You ask a chatbot for advice: which mattress to buy, which vitamin, which bank. It answers in three calm paragraphs. And since August, directly underneath that answer, there is a paid advertisement from a company you have never heard of.
What actually changed
On 24 August, OpenAI, the company that makes ChatGPT, began showing adverts inside the chatbot in 31 European countries, Germany among them. They are not on a separate page. They sit below the reply, separated by a thin line and marked “Anzeige” or “Gesponsert”.
Only people on the free plan and on Go, the cheaper paid plan, see them. The more expensive subscriptions, work accounts and education accounts have no adverts, and neither does anyone the system believes is under 18. OpenAI’s promise is that the advert does not change the answer: its help pages say adverts run on a separate system, that advertisers cannot alter a reply, and that seeing an ad does not mean OpenAI recommends the company.
Why a small card under an answer matters
A search engine hands you a list and you choose. A chatbot hands you one answer, written as though somebody had thought about you. An advert placed directly under that answer borrows some of its certainty. You read the answer and the shop in the same glance.
The German consumer advice centres (Verbraucherzentralen) warned about exactly this on 22 October: a product recommendation you see in a chat can be paid advertising, so check AI answers against other sources before you spend money. That is not a small point. It is the difference between advice and a shop window.
The part that is coming
Today in Europe the advert is chosen from the conversation you are having right now: what you asked, your rough location, your language.
That is about to change. At the end of August OpenAI rewrote its privacy policy to allow personalised advertising, built from your past and current chats, your device and how you use it. In Europe that needs your consent first, so it has not started here yet.
Four German data protection authorities warned about it within two days at the start of October. Berlin’s commissioner, Meike Kamp, said it reaches a new level of intrusion when information from chats, from the devices used and from past behaviour is worked in, and that users have to understand that very sensitive data may be processed and that this is not an anonymous conversation. Lower Saxony’s commissioner, Denis Lehmkemper, said it from the other side: people confide intimate thoughts to chatbots, about their health, their moods, their plans, and that is precisely what a profile is made of.
ChatGPT is the most used chatbot in Germany, so this concerns millions of people. Both authorities cite OpenAI’s own finding that 49 percent of what users type is a question. People treat it as an adviser.
What you can actually do
- Look for the label. “Anzeige” or “Gesponsert” means a company paid. The three dots next to an advert let you hide it or report it.
- Turn personalisation off now, before you are asked. In ChatGPT open Settings (Einstellungen) and find the ad controls. There is a switch for personalised adverts and a second one for past chats and memory.
- Say no when the consent screen appears. Both Hamburg and Berlin advise declining personalised advertising.
- Use Temporary Chat for anything personal: a symptom, a debt, an argument at home. It keeps no history, uses no memory, and shows no adverts.
- Clear what is already stored. Delete chats you no longer need and review the memories ChatGPT has saved about you.
- If adverts bother you more than the limits do, switch to the free ad-free version. You get fewer messages a day and lose some tools. The paid plans carry no adverts at all.
Common misunderstandings
Myth: if it recommends a shop, the shop must be good
An advert is a paid placement. OpenAI says plainly that seeing one does not mean it recommends the advertiser. The paragraphs above were written by the chatbot; the card below was bought by a company. They share a screen, not a judgement.
Myth: refusing personalised adverts stops the adverts
You will still see them. Refusing only stops your history being used. The advert is picked from the conversation in front of you, which is why asking about a mattress and then seeing a mattress shop is not a coincidence. The consumer advice centres note that the current chat alone decides what you are shown.
Myth: I can ask ChatGPT whether the advert is a good deal
By default it cannot see the advert at all: the card sits outside the conversation. You have to open the three dots on the advert and choose “Ask ChatGPT” to hand it over, and even then you are asking the company that sold the placement. Compare the price somewhere the chat cannot reach.
What to watch for next
Watch for the consent screen. When personalised advertising arrives in Germany you will be asked to agree, and both authorities say declining is the safer answer. Declining does not remove them. It only limits what they are built from.
Watch the formats. OpenAI is already testing adverts that appear while it draws an image, several pictures from one advertiser. Every new format is another place where a paid card can look like part of the product.
And watch what you type. The advice every authority gives needs no settings at all: health problems, debts, legal trouble and other people’s secrets do not belong in a chatbot.
The one thing to remember
The answer is the product. The card underneath is the shop. Before you buy anything a chatbot puts in front of you, look for the word “Anzeige” and check the price somewhere the conversation cannot reach.
Image: “Woman using an iPhone” by Kristin Hardwick, released under CC0 1.0 (public domain dedication), via Wikimedia Commons.