ChatGPT ads can match a conversation to an advertiser’s customer list

Advertisers can upload customer or prospect lists to OpenAI and use them to determine who is eligible to see an ad inside ChatGPT. Accepted identifiers include raw or SHA-256-hashed email addresses and phone numbers, plus Google Advertising IDs from Android devices. OpenAI matches those records against users and lets an advertiser include or exclude the resulting audience or bid more aggressively for it.

404 Media tested the system by buying two campaigns. It found that prompt context also shapes delivery: requests such as “free AI video generator with sound” produced an ad for an AI-video tool, while questions about content moderation and app reviews surfaced relevant commercial services. OpenAI’s own documentation says the system considers the current conversation’s context and intent, advertiser-supplied hints and—when personalization is enabled—selected signals from a user’s broader ChatGPT experience.

There are limits. Inclusion targeting and bid adjustments require at least 25,000 matched users, and Ads Manager does not reveal which individuals matched. OpenAI says uploaded files are deleted after processing, generally within 24 hours. Personalized ads and custom audiences are not currently available for campaigns targeting the European Economic Area or Switzerland. Those constraints reduce straightforward one-person targeting, but they do not eliminate segmentation of sensitive or unusually specific groups.

Customer-list matching is routine on Meta and Google. ChatGPT changes the setting. A search engine observes queries; a retailer knows customers; a chatbot can receive extended discussions about work, relationships, health, finances and plans. OpenAI prohibits sensitive targeting and places ads only in contexts it classifies as safe, but enforcement depends on how the platform infers the advertiser’s audience and the conversation’s meaning.

The important disclosure is therefore not that an advertiser receives chat transcripts—it does not. It is that two information streams meet inside OpenAI: identity claims supplied by the advertiser and intent inferred from the conversation. Users need a visible explanation of which stream made an ad eligible, durable controls over audience matching and independent testing of whether sensitive-context restrictions work in practice.