5 and conversion rates are half your baseline, you've learned that the channel doesn't fit your ICP today.
Opascope's benchmark data shows daily ROAS swinging from 0.2x to 2.9x over a 15-day window. That volatility is normal for a new platform. Don't overreact to a single bad day; look at the rolling average.
The honest read on ChatGPT Ads for B2B is this: the channel has real scale (900 million weekly active users, 2.5 billion prompts per day), but the audience you can reach is constrained by tier eligibility. If your buyers are on the free tier, you have a window to test before CPCs rise. If your buyers are on paid plans, the math doesn't work yet.
Model the audience. Run the pilot. Let the data decide.
