ChatGPT Ads launched in 31 European countries. None of them can open an account

In short: ChatGPT Ads went live in 31 European markets on August 24, 2026, the EEA plus Switzerland, on the Free and Go plans only. The campaign location picker offers 40 countries; the billing address field accepts 9, and the only European one is the UK. Until that changes, the useful work is the product feed, the OpenAI pixel and the size of your customer list.

You can point ChatGPT ads at customers all over Europe. You cannot open an account to pay for them: the billing address field accepts nine countries worldwide, and the only European one is the UK.

Ads went live across Europe on August 24, in 31 markets at once, the EEA plus Switzerland. OpenAI calls it its largest expansion to date. They serve on the Free and Go plans only, while Plus, Pro, Business, Enterprise and Edu stay ad-free. Your customers who use ChatGPT free or on Go are seeing ads already.

The other half is missing, which is the account itself. Mergado reaches Ads Manager through US clients and its own test account and described what it looks like inside: the campaign location picker offers 40 countries, but the billing address field accepts nine. The United States, the United Kingdom, Canada, Australia, Brazil, Japan, South Korea, Mexico and New Zealand. That is exactly the set the pilot ran in before ads reached Europe, and the only European country on it is the UK, where they have been running since June.

How the money gets in

Three routes were named at launch: OpenAI’s own Ads Solutions team, agency partners and technology partners. OpenAI has said self-service through Ads Manager will follow “later this summer”, without giving a date. Criteo opened direct booking alongside the launch, across DACH and 28 further European markets, which lets you buy without going through Publicis, WPP, Omnicom, Havas or dentsu.

The published entry prices are steep. From PPC Hero’s test of the channel:

  • Direct with OpenAI: around $250,000 a month
  • Criteo: $50,000 to $100,000
  • StackAdapt: around $50,000
  • CPM: about $15 on an uncontested prompt, near $60 on a contested one, $25 to $35 across brands
  • Starting CPC: $3 to $5

What you would be buying

Selection in the EEA is contextual: the topic of the live conversation, general location, device type. Past chats and memories are excluded at the outset, and personalised ads are not initially available in the EEA or Switzerland at all. Reporting doesn’t show you which conversations triggered an ad.

Eligible new ad groups default to a bid strategy called Maximize results. OpenAI’s own documentation says it “does not guarantee delivery against a specific CPA, CPC, ROAS, or other cost-efficiency target at this time”.

In twelve years of buying on Google and Microsoft, every automated strategy I used aimed at a number it was meant to hold. This one aims at volume.

Manual: Max bid is the only documented way to set a ceiling. You pick it per ad group, every time, and the documentation describes no account-level preference. View-through conversions arrived on a fixed one-day window, excluded from the Conversions total, from CPA, from bidding and from billing.

Now put two measurements next to those prices. SE Ranking studied more than 50,000 commercial prompts and found sponsored placements on 25.94% of them. Of those ads, 14.35% were unrelated to the prompt they sat under, and only 3.63% of advertisers appeared as a source in the answer above their own ad.

Then the conversion rate. On one website, a GA4 comparison of about 1,500 users had ChatGPT Ads traffic converting at 0.21% against 3.71% from Google Ads.

Adthena’s read-across of the European launch runs on roughly 850,000 US and UK queries captured before it happened, and the firm says so itself. European delivery uses a narrower signal set than the US delivery those numbers came from. Read every figure here as direction, not forecast.

What goes in the feed

OpenAI’s feed spec carries 75 live attributes. An ads feed requires thirteen of them, with no condition attached:

  • item_id, title, description, url, brand
  • image_url, price, availability
  • seller_name, target_countries
  • is_eligible_search, is_eligible_checkout, is_ads_eligible

Required is not the same as matchable. You don’t pick keywords here. You connect the catalog and OpenAI picks the product when the ad serves. These are the fields it has to pick on:

  • description (UTF-8, up to 5,000 characters, plain text): describe what the product is for.
  • q_and_a (list of q and a objects): the pre-sale questions your support answers every week.
  • product_category (taxonomy path, > as the separator): Apparel & Accessories > Shoes.
  • material (max 100 characters), color, size, weight, dimensions (LxWxH unit): the constraints people name before they name a product.
  • age_group (enum: newborn, infant, toddler, kids, adult) and gender (lower-case, male).
  • related_product_id with relationship_type (enum: part_of_set, required_part, often_bought_with, substitute, different_brand, accessory): the spec’s route to basket-building and cross-sell.
  • star_rating (0 to 5), review_count, popularity_score (0 to 5), return_rate (0 to 100%): the spec files these under credibility and ranking signals.

Four rules to check before the first upload:

  • title: 150 characters. description: 5,000. Both plain text.
  • Enums are closed and lower-case. An availability outside in_stock, out_of_stock, pre_order, backorder and unknown fails validation.
  • Variants need a group_id shared across the set. Without it OpenAI treats every SKU as its own listing.
  • is_eligible_checkout needs is_eligible_search true, plus seller_privacy_policy and seller_tos.

Coming from Google Shopping, OpenAI maps the columns itself:

  • id: item_id
  • link: url, seller_url
  • image_link: image_url
  • item_group_id: group_id, listing_has_variations
  • preorder: pre_order

It reads them from a delimited file only: UTF-8 .txt, .tsv or .csv. XML, RSS, Atom and JSON need converting first. OpenAI also sets eligibility itself on that path and ignores your columns: search on, checkout off, ads eligibility from the feed configuration.

Upload by SFTP. The Advertiser API has no endpoint that creates the connection. Send a full snapshot at least once a day, under one fixed filename you overwrite in place.

So does the feed get you shown organically? is_eligible_search controls whether a product can surface in ChatGPT search results, so the catalog does feed the organic side. Permission is not placement. Shoptet says the feed alone will not put products in organic answers. OpenAI says ad eligibility “does not guarantee that a product will serve”.

Three things worth doing before billing opens

Get the feed out of your platform. Shoptet ships a dedicated OpenAI feed you connect by copying a URL. Check what yours exports before you pay a developer for those thirteen fields.

Check the OpenAI pixel. If you let the ten-day opt-out window pass, automatic advanced matching became your default on August 17. It sends hashed customer data collected from your own website forms. In Europe that leans on the consent you captured, not the consent OpenAI captured.

Count your matched customers. A custom audience needs 25,000 matched users before it will run at all. Below that, audience targeting is not a decision you get to make.

None of that waits on the billing field. Do it while that field is still what stands in your way.

It will stop standing in your competitors’ way on exactly the same day.

For the AI Max auto-upgrade landing on your Google accounts, see AI Max replacing Dynamic Search Ads.

Running 50+ campaigns and still doing this by hand?

Get in touch →