The ChatGPT Ads Setup Guide: Launching Your First Campaign

The barrier to entry collapsed in 2026. When ChatGPT ads launched on February 9, participation meant a six-figure commitment and a managed relationship. The self-serve Ads Manager opened on May 6 with the spend minimums removed, so the question is no longer whether you can access the surface. It is whether you can set up a test that produces an answer rather than an inconclusive spend.

Most first campaigns fail for structural reasons, not creative ones. The usual causes are the wrong objective, ad groups that mix intents, targeting written like a keyword list, measurement that cannot see conversions, and a test killed before it gathered data. The strategic context sits in the LLM ads explainer; this is the operational companion.

The three-level structure

Ads Manager nests three objects, and the logic is familiar right up until the middle layer.

Campaign sets the objective, budget, dates, and geography.

Ad group holds the bid and the context hint. This layer works nothing like other platforms and determines most of your outcome.

Ad carries the title, copy, image, and landing page.

The structural consequence: because hints live at the ad group level, the clean way to scale is one ad group per audience and intent combination. Three buyer types across three funnel stages is nine ad groups, not one ad group with a hint trying to cover everything.

Step 1: Choose the objective

Three bidding models exist: CPM for reach, CPC for clicks, and oCPC, which bills per click while optimizing toward a tracked conversion event.

Use CPC or oCPC when you have a measurable conversion event and want a short feedback loop. Most demand generation belongs here, and you learn immediately what people click. Use CPM when you are introducing a category and building familiarity ahead of an evaluation that happens later; it needs more volume and more patience before you can read it.

If you are unsure, start with CPC. Move to oCPC once the conversion event is firing reliably, because a conversion-optimized objective with broken tracking optimizes toward nothing.

Step 2: Set budget and geography

The pilot's spend minimums are gone and daily budget is now the practical control. What matters more than the permitted minimum is a realistic test budget. Practitioner guidance converges on roughly $2,000 a month as a floor for gathering usable data, with a 60 to 90 day window before drawing conclusions. This is a judgment call rather than a published rule, but the failure it prevents is real: spending $500 over three weeks, seeing noise, and concluding the channel does not work. The channel may not work for you. A test that small cannot tell you either way.

Geographic control is more granular than at launch: country-level targeting plus US state, DMA, and ZIP options, and platform selection across the iOS app, the Android app, and web. Set these deliberately if your product has regional availability or licensing constraints.

Custom audiences from email or phone lists are also available, with a 25,000 matched-user minimum. That threshold puts them out of reach for most early-stage B2B lists, which is worth knowing before you plan around them.

Step 3: Build ad groups around intent

Each ad group carries one context hint: a plain-language description of the conversations where your ad belongs, capped at 280 characters.

A hint is not a keyword. A keyword asks what string someone typed. A hint asks what the person is trying to accomplish and whether your offer is genuinely useful right now. For a first campaign, three ad groups mapped to decision stage is a clean starting structure:

• Problem-aware. The person describes a problem without naming a solution category.
• Comparison-stage. The person is evaluating options within a known category.
• Purchase-ready. The person is close to a decision, asking about implementation, pricing, or specific requirements.

Give each ad group a focused hint, several matched ads, and the most relevant landing page you have. Writing the hints themselves is the highest-leverage skill in the channel and has its own structure, worked examples, and failure modes.

One limitation to plan around: there is no documented negative or exclusion targeting. You cannot exclude a conversation type the way you would add a negative keyword. Writing hints narrow enough that irrelevant conversations do not match is the only exclusion mechanism you have.

Step 4: Write ads that earn the click on their own

Your ad sits below the answer, not inside it. The user has already received something useful. Your creative has to earn attention against a satisfied question, without the momentum a search result gets from being the thing the person went looking for.

The format is tight. Titles run to about 50 characters and body copy to about 100, so there is no room for a warm-up sentence.

Three things separate ads that work:

Match the ad to the intent stage of its ad group. Running one creative across all three ad groups wastes the structure you just built.

Name the buyer explicitly. The conversational match is already narrowing the audience, so creative should confirm fit rather than broaden appeal.

Send to a dedicated, intent-aligned landing page. Not the homepage. Routing high-intent conversational traffic to a general page that makes the visitor start over is the most common conversion killer in first campaigns. Landing page content is also an auction input, so a page that matches the described situation helps you win placement, not just convert it.

Prepare image assets to the platform's creative specifications before you build, because validation rejects files that do not conform.

Step 5: Set up measurement before launch

Do this first, not after. Retrofitting measurement onto a running campaign throws away the early data, and attribution here is genuinely harder than in search.

A working stack has four parts:

Platform conversion tracking. The OAIQ JavaScript pixel, which sets a first-party cookie, plus the Conversions API for server-to-server events with deduplication. Attribution reporting runs on a 24 to 48 hour delay, so same-day dashboards will look wrong.

Analytics verified to receive UTM-tagged sessions. Confirm GA4 or your equivalent is attributing the traffic instead of lumping it into direct or referral.

CRM lead source tagging. For B2B the platform's conversion count is not the number that matters. Lead quality downstream is, and you can only see it if source is tagged into the CRM.

A post-conversion survey question. Optional, and genuinely useful on the thank-you page. Self-reported attribution catches influence the tracking misses, which on this surface is substantial.

Ads Manager reports impressions, clicks, spend, click-through rate, average CPC, and average CPM. Those are inputs. Judge the channel on cost per qualified outcome.

Expect modest click-through rates. SE Ranking measured 1.30% across roughly 97,000 impressions, and Choice OMG's June 2026 test came in at 0.65% with average CPCs around CA$4.50. A low click-through rate with strong downstream conversion is a good result that a click-through-focused review will kill.

Step 6: Launch, then leave it alone

The instinct after launch is to optimize daily. Resist it. These systems need conversion data before they can optimize delivery, and constant small adjustments prevent the learning that makes the campaign work.

In the first weeks, watch for obvious breakage rather than optimization opportunities: spend with zero conversions, ad groups getting no impressions at all, tracking not firing. After the learning window, the loop is straightforward. Cut hints that spend without converting, move budget to ad groups producing outcomes, and let results decide whether to expand, split, loosen, or retire each hint.

Do not scale before the numbers are stable. Scaling noise produces more expensive noise.

A pre-launch checklist

• Ten to twenty real high-intent buyer conversations mapped, so hints are grounded in buyer language
• Context hints written per ad group and reviewed against the ad policies
• Campaign structure organized by decision stage with one objective per campaign
• Ad copy written for each intent stage rather than reused across them
• Landing pages dedicated and intent-aligned, not the homepage
• Image assets prepared to specification
• OAIQ pixel and Conversions API live, analytics verified, CRM source tagging in place
• Test budget allocated with a 60 to 90 day window agreed internally

That last item is a stakeholder task as much as a marketing one. If the people reviewing this campaign expect week-two results, the test dies before it can answer anything.

Where first campaigns go wrong

Hints written as keyword lists. The most common mistake and the one that produces the worst matching.

One ad group for everything. A single hint cannot cover incompatible conversations, and creative cannot match any of them well.

Homepage landing pages. They convert poorly and they weaken your auction relevance at the same time.

Judging on click-through rate. A research-stage audience clicks less. Cutting on click-through kills campaigns generating quality pipeline.

Killing the test early. Underfunded short tests on an immature platform generate noise, and acting on noise is worse than not testing.

The B2B caveat worth repeating

ChatGPT ads serve Free and Go tier users only. Senior enterprise decision-makers frequently hold paid subscriptions and will not see your ads at all.

That does not rule out B2B, it shapes where it works. Self-serve products, developer tools, and small and mid-market targets fit the reachable audience. If your buyer is a CISO at a large enterprise, the budget is better spent on the organic citation work that puts you inside the answer, which starts with an audit of where you currently appear and how AI search assembles answers in the first place.

The setup mechanics above hold either way. The judgment call is whether the people you are trying to reach are on the tiers where these ads appear.