ChatGPT Ads · OpenAI Ads
ChatGPT Ads: Advertising Inside ChatGPT
Ads in ChatGPT went live in Sweden on 24 August 2026. We built our own campaign the week billing opened — one campaign, ten ad groups, one hundred ads — and the numbers on this page come from that account. Here is how the channel actually works, and what it costs.
What ChatGPT Ads Actually Cost Us
Day one of our own Swedish B2B campaign, clicks objective, after reporting had caught up. OpenAI quotes $15–60 CPM as the typical range — we came in under the floor of it.
~$12
CPM on day one
Below the floor of OpenAI's quoted $15–60 range
~$0.65
effective cost per click
A fifth of the ~$3.30 mid-range planning estimate
~1.8%
click-through rate
Double the 0.91% third-party pilot figure
One day is a small sample and the numbers will move. We plan budgets on OpenAI’s range and treat our own figures as upside — the full working is in our campaign build guide.
We have run the channel, not just read about it
One campaign, ten ad groups, one hundred ads, eighty context hints — live in Sweden since launch week. The numbers on this page come from our own account, not from a vendor deck.
Built for how matching actually works
Ad groups follow buying situations, not your service catalogue, because context hints and bids sit at ad-group level. That single decision is what separates campaigns that deliver from campaigns that do not.
Measurable from the first click
Unique UTM parameters per ad, GA4 that reconciles, and the campaign readable from an AI assistant over MCP — so evaluating a hundred ads costs a question, not an afternoon.
What a ChatGPT Ad Looks Like
The format is deliberately restrained. The ad appears below the model response, never inside it, labelled “Sponsored” with a clickable favicon. You supply three things — and the character counts are a target range, not a ceiling to squeeze under.
| Element | Hard limit | OpenAI’s target range |
|---|---|---|
| Headline | Max 24 characters | 16–24 characters |
| Description | Max 48 characters | 32–48 characters |
| Image | 1:1, 640×640–1200×1200, PNG/JPG | Avoid the logo as the primary visual |
Note that 16 and 32 are floors. Writing shorter leaves the space unused and gives the relevance model less to work with — a mistake we made on forty of our own first hundred ads.
Why Most First Campaigns Underperform
Almost every early mistake comes from running ChatGPT Ads like a search account. Matching happens on conversational context, which changes what a good campaign looks like.
- Ad groups copied from your Google Ads service catalogue
- Context hints written as sentence-long personas
- A handful of broad, generic ads per campaign
- Customer lists you cannot use in Europe anyway
- Reading yesterday's numbers before they have settled
- One ad group per buying situation, bid by intent
- Four-to-five-word hints in the buyer's own words
- ~100 ads so the auction has something to match against
- Context hints plus geography — the targeting that works here
- Reporting read over MCP, after the lag, from one session
Before you budget
Four Things the Product Pages Do Not Lead With
None of these are dealbreakers. All of them change how you plan — and three of the four only became visible to us once we had a campaign in the account.
Only Free and Go users see ads
Plus, Pro and Enterprise are ad-free. Your addressable audience is not every ChatGPT user in Sweden, and the buyers with the most spending power are the most likely to be behind a paid plan. Reaching them is GEO's job, not the ad account's.
There is no exact-match keyword targeting
Context hints are short descriptive phrases — [product or category] + [use case or context] + [qualifier], five to ten per ad group. They guide matching, they do not guarantee delivery for a given query. Vague hints like "software" or "AI" are explicitly discouraged.
Custom audiences do not exist in the EEA
Campaigns targeting the EEA or Switzerland cannot use customer lists at all, regardless of size. Targeting here is context hints plus geography — which makes the hint work the entire targeting layer, and puts every advertiser on the same footing.
The traffic is mobile, and the reporting lags
On our first day roughly six in ten impressions came from the mobile apps, and CTR was strikingly even at 1.7–1.9% across every segment. Our first reading also undercounted clicks by about half before reporting caught up. Never decide on fresh numbers.
Ads reach Free and Go users. GEO reaches everyone else — and both run on the same underlying signals.
Recommended tooling
Run ChatGPT Ads From Your AI Assistant
A hundred ads across ten ad groups is a lot of surface, and Ads Manager is not built for the repeated daily reading a learning budget needs — especially when reporting lags by hours and you have to check the same day more than once.
Cogny’s OpenAI Ads MCP is now live: an MCP server — Model Context Protocol, the open standard that lets an AI assistant call external systems as tools — exposing campaigns, ad groups, chat-card ads and insights from OpenAI Ads. You ask in plain language and get the answer, instead of clicking through reporting views. Connect it with an API key from Settings → API keys in Ads Manager; each key is scoped to one ad account, is stored encrypted, and can be revoked whenever you want.
Disclosure: Cogny is an AI marketing platform built by the same founders as Growth Hackers. We use it on our own ChatGPT Ads campaign, alongside Google Ads, GA4 and Search Console in the same session — which is what makes comparing the new channel against the ones you already run a question rather than an export.
Read about the OpenAI Ads MCPHow We Build and Run Your Campaign
The same sequence we used on our own account. Most of it can be finished before access to the channel opens — which is exactly what let us launch the week billing came through.
Channel fit and budget maths
We work backwards from your order value and conversion rate to a defensible cost per click, before anything is built. For low-margin e-commerce the honest answer is sometimes no.
Buying situations and context hints
We map the problems your buyers are solving when they open ChatGPT, turn them into one ad group per situation, and write five to ten short hints per group in the words buyers actually use.
Around 100 ads and square creative
Ten distinct ads per ad group inside the 16–24 and 32–48 character ranges, plus the 1:1 creative most marketing teams do not already have on file.
Measurement before launch
Unique utm_term per ad, ad group in utm_content, and GA4 that reconciles — so a hundred ads pointing at the same landing pages stay separable afterwards.
Operate over MCP
We connect the account to Cogny's OpenAI Ads MCP so campaigns, ad groups, ads and insights can be read and adjusted from an AI assistant — alongside Google Ads, GA4 and Search Console in the same session.
Is ChatGPT Ads Right for You?
Tell us your order value and where your buyers are today. We will work backwards to a defensible cost per click and say plainly whether the channel is worth your budget yet — and what to do instead if it is not.
Book a free analysisFrequently Asked Questions About ChatGPT Ads
Something we did not cover?
We are happy to answer specific questions about the channel, your vertical, or what we are seeing in our own account right now.
Contact usEverything We Have Written About ChatGPT Ads
The long-form versions: the launch, the campaign build, the migration from Google Ads, who is actually on the other side of the ad, and why paid and organic are the same strategy here.