How to Set Up ChatGPT Ads: A Practitioner's Walkthrough

ChatGPT opened up to advertisers in January 2026, and most of what’s been written about it since has come from people who read the announcement, not people who built a campaign. We’ve done the build. This is the walkthrough I wish existed when we started: what the platform supports today, how the pieces fit together, and where the sharp edges are.

One caveat up front. This platform is moving fast. Everything below is accurate as of mid-2026, from hands-on work in a live account. If you’re reading this a year later, verify against the current docs before you take anything as gospel.

What ChatGPT Ads Actually Is

Buyers have moved a real chunk of their research inside ChatGPT. Questions that used to be ten Google searches are now one conversation. Ads on the platform appear natively inside those conversations, which means your brand shows up at the moment someone is actively working through a decision, not scrolling past it in a feed.

That’s the pitch. The reality check: this is an early-stage ad platform, and the capability list is short. Knowing exactly what’s supported saves you from planning a campaign the platform can’t run.

What the Platform Supports Today

Here is the honest capability map as it stands.

Objectives. The Advertiser API currently supports Reach campaigns only, billed on CPM. Clicks and Conversion objectives exist on the UI roadmap, but if you’re building through the API today, you’re buying impressions. Plan and measure accordingly.

Ad format. The unit is a chat card: a native card that renders inside the conversation. There’s no responsive search ad equivalent, no video unit, no carousel. One format. That simplicity is actually useful in a pilot, because creative variables stay contained.

Reporting. The API is impression-only for reporting. You will not get click or conversion columns back from it. Everything downstream of the impression has to be measured on your side, which is why tracking setup matters more here than on any mature platform. I’ve covered that in a separate post on ChatGPT Ads measurement, because it deserves its own walkthrough.

Campaign Structure

The hierarchy will feel familiar if you’ve built on any major platform: campaigns contain ad groups, ad groups contain ads. Both campaigns and ad groups can be created via the Advertiser API, and the chat card ads attach at the ad group level.

Our structure advice for a first build: keep it small. One campaign per objective (which today means one Reach campaign), two or three ad groups split by theme or audience angle, and a small set of chat card variants per group. The temptation on a new platform is to port over the sprawling structure from your Google account. Resist it. You don’t have the reporting depth to learn from a complicated structure yet, so a complicated structure just spreads budget thin and teaches you nothing.

The Built-Paused Workflow

Here’s a detail I genuinely like: campaigns and ad groups created through the API come out paused. Nothing serves until someone deliberately enables it.

That sounds minor. It isn’t. It means you can build the entire campaign, review every ad group, every card, every destination URL, and every budget line before a single dollar moves. On platforms where API-created objects go live immediately, one malformed request can start spending in seconds. Here, the enable step is a human decision.

We’ve made that our standard operating rhythm, and I’d recommend it to anyone regardless of platform: build everything paused, run a full review pass with human eyes on every element, then enable deliberately. AI and automation can assemble a campaign fast. A senior person should still be the one who flips the switch. On ChatGPT Ads, the platform enforces the first half of that discipline for you.

API vs UI: Know Where the Line Is

This is the part that catches people. The capability split between the API and the UI is real, and if you don’t map it before you start, you’ll plan work the API can’t do.

The API can: create campaigns, create ad groups, build chat card ads, and pull impression-level reporting.

The UI only: conversion-event creation and attaching those events to campaigns are UI-only operations today. If your workflow assumes you’ll script the conversion setup the way you might in Google Ads, you’ll hit a wall. Budget human time for the UI steps.

Neither, yet: Clicks and Conversion objectives are roadmap items. You cannot bid toward a conversion today no matter which door you walk through.

Our practical workflow ended up being a hybrid. The structural build runs through the API, because it’s repeatable and reviewable. The conversion-event work happens in the Ads Manager UI as a checklist a person walks through. Write that checklist down. On a platform this new, the person doing the UI steps six weeks from now will not remember what the person who did them the first time figured out.

Budget Expectations

Two things to internalize before you set a number.

First, you’re buying CPM impressions in an environment where nobody has years of benchmark data. Treat your first month’s CPMs as discovery, not as a number to judge. What you’re really buying early is learning: does this audience exist on the platform, does the chat card format hold attention, does tagged traffic behave well once it lands on your site?

Second, size the pilot so you can afford patience. Early accounts on this platform tend to show strong engagement signals well before they show conversions, because the buying cycles in most categories run longer than the first weeks of a campaign. A budget that panics after week two will get pulled right before the data would have started meaning something. Set a number you can hold steady for a full quarter without anyone getting nervous, even if that means a smaller monthly figure.

And keep a human hand on the spend. New channel, thin reporting, no conversion bidding: this is exactly the situation where you want senior judgment reviewing pacing weekly, not an automated rule making decisions off two weeks of impression data.

A Setup Checklist

Condensing the above into the order we actually run:

1. Get access and find your pixel ID in the Ads Manager, and deploy the measurement pixel through Google Tag Manager before any campaign work. Tracking first, always.
2. Tag every destination URL with UTMs so paid ChatGPT traffic is separable in GA4. Without this, you cannot tell paid from organic ChatGPT traffic later, and you will want to.
3. Build the campaign and ad groups via the API (or UI if you prefer). Keep the structure small. Everything arrives paused.
4. Write and attach your chat cards. A handful of genuine variants beats a wall of near-duplicates.
5. Do the UI-only conversion-event work in the Ads Manager and keep notes on the steps.
6. Run a full human review of the paused build: URLs, budgets, cards, tracking.
7. Enable deliberately, then leave it alone long enough to learn something.

Should You Do This At All?

Setup is the easy question. Whether ChatGPT Ads deserves a slice of your budget right now is the harder one, and the answer is genuinely “it depends” in ways I’ve laid out in a separate post: Should You Advertise on ChatGPT Yet? If you’re on the fence, read that before you build anything.

If you’d rather have someone who has already made the early-platform mistakes run the build, that’s work we do. We manage ChatGPT Ads campaigns end to end: pixel deployment, the API build, the UI checklist, and the honest monthly verdict on whether the channel is earning its keep.