ChatGPT ads can target people by uploaded phone or email

Joseph Cox of 404 Media says it took him five minutes to get an ad into the ChatGPT sessions of strangers. He used the same tools any advertiser gets, and ChatGPT has one billion monthly active users. In 404 Media's tests of ChatGPT's ad system, OpenAI let anyone upload raw email addresses, phone numbers or a unique identifier from a person's phone, then show ads to exactly those people.

Cox's ad warned users that ChatGPT is not private and that OpenAI hires human contractors to read users' prompts. He aimed it at people asking ChatGPT about privacy, AI and whether their chats are private. That shows the second kind of targeting 404 Media describes, which uses the prompts users type. Many users probably assume those prompts are private. 404 Media has reported before that humans may review them, and in this case they were used to choose who saw an ad.

The ease of the test matches how OpenAI has rolled out ads. Testing in ChatGPT was announced on January 16, and a pilot for US users on the Free and Go tiers went live on February 9. On May 5, the self-serve Ads Manager opened to every US business with no minimum spend. The February pilot excluded health, mental health and politics. Privacy was not on that list.

404 Media says uploading a list of contacts to target people is somewhat standard practice in the ad industry. Advertisers do the same with customer lists on other platforms. The outlet sees two differences. First, ChatGPT users may not realize they are seeing an ad because an advertiser uploaded their personal data. Second, targeting based on prompts is new.

Not every user is in this pool. Ads appear on the Free and Go plans. Plus, Pro and Business plans have never shown ads, and Free users can choose fewer daily messages in exchange for no ads.