ChatGPT Ads in Sweden: Everything About OpenAI's Ads in ChatGPT

Updated 1 September 2026: our own campaign is now live and has spent its first full day of budget. We have added actual CPM, CPC and click-through rate from day one to the cost section — revised later the same day once reporting caught up — plus the delivery split by platform (iOS app, Android app, web) and an important EEA limitation on custom audiences that the platform only reveals once you build a campaign. Earlier update, 31 August: the sections on context hints and character limits were corrected against OpenAI's official creative guidance.

On 24 August 2026, OpenAI started serving ads in ChatGPT in Sweden. Sweden is part of an expansion into 31 European markets — by OpenAI's own description their largest geographic advertising expansion to date, six months after the first tests began in the US.

We have been in conversation with OpenAI about what this means in practice. Here is what we know about format, pricing and targeting — and what you can actually do today.

A note on names first. Most people search for this as ChatGPT Ads, because the ads appear in ChatGPT; OpenAI’s own product and interface are called OpenAI Ads and Ads Manager. They are the same thing, and both terms are used below. If you want the condensed version rather than the full walkthrough, our ChatGPT Ads guide collects the formats, the pricing and our own current numbers on one page.

What launched, and where

Ads went live simultaneously across 31 European countries: all 27 EU member states plus Iceland, Liechtenstein, Norway and Switzerland. That is a regulatory perimeter — the EEA plus Switzerland — rather than a list ranked by market size. Together with the nine markets that already had ads (the US, UK, Canada, Australia, Brazil, Japan, South Korea, Mexico and New Zealand) plus India, the platform now supports 41 countries.

The most important thing to understand up front: ads are shown only to users on ChatGPT's Free and Go plans. Plus, Pro and Enterprise remain ad-free. So your addressable audience on the platform is not "every ChatGPT user in Sweden" but the subset who do not pay for Plus or above.

For B2B marketers that is a meaningful caveat. The professional users with the most buying power are also the most likely to hold a paid subscription — and you cannot reach them with ads. You can still reach them organically, through GEO.

What the ad looks like

The format is deliberately restrained. The ad appears below the model response, never inside it. It is clearly labelled "Sponsored" with a clickable favicon, and OpenAI is explicit that advertising does not influence what ChatGPT answers — advertisers also do not receive user conversations.

You supply three things: a headline, a description and a square image.

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, not ceilings. OpenAI's own creative guidance is that the title should be 16–24 characters and the copy 32–48 where possible. Ads shorter than that leave the space unused and give the relevance model less to work with. In Swedish this is a relief rather than a constraint: the target range leaves room for a concrete offer instead of just a product name.

What it costs

OpenAI quotes a typical range of $15–60 CPM. That is expensive compared to display, roughly in line with LinkedIn, and hard to compare directly to search because the payment model differs.

The platform now supports both CPM and CPC bidding, as well as conversion optimisation. If you choose a click or conversion objective the CPM figure still matters as a benchmark, since it is the underlying auction currency. At a reported click-through rate of around 0.91%, the conversion looks like this:

CPM	Clicks per 1,000 impressions	Effective cost per click
$15	9.1	~$1.65
$30	9.1	~$3.30
$60	9.1	~$6.59

Around $3.30 a click at the middle of the range. For B2B with high order values that is entirely defensible. For low-margin e-commerce it probably is not — at least not until there is conversion data to optimise against.

Two important caveats about that click-through rate. The 0.91% figure comes from a third-party measurement early in the US pilot, and it is contested: in OpenAI's own material Figma reports that ChatGPT Ads deliver higher click-through rates than traditional search for their campaigns, alongside stronger-than-average onsite engagement after the click. Treat 0.91% as a pessimistic planning number, not a verdict.

And even if it holds for you, it mostly describes the interface. Google users scan and click. ChatGPT users are mid-task and reluctant to break that flow. The metric that actually decides this is conversion rate — and there HubSpot reports competitive CPAs and downstream revenue signals from the channel.

Day-one data from our own campaign

Since this article was published, our campaign has gone live and spent its first full day of budget. We launched in stages, with a first ad group serving while the rest of the structure rolls out. The numbers, from a Swedish B2B campaign with a clicks objective:

Metric	Day-1 actual	Compare against
CPM	~$12	Below the floor of OpenAI's $15–60 range
Effective CPC	~$0.65	A fifth of the mid-range estimate of ~$3.30
CTR	~1.8%	Double the 0.91% third-party figure

One day is a small sample and the numbers will move. But the direction is clear on two points: the click-through rate sits closer to Figma's account than to the pessimistic third-party measurement, and the effective cost per click came in far below what we planned for — the CPM even landed under the floor of OpenAI's quoted range. We are leaving the table above in place as a conservative reference.

One lesson about the reporting itself: the numbers lag. Our first reading, taken right as the daily budget was spent, showed substantially fewer impressions and clicks than were later reported for the same day — clicks were revised up by about half — and the spend columns in parts of the interface can temporarily show zero. Do not read the campaign too early, and do not make decisions on fresh numbers.

Reporting also breaks delivery down into three segments: platform (Android app, iOS app, web), device (desktop, mobile) and geography (country). On day one, roughly six out of ten impressions came from the mobile apps — the iOS app alone accounted for just under half of all delivery — and in the device view nearly two thirds of both impressions and clicks were mobile. Click-through rate was strikingly even across every segment, around 1.7–1.9%, with mobile slightly ahead of desktop and the iOS app highest. The practical conclusion: the traffic is predominantly mobile, so your landing pages had better be too.

Targeting does not work the way you are used to

This is the biggest adjustment, and the one that causes most early campaigns to underperform.

There is no exact-match keyword targeting. Instead you write context hints — short descriptive phrases signalling which conversations, needs and situations the ad group is relevant for. They are set at ad-group level alongside the bid.

This is where it is easy to draw the wrong conclusion, and we did so ourselves at first. Context hints are not sentence-length descriptions of a person. OpenAI's own guidance is explicit: write them as descriptive phrases of roughly four to five words, following the pattern [product or category] + [use case or context] + [qualifier]. Five to ten per ad group.

They sit closer to keywords than you might assume — OpenAI writes that if you are familiar with search keywords, context hints are "a good place to include keywords or phrases" describing relevant contexts. The difference is that they do not guarantee delivery for a specific query. They are broad contextual signals, not exact targeting rules.

Hints that work: "marketing agency in Stockholm for B2B", "project management tool for marketers", "GDPR-safe measurement with Consent Mode". Hints that are too vague, and explicitly discouraged: "software", "marketing", "AI".

On top of that you layer geography and your own audiences:

• Geographic targeting is set at campaign level across supported countries and, where available, states or regions, cities, DMAs and postal codes.
• Audience creation works from customer or prospect lists using email or phone identifiers, including supported hashed values. Minimum 25,000 matched, with 100,000+ recommended.
• Audience activation means including or excluding audiences at campaign level, with ad-group bid modifiers from 0.1× to 10×. Individual matches are never shown or selectable.

The 25,000 matched-contact threshold is worth pausing on. Most Swedish B2B companies will not come close. For them, first-party audiences are not a starting point but something to build towards — and the first realistic use is likely excluding existing customers rather than targeting.

Update: for European campaigns the question is moot for now. When we built our own campaign, the interface showed a notice that does not appear in the public material: "Custom audiences are not supported for campaigns that target locations within EEA or Switzerland, where personalized ads are not yet available." Custom audiences cannot be used at all for campaigns targeting the EEA or Switzerland — regardless of list size. In practice, targeting in Europe is context hints plus geography, nothing else. That makes the 25,000-contact threshold irrelevant for Swedish campaigns today, and it levels the field: no competitor can use their customer lists here either.

Why 100 ads per campaign

OpenAI recommends roughly 100 ads per campaign. It sounds like a volume heuristic but it follows directly from how matching works.

The relevance model reads your context hint and your ad copy together. An ad that literally mirrors the situation the user is in scores higher than a generic line that could sit under any conversation. That inverts the usual instinct to write a handful of strong, broadly applicable ads.

Here, the breadth of specific creative is the targeting. If you want to cover ten distinct buying situations you need ten ad groups with their own context hints — and enough ad variants in each for the auction to have something to choose between.

What you cannot do yet

Ads serving in Sweden does not mean anyone can buy them. Access is staged: initially through OpenAI's Ads Solutions team, agency partners and technology partners. Self-service through Ads Manager follows later.

In practice that means two things. Early advertisers in Sweden get in through relationships rather than by signing up for an account. And per-country billing setup is still rolling out — a constraint we were sitting in ourselves with our Swedish legal entity when this article was published. It has since cleared: our campaign is now live, and the first day's numbers are in the cost section above.

The work that determines whether the campaign performs — account structure, context hints, ad copy, square creative, measurement — can all be finished before access opens. That is exactly what let us go live the same week billing came through.

What we recommend doing this month

Our own campaign is now live — in stages, with a first ad group serving and the full structure of ten ad groups and one hundred ads rolling out. Here is how we would prioritise starting today.

1. Decide whether the channel is right for you at all

Plan for somewhere between $1.50 and $3.30 a click — our first day came in below that range at around $0.65 once reporting settled, but one day is not something to build a budget on. Order value and conversion rate need to carry the cost; work backwards from your actual customer value before building anything.

2. Map the needs you want to appear alongside

What problems are your buyers trying to solve when they open ChatGPT just before becoming customers? That list becomes your context hints — rewritten as short phrases of four to five words, five to ten per ad group. Keep each ad group focused: if you need to cover meaningfully different products or audiences, the answer is more ad groups, not more hints in the same one.

3. Produce square creative

The 1:1 format is probably something you do not already have. Avoid the logo as the primary visual.

4. Make sure measurement holds up

Give every ad unique UTM parameters from the start. With a hundred ads across ten groups pointing at the same landing pages, it is the only way to see afterwards what actually worked.

Plan the reading, too, not just the tagging. Platform reporting lags by hours, so you will be re-checking the same day repeatedly — worth connecting Cogny’s OpenAI Ads MCP, which exposes campaigns, ad groups, ads and insights to an AI assistant so a check costs a question rather than a session in Ads Manager. Disclosure: Cogny is built by the same founders as Growth Hackers.

5. Keep doing GEO regardless

This is the most important point. Paid placement and organic citation on AI platforms are driven by the same signals: clear positioning, structured content, documented expertise on the questions buyers actually ask. We have written about why they are the same strategy. Ads arriving in Sweden does not change that conclusion — it makes it more urgent, because the same work now pays off in two channels.

And unlike the ads, GEO does not require you to wait for billing to roll out. Read more about how we work with GEO optimization, or book a free analysis of how your brand shows up in AI answers today.