Advertising on ChatGPT: the guide to advertise in Spain

What is advertising on ChatGPT

Advertising on ChatGPT is OpenAI's ad system that shows sponsored blocks below chat responses. It is managed from OpenAI Ads Manager at ads.openai.com, and you pay per thousand impressions (CPM) or per click (CPC). Unlike Google Ads, the ad is not chosen by a keyword someone types, but by the context of the conversation already happening.

Dates matter here more than on any other channel, because almost everything published on the topic in the first half of 2026 is already outdated. OpenAI started the test in the United States on February 9, 2026, and on August 25 expanded it to Spain and about thirty European markets. You will still find well-ranked articles saying it is "not yet available" in Spain. It is.

That said, it is still a beta. Pieces that anyone managing accounts takes for granted are missing, and it helps to know which ones before you promise anything to a client or leadership.

What an ad looks like in ChatGPT

The ad appears below the end of the response, visually separated and labeled as sponsored. It is not inserted inside the text the model writes. Each ad unit has six elements: advertiser name, favicon, title, description text, image, and landing page.

One detail changes how you write copy: the user can open the ad menu and tap "Ask ChatGPT." That shares your ad with the chat and lets them ask questions about it. By default the model does not see the ads shown to you, so until the user shares it, it does not exist for the conversation. If your headline promises something your landing page does not deliver, the user has an instant checker at hand.

There are also places where your ad will not show, even if you pay: in the ChatGPT Atlas browser during the test, in temporary chats, on accounts identified as under 18, and near sensitive or regulated conversations (personal health, mental health, politics).

Who sees ads: the filter almost nobody checks

Ads can show to users on Free and Go plans. Plus, Pro, Business, Enterprise, and Edu accounts do not carry advertising.

It is worth pausing here, because this is the most important strategic decision on the channel and is often overlooked. If you sell to profiles that pay for Plus or Pro to work with AI daily (developers, agencies, consultancies, much B2B SaaS), your buyer is not in the inventory. You can have the best ad in the world and not reach them.

The reverse also holds: if you sell to consumers or SMBs using the free version, your audience is there, and they are also saying out loud what on Google they would type in three words.

How ChatGPT decides which ad to show

The system selects based on relevance and expected results, and considers conversation context and intent, landing page, ad title and text, context hints from the advertiser, and configured targeting.

The auction is second-price weighted by relevance. In plain terms: bidding higher does not buy the placement if your ad does not fit the conversation. A relevant ad with a moderate bid can beat a generic one with a high bid, much like Google's quality score did in its day.

In Spain, personalization is off

This is the nuance that separates a copied article from a useful one. Personalized ads are not initially available in the European Economic Area or Switzerland. In the United States, with personalization on, the system can use past chats, memory, and ad interaction history to decide what to show you.

In Spain, it does not. Here selection relies on the current conversation thread, plus basic context such as language and general location. That has two practical consequences: you will not build retargeting or audience strategies based on historical behavior, and all weight falls on matching conversation context. What on other channels you would solve with audience data, here you solve by writing context hints well.

Context hints: the new keyword research

Context hints are set at ad group level and describe conversations, topics, or terms where your product may be relevant. They are not exact match and do not guarantee you appear when someone mentions a word.

The usual mistake is treating them like a keyword list. If you are a dental clinic and put "dental implants," you are describing your service, not your customer's conversation. Nobody writes that to ChatGPT: they write "I'm missing two molars, I got a quote of 3,000 euros for two implants and I don't know if that's expensive."

The way to work them is to write full situations, doubt and context included. These are the three conversation types that work best:

• Comparing options: the user already knows what they want and is choosing between alternatives or hesitating between two ways to solve it.

• Price doubt or whether it is worth it: they ask how much something should cost, or whether hiring it beats doing it themselves.

• Problem described without a product name: they describe a symptom ("my site doesn't show on Google," "my summer electricity bill is too high") without yet knowing which service fixes it.

That third group is the most valuable and the one Google barely gives you, because in a search engine that person does not know what to type. In a conversation they tell the whole story.

How much it costs to advertise on ChatGPT

There are two campaign objectives with their buying models. Reach optimizes impressions and is paid on CPM. Clicks optimizes clicks and is paid on valid click. Maximum bid is set at ad group level, and Ads Manager warns if your bid will limit delivery. There are also conversion-oriented campaigns, which require measurement set up first.

The only cost reference OpenAI publishes is for CPC: it recommends starting with a maximum bid between 3 and 5 dollars per click. Be skeptical of articles giving an "official" CPM for Spain, because there is none: they are estimates from early US tests.

With that reference you can do the math you really need before asking for budget. At that bid range, every 100 clicks cost on the order of 300 to 500 dollars. If your site converts cold traffic to leads at 3 percent (normal for services), those 100 clicks are about three contacts. To learn something with minimal reliability you need several hundred clicks, so an honest test on this channel starts at about 1,000 to 1,500 euros spread over four to six weeks. At 300 euros a month you will not conclude anything.

There is an incentive that changes the math right now: OpenAI offers 500 dollars in ad credit when you spend your first 500 dollars. It is an acquisition offer and may disappear, but while it stands it covers much of the learning cost.

How to set up the account (and three mistakes you cannot undo)

Setup has five steps: create the account at ads.openai.com, complete onboarding with identity and business verification, fill account information, set up billing, and invite the team. Until setup is complete and approved, campaigns do not deliver even if created.

Three things are decided once with no going back:

• Country, currency, and time zone cannot be changed after you submit the account. If you pick the wrong currency or time zone, you are stuck with misaligned reporting forever.

• One email, one advertiser account. There is no MCC-style structure or agency umbrella yet, so if you manage several brands you need separate accounts and emails from the start.

• Verification is not instant. It can take weeks, because there is manual review that the advertiser and product fit policies. Do not commit to a launch date before approval.

For billing, the model is postpaid by card: spend accumulates and charges when a threshold is hit, which is not a campaign spend cap. When validating the card the bank may show a temporary authorization of about 100 dollars that is not a real charge. If there is a payment issue, delivery stops.

How to measure campaigns properly

Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM, and conversions. That is enough to know if you are delivering, but not if you are making money. For that there are three layers and it is worth setting up all three.

The first is UTM parameters on the destination URL. Static parameters are kept on click, so you can isolate ChatGPT Ads traffic in analytics. If you do not tag it, it will mix with ChatGPT referral traffic and you will not know what came from the ad versus an organic model recommendation. That distinction is exactly what you want to prove.

The second is OpenAI's measurement pixel on the site, which records actions after the click. The third is the Conversions API, to send conversions from your server or CRM. That is what saves leads closed by phone days later, and what survives blockers and cookie loss. If you send the same event via pixel and API, you must send the same event ID in both sources so they deduplicate; otherwise you double count and optimize on inflated data.

Before spending a euro here, check that your GA4 and Tag Manager measure real conversions and that you know how to count leads, not visits. A new channel with weak measurement is the fastest way to spend budget without defending results.

What you cannot do in ChatGPT Ads

This part is worth having clear because it is where inflated expectations are highest. Paying does not buy recommendation: ads run on systems separate from the chat model and advertisers cannot influence, order, or modify ChatGPT responses. Seeing an ad does not mean OpenAI endorses it.

You also do not receive personal data. Advertisers do not access conversations, history, memory, name, email, precise location, or IP: only aggregated performance reports. If someone offers to segment by ChatGPT user profile, they do not know what they are talking about.

Some categories are closed: political advertising is not allowed, and ads do not appear next to sensitive content such as personal or mental health. Regulated sectors such as health or financial services may be eligible if they meet stricter eligibility criteria.

Ads do not replace being recommended by AI

Here is the part almost no article connects, and it is what decides the most money. An ad puts you below the response. Having the model name you inside the response is something else, not for sale, and still depends on organic work.

The difference matters for how people read: if ChatGPT recommends three providers in the text and you appear only in the sponsored block below, you compete at a trust disadvantage. If you appear in both places, the ad confirms what the answer already said. That is the scenario to aim for, and why working organic visibility in AI and ads together beats either one alone.

The platform also uses your landing page as a signal for whether your ad fits a conversation. A clear, crawlable landing that answers the user's specific doubt helps twice: it improves relevance in the auction and converts better. What already rewards Google with AI Overviews applies here too.

How to start without burning budget

If the channel fits you, this is the order that makes sense today, with the beta as it is:

1. Check that your customer is in the inventory. If they sell to Plus or Pro users, stop here and put budget elsewhere.

2. Start verification as early as possible, with country, currency, and time zone decided. It is the slowest step and not fully in your control.

3. Write conversations before ads. Ten to fifteen real situations from your customer, with their doubt as they would tell it. Context hints and headlines come from that.

4. Set up measurement before going live: UTMs, pixel, and if you sell services with offline close, Conversions API.

5. Start with Clicks, not Reach. On a channel with no history you need intent signals, not impressions you cannot value yet.

6. One landing per conversation, not the home page. Destination URL is both a relevance signal and a conversion lever.

7. Reserve budget to learn, on the order of four to six weeks, and use welcome credit while it lasts.

One honest warning: this is a beta with basic objectives, no agency structure, and personalization off in Europe. It can be an excellent channel to get ahead of competitors, as Google Ads was when almost nobody bid, but it is not where to put the budget your sales depend on this quarter. If the numbers do not work yet in your case, Google Ads and Meta are still the floor you build on.