ChatGPT Ads in Finland: guide, prices and first results

ChatGPT advertising opened in Finland and the rest of Europe in late August 2026. Ads started showing to European users on August 24, and the self-serve ad manager opened to advertisers on August 31 at ads.openai.com. In the United States, advertising launched in February, and according to OpenAI it passed a one billion dollar annualized run rate in under 200 days.

Over the first month we’ve built campaigns for Mediashake’s own account and for our clients. This guide pulls together everything we’ve learned: how the platform works, how to build an account in practice, what a click has cost in Finland and what budget to start with. We updated it on September 24, 2026.

We also covered the topic in a webinar on September 22. Watch the recording (in Finnish).

In this guide:

1. Why ChatGPT advertising matters
2. Results from Finland: the first month in numbers
3. Where and to whom the ads show
4. How an ad gets picked
5. How ChatGPT advertising differs from Google and Meta
6. Account structure in practice
7. How much ChatGPT advertising costs
8. Conversion tracking and reporting
9. Where to start and with what budget
10. The most common mistakes
11. Is now the time to start?

The short version for busy readers

ChatGPT shows ads only to users on the free Free tier and the low-cost Go tier. The structure is the same as in Google Ads and Meta: campaigns, ad groups and ads. Targeting is based on the context of the conversation and the advertiser’s context hints, not keywords. Pricing is based on impressions, clicks or conversions, and the minimum daily budget is 15 € per campaign. The ad format is stripped down: a headline of up to 50 characters, a 100-character description and one square image.

On our client accounts, cost per click has been 0.15–1.42 €, well below OpenAI’s recommended starting bid. Conversion tracking works with a pixel and a Conversions API, much like on Meta. Financial, health and legal services can’t advertise in Europe for now.

Why ChatGPT advertising matters

ChatGPT is an everyday tool in Finland, not an experiment. By our estimate it has more than 1.5 million monthly users in Finland, and according to Statistics Finland, 41% of Finns aged 16–89 used generative AI during a three-month period in 2025, up from 23% a year earlier.

Most users don’t pay. OpenAI reported 50 million paying subscribers in February 2026, against more than a billion weekly users. That means over 95% use ChatGPT for free, and those are exactly the people who see ads.

Buying has moved into conversations. In a Gartner survey, 45% of B2B buyers said they used generative AI in their most recent purchase, and according to NielsenIQ, 42% of US consumers used an AI tool to shop in the previous month.

Search picks up a keyword, ChatGPT gets the whole situation

In a search engine people type a couple of words: “air source heat pump house”. In ChatGPT they describe the whole situation: “We have a 120 square meter 1950s house with oil heating. We want to cut our heating bill before winter, the unit can’t be noisy on the bedroom side, and our budget is about 3,000 € installed.”

The same goes for business purchases. In search, a buyer moves through queries, vendor pages and comparison articles, and each query is a couple of words. In ChatGPT, every stage of the purchase is a question that adds context. That’s why ChatGPT knows far more about the context of a conversation than a search engine knows about a single query.

The money is already in play. According to Kantar Media and IAB Finland, 249 million euros went into search advertising in Finland in 2025. ChatGPT advertising is the first channel that genuinely competes for the same pre-purchase questions.

Results from Finland: the first month in numbers

On our client accounts, actual cost per click has ranged from 0.15 to 1.42 € by account, cost per thousand impressions from 7 to 16 €, and click-through rate from 1.0 to 4.8%. The accounts include both B2B and consumer advertisers. Those prices are well below OpenAI’s recommended starting bid of 3–5 dollars per click.

The numbers cover a short period, and the channel is new. Prices may still move significantly as more advertisers join and the system learns.

Daily delivery swings more than you’d expect from other ad channels. One day brings plenty of impressions, the next very few, with the same budget and settings. Early on, impressions also piled up quickly on very narrowly defined topics, which suggests the system samples broadly at first before it narrows down.

Traffic quality varies. On some accounts, the ratio between Google Analytics sessions and ad clicks has been unusually weak. That points to accidental or curiosity clicks in the channel’s early days, so check traffic quality in your own analytics, not just in the ad manager.

Product-based advertising is cheap, but delivery stays small. Ads built from a product feed have had a low cost per click, a good cost per conversion and roughly double the click-through rate of regular ads. Volume, though, has stayed small.

Conversions have come in, but volumes are small. That’s the nature of a new channel weighted toward the consideration stage before a purchase decision. It isn’t necessarily a sign the channel doesn’t work.

Where and to whom the ads show

The ad appears below ChatGPT’s answer, labeled as an ad, with the advertiser’s name, logo, headline, description and image. Ads are shown only to adult users on the free Free tier and the low-cost Go tier, both logged in and logged out.

Plus, Pro, Business, Enterprise and Edu accounts see no ads at all, and neither do Temporary Chats or accounts identified as belonging to someone under 18. Users on the pricier subscription tiers are out of advertising’s reach, which is worth factoring into your audience math.

According to OpenAI, ads don’t influence ChatGPT’s answers, and advertisers don’t see users’ conversations. Ads aren’t placed next to sensitive topics such as health, mental health or politics.

In Europe, ad personalization is off. The ad is chosen based on the current conversation, general location and language. In the United States, past conversations, memory, ad history and customer lists can also play a part.

Advertising is a different thing from showing up organically in ChatGPT’s answers. We cover the organic side, AI visibility, in more depth on our AI search expertise page.

How an ad gets picked

Four signals shape which ad is shown. The main signal is the context of the conversation, which comes from the user. The advertiser controls the other three: the landing page, headline and description, the context hints, and the bid.

Ad slots are allocated through a relevance-weighted second-price auction. The winner pays based on the second-highest bid, and a more relevant ad can beat a pricier one. The same principle is familiar from Google search ads and Meta’s auction, so you can set your maximum bid at your true maximum without fear of overpaying.

Not every industry can advertise in Europe

Under OpenAI’s ad policies, advertising for financial, health and legal services is currently prohibited outside the United States. That covers banks, insurers, loans, investment services, clinics, labs, dental care and law firms, for example. Alcohol, gambling, dating and political advertising are prohibited everywhere.

The line is drawn at the claim, not the industry label. Financial education, a budgeting app without a loan offer or general wellness without health claims are allowed. The policy has been updated almost monthly, so check again before planning a campaign.

How ChatGPT advertising differs from Google and Meta

Account structure in practice: Mediashake’s own account

If you’ve built a campaign in Google Ads or Meta, you can build one in ChatGPT. The hierarchy is the same: the account level holds settings and tools, the campaign sets the objective, budget and location, the ad group handles targeting, and the ad carries the message. Here’s how it looks on Mediashake’s own account.

Mediashake’s first campaign: seven ad groups by service and buying situation. Finnish interface.

Account level: settings and tools before any campaign

When you create the account, you pick the country, currency and time zone, and they lock. A wrong choice means a new account. Your company name and logo appear in the ad as is. You also add payment details, users and, if needed, API keys.

Business verification is nothing to worry about: in our case, filling in the company details was enough and the review cleared in under an hour.

The account level also holds the things campaigns refer to: conversion events, product feeds and custom audiences. Set up conversion events before your very first campaign, because without them the conversion objective has nothing to optimize toward.

Campaign level: objective, budget and location

There are three campaign objectives: impressions, clicks and conversions. The objective and budget type can’t be changed after creation, so choose them carefully. At the campaign level you also pick countries and platforms: the Android and iOS apps, Android and iOS web, and desktop web. Conversion events are attached to the campaign for reporting and optimization.

We started our own campaign on the clicks objective with a 20 € daily budget, and attached two conversion events for reporting from day one. With some clients we’ve used the conversion objective from the start, with no delivery problems.

Campaigns can be bulk-imported and bulk-edited as a spreadsheet or through the API, which speeds up building larger structures considerably. We pushed our own campaign to the account through the API.

Ad group level: context hints and bid

The most interesting difference from other platforms is contextual targeting. For each ad group you write context hints: sentences that say what you offer, to whom and in what situation. The hints go into one text field separated by commas, and in our experience 2–5 hints per ad group works best.

The difference from a keyword is big. “AI visibility” is a weak hint. “AI visibility audit for a company whose competitors show up in ChatGPT’s answers but it doesn’t” is a good one, because it describes the situation where a buyer needs the service. The user doesn’t have to use the same words for the ad to be relevant.

Write hints in your customer’s language, not your internal vocabulary. We noticed our own hints used words like “GEO” and “kick-off project” that customers don’t use.

Context hints aren’t targeting rules. They don’t guarantee showing up in specific conversations, and they don’t limit location, time or audience. There are no negative hints or match types. Split ad groups by theme, such as service or buying situation, because the ad group is the level where you compare results.

The default bid strategy is the automatic Maximize results, with a manual maximum cost per click as the alternative. Landing page tracking parameters can be set at the campaign, ad group or ad level.

Ad level: a stripped-down format, many versions

Creating an ad in the ad manager. On the right, a preview of how the ad looks in ChatGPT.

An ad has a headline, a description and one square image of up to 1200 × 1200 pixels. The headline maxes out at 50 characters and the description at 100, so the fields are in Google Ads territory but clearly shorter than Meta’s ad copy. OpenAI recommends going even tighter: a 16–24 character headline and a 32–48 character description. Each image needs its own ad.

It’s hard not to think of the early days of Facebook advertising, when ad images were the size of a postage stamp and the copy was tiny. The image shows up in ChatGPT as a square of roughly 100 pixels, so room for creative testing is limited. Pick images where the person is large and the face is visible. Text images work only with 2–4 large words. Specific, benefit-led copy works better than slogans, so precise copy beats clever copy.

One ad group, six ads, two image types: photos from Mediashake’s own image library and text cards. Finnish-language ads.

Make several clearly different ads. OpenAI’s guidance says a diverse set of ads helps the system find more situations where your ad is relevant. People describe the same need in very different ways, and every distinct angle reaches different conversations. Our own first campaign had 43 ads across seven ad groups, and the goal is to grow that past a hundred.

One setting worth noticing right away: text customization is on by default, meaning AI creates adapted and translated versions of your headlines and descriptions that can be shown without separate approval. A familiar pattern from Google Ads automation, so go through the automation settings before launching a campaign.

The landing page and OpenAI’s crawlers

The landing page is chosen per ad, and it matters for targeting too. According to OpenAI, it checks the landing page against its ad policies and may use the page’s content to judge when an ad is relevant. So for each need, send people to a page that talks about the same thing as the ad and the context hints.

Make sure OpenAI’s crawlers can reach the page. OAI-AdsBot must be allowed, and OpenAI recommends allowing OAI-SearchBot too. Check robots.txt, your firewall, CDN and bot protection: a 403 error can block ad review and feed processing. For product feeds, the image host also has to let OAI-SearchBot through.

How much ChatGPT advertising costs

The minimum daily budget in Finland is 15 € per campaign. You can set a daily budget or a total campaign budget. The daily budget is an average over a seven-day period: spend on a single day can go up to double the daily budget, but spend over the seven days won’t exceed seven times the daily budget. The same practice as on Google.

There are three pricing models. Impression campaigns are bought per thousand impressions. Click campaigns are bought per click, and OpenAI recommends starting with a maximum bid of 3–5 dollars, although actual prices on our client accounts have been clearly lower. With the conversion objective you set a bid cap per conversion, but billing still happens per click or impression, and the bid cap is not a guaranteed acquisition cost.

Conversion tracking and what the report shows

Conversion tracking resembles Meta’s model: the OpenAI Pixel on your site, the Conversions API on the server side, and the most reliable measurement comes from using both. Conversions appear in reporting with a 24–48 hour delay.

Conversions are credited on a last-touch basis. By default, the window is 30 days from a click and one day from an impression, and you can switch the windows in the column settings to 7, 14 or 30 days and 0 or 1 day.

The report shows impressions, clicks, click-through rate, spend, cost per click and conversions at the campaign, ad group and ad level, broken down by device, country and platform. The report doesn’t show what kinds of conversations the ad appeared in, demographics, a split between branded and non-branded queries, or competitor share.

Tracking parameters persist through clicks, so you’ll see the traffic in Google Analytics right away. Custom audiences based on your customer lists are created at the account level, but they aren’t available in campaigns targeting the EEA. The ad manager blocks selecting them in a campaign targeting Finland.

Add a low-threshold conversion if direct leads or sales are few. In ecommerce that could be an add-to-cart, in B2B a qualitative event such as reading content or browsing several pages. Say out loud from the start how it relates to actual leads and sales, so page browsing isn’t mistaken for a business result.

For ecommerce: feeds work, but a Google feed needs changes

Product feeds can be imported as CSV files, from a URL or over SFTP, and they get their own campaign type. According to OpenAI, feed-based ads have been among the platform’s best performers, and on our client accounts their cost per click has been low.

A Google or Meta feed won’t work as is. OpenAI has a Google-compatible feed profile, but the format is confirmed when the feed is registered. An XML Merchant Center feed has to be converted to CSV or TSV, and the merchant name and target countries come from the registration, not the feed. Products expire after two weeks, so automatic feed updates are a must.

There’s no ROAS optimization yet, so ecommerce advertisers optimize for conversion volume rather than revenue. Reporting does show order-based revenue and ROAS once purchase events are connected to tracking.

Where to start and with what budget

We recommend two ways to start. In both, the budget is additional, not taken from campaigns that work.

Pilot: 15–30 € a day for one to three months. One tightly scoped focus area and 2–4 ad groups. That’s roughly 500–2,700 € in media, which at the click prices we’ve seen on client accounts means about 1,000–6,000 clicks. The goal is to try the channel, watch traffic quality and scale later.

Learning test: 50–100 € a day for three months. Several services and ad groups, roughly 4,500–9,000 € in media. The goal is to properly learn which topics and ads work, and at what price.

What matters most is that your account is fully in order from day one. Before the first euro, get three things in order: a situation map of the questions and buying situations where you need to show up, dozens of different ads for different needs with their own landing pages, and measurement with tracking parameters and conversions from day one.

How to track a pilot

The ad manager’s numbers alone won’t tell you whether to continue. Track four things in a pilot:

1. Compare against Google non-branded traffic. Same landing page, same period. Non-branded search is the closest benchmark, because in both the visitor doesn’t know you yet.
2. Traffic quality in Google Analytics. Share of engaged sessions, pages per visit and return visits tell you whether clicks turn into real visitors.
3. Cost per engaged visitor. Media spend divided by engaged sessions is the first comparable number while conversions are still few.
4. Cost per conversion, including low-threshold conversions. Agree on how they relate to actual leads and sales before the pilot starts.

Decide after the pilot based on these: continue, change or stop.

The most common mistakes in ChatGPT advertising

These are mistakes we made ourselves or saw in the first month:

1. Context hints written as keywords or in internal jargon. “GEO services” doesn’t describe a situation where a customer needs help. Write a sentence with the offer, the audience and the situation.
2. Too few ads. Three ads and one landing page aren’t enough, because people describe the same need in many ways. Make dozens of clearly different versions.
3. The same image in every ad. The image shows up as a roughly 100-pixel square, and using one image everywhere takes away the chance to learn. Mix photos and text cards.
4. Ad copy that describes the channel. “Ad below the answer” or “now in Finland” tells the user nothing. They’ve just read an answer, so the ad has to be the next step.
5. Launching without tracking. Without conversion events and tracking parameters, you can’t evaluate the first weeks’ data after the fact.
6. Judging results by the ad manager’s numbers alone. Traffic quality only shows up in your own analytics, and the gap can be large.

Is now the time to start?

Yes, with a scoped pilot, if:

• search advertising has worked well for you as a channel
• you have low-threshold conversions that let you measure results in the short term
• you can set aside a separate test budget that doesn’t eat into channels that work.

Wait if you’re in financial, health or legal services, your audience is a small, tightly defined B2B group where use of ChatGPT’s free version is likely low, every media euro carries a strict return target, or your other marketing channels haven’t worked and a new channel would be a fix.

The channel is just getting started, so keep expectations modest. New ad channels still always carry a potential early-adopter advantage: competition is thin, prices reflect that, and new mechanics like context hints are only learned by doing.

We help both with launching ChatGPT advertising and, more broadly, with how your company shows up in AI search. Read more about AI search as an expertise area or ChatGPT as a channel, or watch the webinar recording (in Finnish), where we walked through Mediashake’s own ad account level by level.

Sources

• OpenAI: A milestone in expanding access to AI, August 31, 2026
• OpenAI Help Center: Ads in ChatGPT, Ads in ChatGPT: The Basics, Create ads, Daily budgets, Web crawlers
• OpenAI: Ad policies and product feed specification
• Statistics Finland: Use of information and communications technology by individuals 2025
• Gartner: B2B buyer survey, May 20, 2026
• NielsenIQ: 42% of consumers now use AI tools to shop, May 5, 2026
• Kantar Media and IAB Finland: Media advertising volume 2025
• Mediashake client account data, August–September 2026