OpenAI to Test Visual Ads in ChatGPT, Starting With Image Generation
OpenAI will start testing visual image ads in ChatGPT in the United States later this month. The company announced the format on Monday, Oct. 5. The Verge
The first placement is tied to image generation. The ads will appear when users ask ChatGPT to generate images. OpenAI says the ads will stay separate from generated images and will not influence ChatGPT's answers.
The test will run with an initial group of advertisers. OpenAI first brought ads to ChatGPT in February. Ads will not appear for Plus, Pro or Enterprise subscribers.
Alongside the format, OpenAI expanded measurement tools, attribution partnerships for linking ad views to later outcomes, and brand suitability offerings that help advertisers avoid sensitive contexts. OpenAI The company lists the update as "Building advertising for the way people use AI," dated Oct. 5, 2026.
On the buying side, OpenAI is building campaign management into the chat interface itself. Its Ads Manager lets advertisers use natural-language prompts to create, update and analyze campaigns directly in ChatGPT. That workflow was outlined in the Sept. 16 article "Reimagining advertising with AI" on openai.com.
Documentation is split between buyers and users. The Help Center hosts an "Ads in ChatGPT" FAQ covering eligibility, personalization, privacy and controls. A separate guide, "Create Ads for ChatGPT Ads," advises advertisers to keep ad images simple and relevant and avoid overly abstract or cluttered visuals.
The stated framework for the program dates to February. OpenAI says testing ads helps support free access. It says ads will be clearly labeled and will include strong privacy protections.
The broader context here is why images come first. Image prompts state colors, objects and styles directly, which makes matching an ad simpler than parsing open-ended conversation. Like a sidebar next to results, that setup is easier to check and keeps paid content away from factual or advice answers.
In my view, the structural separation matters more than how the ads look. Keeping paid units apart from model output, with independent measurement and suitability controls, answers practical questions about verification and placement. For the longer term, plain-language tools should invite more experimentation, and if labeling and eligibility rules hold, ad-supported computing could help fund broad free access without eroding trust in the answers.
