How ChatGPT Targets Ads To You Based On Your Prompts, Phone Number, and Email
404 Media Joseph Cox ● Covered by 2 sources
ChatGPT can now show ads based on your prompts, email, phone number, or phone ID. That means “private” chats can feed ad targeting too.
Based on reporting by 404 Media, Joseph Cox — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI is letting advertisers target people inside ChatGPT using personal data, and 404 Media’s own tests found just how far that can go. An ad buyer can upload email addresses, phone numbers, hashed versions of both, or Android advertising IDs, then aim ads at matched users. OpenAI also lets advertisers add “context hints” tied to topics and keywords, so the system can match people based on what they type into ChatGPT as well as who they are.
That mix is the uncomfortable part. ChatGPT now has about one billion monthly active users, and some of them may not realize their prompts can help shape the ads they see. OpenAI has already rolled out ads in the U.S. in February and then expanded them to other places. This month, it pushed ads next to image generation results too. That’s a steady climb from simple placement to something much more personal.
404 Media tested the system by running ads that pointed back to its reporting on Project Lily, the program in which contractors read real user prompts. OpenAI’s ad tools let advertisers describe the kind of conversations they want to reach, and 404 Media tried targeting people interested in privacy, ChatGPT chats, and current technology and AI topics. OpenAI’s own wording says those hints guide matching, but are not exact-match rules.
There are limits, but they don’t make the setup feel especially gentle. To be eligible for “inclusion,” a list needs more than 25,000 entries. OpenAI says uploaded files are deleted after processing, usually within 24 hours, and says advertisers cannot see which individuals matched or pick specific people from an audience. Users can switch off “ads personalization,” but the setting still allows ads based on current chats.
And now there are tools built to map the whole thing from the outside. ChatGPT Ad Library says it has recorded at least 43,410 ads across 415,289 placements by feeding prompts into ChatGPT and logging the sponsored cards that came back. Zach Edwards of DecryptAds called this the familiar playbook: new monetization, quiet privacy costs, and a lot of hope that users won’t notice until later.
My take — AI-written commentary, not fact-checked reporting
This is the cleanest proof yet that “chatbot” and “ad network” are becoming the same product. The industry keeps dressing surveillance up as relevance, and then acts surprised when people don’t love being auctioned off by their prompts. If a company needs a 25,000-person threshold to make it sound civilized, the mess is already doing the talking.
Read more about this at: 404 Media