How to advertise on ChatGPT: a setup walkthrough

How do you set up ChatGPT ads?

Connect an account, install conversion tracking, write context-fit creative, set a CPC bid, and launch one small campaign. The whole sequence takes an afternoon. What separates a useful test from a wasted one is order: measurement before traffic, one context before five, two weeks of patience before the first edit. The step-by-step is below. This section is the why behind it.

If you are new to the channel itself, start with the complete guide and come back. This piece assumes you have decided to run a test and want the mechanics, in the order that keeps the data clean enough to learn from.

What do you need before you launch?

Three things, in this order: a funded ad account, working conversion tracking, and creative written for context rather than keywords. The order is the point. Conversion-optimized bidding cannot optimize toward a signal it never receives, so a pixel installed the week after launch means the platform spent your first week guessing.

Tracking has two parts. The conversion pixel reports from the browser, and the CAPI reports server-side from your backend. Server events deduplicate against pixel events by event id, so running both is the safe default rather than a redundancy. The conversion-tracking guide walks the install end to end, including which events to send and the one formatting detail that trips people up.

The third requirement, creative, is the one people underprepare. On a keyword platform you can launch with a thin ad and let the keyword do the targeting. Here the creative is the targeting input, so write it before you open the campaign builder, not in the last field before launch.

How do you connect and get approved?

Sign up at ads.openai.com, register as an advertiser, and add a billing method. The self-serve Ads Manager opened to US advertisers on May 5, 2026, and is rolling out gradually, so access is not always instant. Build the account before the day you want to launch rather than assuming you can do both at once.

Two practical notes. First, agencies should set up one ad account per client rather than running several brands from one account, because spend, conversions, and approvals all stay cleaner separated, and each client’s conversion signal needs to be its own for conversion bidding to work later. Second, ad review applies here as on every platform, so check OpenAI’s advertising policies before you build creative. Excluded categories are the usual ones, and the rules tighten as the platform grows, so a rejection is cheaper to avoid than to appeal.

How should you structure the first campaign?

One campaign, one context theme, one bid type. ChatGPT has no keyword targeting, so the unit you are testing is a context: the kind of conversation you want your ad to continue. Pick the context where your product is the obvious next step, and aim there before you broaden.

You do have two optional levers now. Custom audiences let you upload a hashed email or phone list to include or exclude specific people, and US geo targeting narrows delivery to a state, DMA, or ZIP. Both shipped in May 2026. Use at most one of them on the first test. Every filter you add shrinks the data you get back, and a thin auction is already short on signal, so the instinct to define the audience tightly works against you early.

Budgets gained options in May too. You can set a daily budget, which caps spend per day and is the simpler choice for a first test, or a lifetime budget across the campaign’s run. Start with a daily budget you can read quickly and lose without flinching.

What creative works in a conversation?

Creative that reads like the next helpful line in the conversation, not a banner that wandered in. The reader is mid-task. An ad that names the specific thing your product does and why it fits the moment earns the click; an ad that shouts a slogan gets scrolled past, and the auction reads that indifference as low relevance, quietly raising your cost.

The difference is easier to see than to describe:

• Don’t: “The #1 project management tool. Sign up free today.” It is generic, it is about you, and it would fit under any answer, which means it fits none.
• Do: “For a small team weighing Asana against Linear, here is how the tradeoffs actually shake out.” It belongs in the conversation it targets and earns the click by being useful first.

Write 2 or 3 variants and let them run against each other. Vary the angle rather than the wording: one variant leading on price, one on the specific feature, one on the outcome. Because matching is contextual, you are testing which framing fits the conversation, which is a different question from which headline is catchiest. Keep the dynamic CTA concrete, a clear next step rather than a generic prompt to click.

How much should you budget for the first test?

Enough to clear noise, not enough to hurt. A few hundred dollars over two weeks reads early click-through and conversion signal. The cost breakdown works the full budget-to-CPA arithmetic at a $25 CPM, including a worked $2,500 example and the CPA math by business type. There is no minimum spend, so the floor is set by how much data you need, not by a platform rule.

Start on CPC at $3 to $5. Paying per click caps your downside while click-through rates are unproven, and it hands the platform a clean signal about which impressions earn attention. Hold CPM and conversion bidding for later, once you have a proven click-through rate and enough logged conversions to bid against.

How do you read the first two weeks?

Treat the first two weeks as measurement, not performance. Early CPMs and conversion counts swing 20% or more week to week while the learning phase runs, so a great day-three number is as unreliable as a terrible one. The numbers you act on are the ones from week three onward, against your own account.

Watch three things in order. Click-through rate tells you whether the creative fits the context, and it is the first signal to stabilize. Conversion rate tells you whether the click was worth buying, and it needs your tracking to be solid. Cost per acquisition is the verdict, but it is the last number to trust, because it depends on the other two settling first. Resist editing on day four. Every material change can reset the learning phase, and a reset spends more of your budget recalibrating instead of converting.

How do you scale once a test works?

Raise the budget before you add complexity. If a single context is converting at a CPA your economics like, the lowest-risk next move is more budget on the same campaign, because you already know it works and the only question is how far it scales before the CPA drifts. Push it until the cost per acquisition starts to climb, then hold.

After that, broaden deliberately. Add a second context theme as its own campaign rather than crowding the first, move proven creative to CPM once your click-through rate clears the break-even, and turn on conversion-optimized bidding once the pixel has logged enough conversions for the platform to learn from. Each of those is a separate change, made one at a time, so you can tell which one moved the number.

What does daily management actually involve?

More than launching, which is the part people underestimate. Once a campaign is live, the work is recurring and unglamorous: review bids against yesterday’s clearing prices, check pacing so the daily budget neither starves nor dumps, watch for anomalies (a context that suddenly costs double, a variant that collapsed), and rotate creative before it fatigues. None of it is hard. All of it is constant.

That cadence is what the platform’s volatility demands and what a person doing it by hand tends to skip. This is where aematic, the ad management platform for OpenAI Ads, comes in: it reads your actuals every day, moves bids inside the rules you set, and surfaces what changed and why so the judgment stays yours.