ChatGPT is introducing ads in Switzerland while changing its web search and its handling of business data. You are directly affected if you use the free version or the lower-priced Go plan, while companies face a separate set of privacy promises for highly capable models. These developments belong together because ChatGPT increasingly serves as a search engine, decision aid, and workplace tool.
Where ads appear
OpenAI plans to serve ads in 31 European markets. According to the report on the European rollout, the company did not explicitly list Switzerland, but the Swiss launch is expected on August 24, 2026. Ads will appear in the free version and paid Go plan, while Plus, Pro, and Enterprise will remain ad-free.
For you, this creates a clear dividing line based on subscription level: users who pay nothing or less will see advertising, while the more expensive plans avoid it. The sources provide neither specific prices nor details about how many ads may appear in one conversation. They also do not say whether the launch will occur simultaneously across all Swiss language regions or offer the same range of campaigns in each one.
Advertisers can initially book placements directly through OpenAI or through agency and technology partners. A self-service Ads Manager is expected later, allowing companies to create and manage campaigns on their own. Campaigns can target regions and audience groups and can be optimized for conversions, meaning desired actions such as purchases or sign-ups.
For measurement, OpenAI names the OpenAI Pixel, a Conversions API, and connections to outside measurement providers. A pixel is technology that attributes activity on a website to an advertising campaign; an application programming interface (API) transfers data between services. The supplied sources do not explain which conversation data, if any, is used to build audiences or measure results.
Ad placement becomes especially sensitive when you ask ChatGPT for a travel destination, suitable software, or a product. In these situations, the conversation is already part of a purchasing decision. According to the lead report, OpenAI says ads will not influence the assistant’s answers, but that separation has not yet been independently demonstrated in everyday use.
How well the ads fit
The premise behind advertising in a chatbot is that the system can infer intent from a conversation more accurately than an ordinary ad slot can. An analysis by Searchable, covered by AI News, presents a less tidy result: roughly one-third of the examined ChatGPT ads reportedly appeared in irrelevant conversations. The analysis covered more than 11,000 ads from real conversations between July 4 and August 4, 2026.
The available summary does not describe how the conversations were selected or exactly how irrelevance was judged. The finding should therefore not be treated as a complete picture of every ChatGPT ad. It does indicate that access to conversational context does not automatically produce precise advertising, even when a chatbot appears to know a great deal about the current request.
For you, an irrelevant ad may simply be an annoyance. A more serious problem would be an ad that fits the subject but looks like a neutral part of the answer. Relevance alone is not enough; paid material also needs to be clearly labeled and separated from the AI-generated response.
How search is changing
ChatGPT’s underlying search behavior is changing alongside its advertising model. Observations by Promptwatch indicate that ChatGPT is using the “site:” operator much more often in so-called fan-out searches. Fan-out means the system turns one request into several search queries, while the operator limits a search to a particular website or domain.
The pattern documented by Simon Willison shows an increase from about 0.3 to 0.5 percent to 16 or 17 percent on August 8, 2026. However, these figures cover only the queries that Promptwatch tracks automatically. They suggest a product change but are not a comprehensive measurement of all ChatGPT search traffic.
The jump coincided with the rollout of GPT-5.6. OpenAI had said only that it was making answers for Plus and Pro users more factually reliable and more focused. Based on his own testing, Willison suspects that ChatGPT uses a search tool with fields for a query, recency, and domains rather than literally typing the “site:” operator; the exact internal mechanism remains hidden.
Promptwatch also observed that Reddit became much less likely to appear as a source starting on August 18. Willison could not establish whether OpenAI had explicitly instructed the system to avoid Reddit. Restricting searches to selected domains may give you more focused results, but it can also remove sources without visibly explaining why they were excluded.
For companies, this is creating a field called Generative Engine Optimization, meaning the optimization of content for mentions in generative-system answers. It resembles search engine optimization but focuses on chatbots rather than traditional result pages. Because these changes were detected by outside monitoring services and OpenAI discloses little about its search logic, the evidence remains limited.
What the privacy promises cover
The clearest new privacy promises do not apply to personal ChatGPT accounts. They concern eligible business customers using the API. OpenAI has reaffirmed Zero Data Retention (ZDR), meaning customer data is not stored after processing, for certain customers using highly capable models. The company’s own announcement also previews Private Safety Processing, an automated method intended to detect abuse without permanently retaining the processed content.
According to OpenAI, the system can identify patterns across multiple related interactions. The company says it receives only a narrowly defined safety signal describing the type and severity of an activity, not the actual prompts and responses. Customer data is supposed to remain on the customer’s own infrastructure or be stored in encrypted form with the customer retaining the keys.
OpenAI is trying to combine two goals: avoiding the storage of confidential business data while still detecting misuse spread across several sessions. The reporting gives the example of someone distributing requests for malware development across different conversations to evade detection. TechCrunch describes the system as a preview for select customers, not as a broadly available and independently verified service.
A technical white paper is expected in September, according to The Decoder. Until then, central claims about how the system works rest on OpenAI’s account. The sources also conflict over the names of Anthropic’s affected models: The Decoder refers to Fable 5, while TechCrunch names Mythos-class models and future models with similar capabilities. Both report a 30-day retention period for certain Anthropic models.
These enterprise commitments do not provide an equivalent guarantee for private users. The sources describe neither a ZDR policy for free ChatGPT conversations nor exactly whether and how conversation content is evaluated for advertising. In Switzerland, that leaves unanswered questions about what data supports regional targeting, ad measurement, and outside measurement services, as well as where that processing occurs.
Pros and Cons of the new ChatGPT rules
Pros:
- Subscription choice – Plus, Pro, and Enterprise are expected to remain available without ads.
- More focused search – Domain-specific searches can help ground answers in more targeted sources.
- Less storage for business use – ZDR is intended to let eligible API customers use capable models without permanent retention of their content.
- Safety across sessions – Private Safety Processing is designed to detect distributed abuse patterns without people reviewing the underlying conversations.
Cons:
- Irrelevant advertising – The analyzed sample suggests that roughly one in three ads did not match its conversation.
- Unclear data flows – The sources do not explain which personal chat data supports ad selection, delivery, and measurement.
- Invisible source selection – Stronger domain filtering can narrow search results without showing you the rules behind that choice.
- Limited independent scrutiny – The broad privacy and safety claims come mainly from OpenAI and have not yet been comprehensively verified.
What this means for you
If you are starting out, first check which plan you use and whether ads are clearly identified as paid placements. When ChatGPT recommends a trip, software package, or product, you can read the main answer separately from the ad and open the cited sources. A new privacy announcement by itself is not a reason to place confidential personal or business information in an ordinary ChatGPT conversation.
If you are a more advanced user, examine the range of sources behind web-search answers. You can check whether responses repeatedly rely on a small number of domains and whether relevant opposing views are absent. With ads, it also helps to distinguish topical relevance from the actual quality of the promoted offer; fitting the conversation does not make an ad an independent recommendation.
For organizations, the practical first step is to distinguish an ordinary ChatGPT account from an eligible API setup with ZDR. Teams handling sensitive information can determine whether they qualify for the select rollout of Private Safety Processing and which commitments are contractual. The supplied sources specify neither general availability nor pricing or special terms for Swiss companies.
ChatGPT is becoming more commercial in Switzerland while its search grows more focused and less transparent. The new privacy mechanisms could make sensitive business use easier, but they do not automatically extend to personal conversations and are not yet fully documented. The largest unresolved risk is the combination of personalized advertising, hard-to-see source selection, and unclear data flows.
Sources
- OpenAI bringt Werbung in ChatGPT nach Europa – Netzwoche, 2026-08-19
- A third of ChatGPT ads appear in irrelevant conversations – AI News, 2026-08-20
- ChatGPT search now uses the site:operator at scale – Simon Willison’s Weblog, 2026-08-20
- OpenAI will Missbrauch aufspüren und trotzdem keine Kundendaten speichern – The Decoder, 2026-08-20
- Offering Zero Data Retention for frontier models – OpenAI, 2026-08-19
- OpenAI seeks to one-up Anthropic with new customer privacy protections – TechCrunch, 2026-08-19
